diff --git a/.2119/verdicts/REQ-003.1.6--de6caae0b7d5.json b/.2119/verdicts/REQ-003.1.6--de6caae0b7d5.json
deleted file mode 100644
index 0f04539..0000000
--- a/.2119/verdicts/REQ-003.1.6--de6caae0b7d5.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.1.6--de6caae0b7d5",
-  "requirementId": "REQ-003.1.6",
-  "hash": "de6caae0b7d5",
-  "verdict": "pass",
-  "summary": "Review must create the identified instruction packet directly in .2119/reviews, and init is verified to add an effective .gitignore rule for files there; missing or adjacent-directory outputs fail.",
-  "timestamp": "2026-07-22T14:42:10.024Z"
-}
diff --git a/.2119/verdicts/REQ-003.1.6.json b/.2119/verdicts/REQ-003.1.6.json
new file mode 100644
index 0000000..72acdbc
--- /dev/null
+++ b/.2119/verdicts/REQ-003.1.6.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-003.1.6--1795cb813d37",
+  "requirementId": "REQ-003.1.6",
+  "hash": "1795cb813d37",
+  "verdict": "pass",
+  "summary": "Compiled CLI test confirms two generated instruction files land in .2119/reviews/ and clean init makes review packets effectively gitignored.",
+  "timestamp": "2026-08-03T15:57:48.088Z"
+}
diff --git a/.2119/verdicts/REQ-003.2.2--bea345f6b733.json b/.2119/verdicts/REQ-003.2.2--bea345f6b733.json
deleted file mode 100644
index 4e7f744..0000000
--- a/.2119/verdicts/REQ-003.2.2--bea345f6b733.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.2.2--bea345f6b733",
-  "requirementId": "REQ-003.2.2",
-  "hash": "bea345f6b733",
-  "verdict": "pass",
-  "summary": "pass and fail serialize verdict records as plain JSON in .2119/verdicts/, while init and verdict writes repair ignore rules so those records remain trackable.",
-  "timestamp": "2026-08-02T22:29:35.122Z"
-}
diff --git a/.2119/verdicts/REQ-003.2.2--f06967fba1ab.json b/.2119/verdicts/REQ-003.2.2--f06967fba1ab.json
deleted file mode 100644
index 10293bd..0000000
--- a/.2119/verdicts/REQ-003.2.2--f06967fba1ab.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.2.2--f06967fba1ab",
-  "requirementId": "REQ-003.2.2",
-  "hash": "f06967fba1ab",
-  "verdict": "pass",
-  "summary": "writeVerdict (called by pass/fail) and init both invoke ensureReviewStorageRules, which appends the 2119 gitignore block LAST in root .gitignore with !.2119/, !.2119/verdicts/ and !.2119/verdicts/**, plus last-position repairs to any existing .2119/.gitignore and .2119/verdicts/.gitignore; verdicts are written as pretty-printed 2-space JSON at .2119/verdicts/<reviewId>.json. Verified empirically: with a pre-existing '.2119/' rule and even with '*' in .gitignore, the verdict file is stageable without -f while .2119/reviews/ stays ignored.",
-  "timestamp": "2026-08-01T06:31:16.680Z"
-}
diff --git a/.2119/verdicts/REQ-003.2.2.json b/.2119/verdicts/REQ-003.2.2.json
new file mode 100644
index 0000000..b15f960
--- /dev/null
+++ b/.2119/verdicts/REQ-003.2.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-003.2.2--ba3a7f951d2b",
+  "requirementId": "REQ-003.2.2",
+  "hash": "ba3a7f951d2b",
+  "verdict": "pass",
+  "summary": "pass/fail write stable plain JSON after repairing final unignore rules, and init installs the same trackable verdict path",
+  "timestamp": "2026-08-03T17:05:00.760Z"
+}
diff --git a/.2119/verdicts/REQ-003.6.3--b1ff7ba4feab.json b/.2119/verdicts/REQ-003.6.3--b1ff7ba4feab.json
deleted file mode 100644
index f75ee66..0000000
--- a/.2119/verdicts/REQ-003.6.3--b1ff7ba4feab.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.6.3--b1ff7ba4feab",
-  "requirementId": "REQ-003.6.3",
-  "hash": "b1ff7ba4feab",
-  "verdict": "pass",
-  "summary": "Real CLI test: passes a verdict for FIX-001.1.1, runs review --audit, asserts audit file generated for the passing-verdict req only (unreviewed FIX-001.1.2 gets none), body carries construct-counterexample and pass-only-if-none directives, and the verdict file is byte-identical before/after. Minor untested near-miss: a failing (not merely absent) verdict is not exercised.",
-  "timestamp": "2026-07-10T23:03:33.182Z"
-}
diff --git a/.2119/verdicts/REQ-003.6.3.json b/.2119/verdicts/REQ-003.6.3.json
new file mode 100644
index 0000000..43acdc2
--- /dev/null
+++ b/.2119/verdicts/REQ-003.6.3.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-003.6.3--4fa1d92e29a1",
+  "requirementId": "REQ-003.6.3",
+  "hash": "4fa1d92e29a1",
+  "verdict": "pass",
+  "summary": "Packaged CLI test rejects missing or mis-scoped audits, non-adversarial directives, stale passes, and any byte change to current, failing, orphan, or multiple verdict records",
+  "timestamp": "2026-08-03T16:15:20.573Z"
+}
diff --git a/.2119/verdicts/REQ-003.7.2--d253b491fe00.json b/.2119/verdicts/REQ-003.7.2--d253b491fe00.json
deleted file mode 100644
index 883cd70..0000000
--- a/.2119/verdicts/REQ-003.7.2--d253b491fe00.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.7.2--d253b491fe00",
-  "requirementId": "REQ-003.7.2",
-  "hash": "d253b491fe00",
-  "verdict": "pass",
-  "summary": "Test truncates a committed verdict to invalid JSON and asserts check exits 1 with the verdict's file path, the [REQ-003.7.2] rule tag, and the reason 'unparseable JSON' in stderr — verifies loud violation naming file+reason, not a silent skip/pass.",
-  "timestamp": "2026-07-10T17:51:34.034Z"
-}
diff --git a/.2119/verdicts/REQ-003.7.2.json b/.2119/verdicts/REQ-003.7.2.json
new file mode 100644
index 0000000..677b2c7
--- /dev/null
+++ b/.2119/verdicts/REQ-003.7.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-003.7.2--9d4edf786d45",
+  "requirementId": "REQ-003.7.2",
+  "hash": "9d4edf786d45",
+  "verdict": "pass",
+  "summary": "Annotated test rejects truncated JSON and an empty-summary record, with check naming the stable verdict file and concrete parse or validation reason",
+  "timestamp": "2026-08-03T15:57:38.695Z"
+}
diff --git a/.2119/verdicts/REQ-003.7.3--dd9420ef82ff.json b/.2119/verdicts/REQ-003.7.3--dd9420ef82ff.json
deleted file mode 100644
index 69ed135..0000000
--- a/.2119/verdicts/REQ-003.7.3--dd9420ef82ff.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.7.3--dd9420ef82ff",
-  "requirementId": "REQ-003.7.3",
-  "hash": "dd9420ef82ff",
-  "verdict": "pass",
-  "summary": "Test renames a valid verdict to a mismatched filename and asserts check exits 1 with 'filename does not match reviewId' — genuinely verifies filename-mismatch is treated as malformed rather than counted.",
-  "timestamp": "2026-07-10T17:51:34.130Z"
-}
diff --git a/.2119/verdicts/REQ-003.7.3.json b/.2119/verdicts/REQ-003.7.3.json
new file mode 100644
index 0000000..d1d0bbf
--- /dev/null
+++ b/.2119/verdicts/REQ-003.7.3.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-003.7.3--957c25b9bde5",
+  "requirementId": "REQ-003.7.3",
+  "hash": "957c25b9bde5",
+  "verdict": "fail",
+  "summary": "The test accepts <requirementId>.json despite the exact <reviewId>.json requirement and never verifies prune agreement; it only rejects a legacy-shaped filename with the wrong hash.",
+  "timestamp": "2026-08-03T15:52:09.333Z"
+}
diff --git a/.2119/verdicts/REQ-003.8.2--5262e7c71e7b.json b/.2119/verdicts/REQ-003.8.2--5262e7c71e7b.json
deleted file mode 100644
index 98d9ee8..0000000
--- a/.2119/verdicts/REQ-003.8.2--5262e7c71e7b.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.8.2--5262e7c71e7b",
-  "requirementId": "REQ-003.8.2",
-  "hash": "5262e7c71e7b",
-  "verdict": "fail",
-  "summary": "The current template cannot preserve calibration case 013's expected PASS: its conjunct rule requires rejected counterexamples for empty note, mismatched owner, and unparseable stamp, which that case's evidence does not provide.",
-  "timestamp": "2026-08-02T22:33:18.373Z"
-}
diff --git a/.2119/verdicts/REQ-003.8.2--5ec997925e8d.json b/.2119/verdicts/REQ-003.8.2--5ec997925e8d.json
deleted file mode 100644
index 09c2ee1..0000000
--- a/.2119/verdicts/REQ-003.8.2--5ec997925e8d.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.8.2--5ec997925e8d",
-  "requirementId": "REQ-003.8.2",
-  "hash": "5ec997925e8d",
-  "verdict": "fail",
-  "summary": "Calibration case 013-pass-control-negative-positive expects PASS, but the template requires a rejected counterexample for every conjunct while its evidence never independently rejects an empty note, mismatched owner, or unparseable stamp; a reviewer following the template must FAIL that case.",
-  "timestamp": "2026-08-02T22:31:16.326Z"
-}
diff --git a/.2119/verdicts/REQ-003.8.2--5fff49bf72aa.json b/.2119/verdicts/REQ-003.8.2--5fff49bf72aa.json
deleted file mode 100644
index 89fee85..0000000
--- a/.2119/verdicts/REQ-003.8.2--5fff49bf72aa.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.8.2--5fff49bf72aa",
-  "requirementId": "REQ-003.8.2",
-  "hash": "5fff49bf72aa",
-  "verdict": "pass",
-  "summary": "src/review.ts template prose (audit prompt, dispatch prompt, renderInstructions) is byte-identical to a92ad03, the commit that seeded eval/calibration, so no clause has been dropped since the corpus existed; every one of the 13 cases maps to a live clause — 'comment' is a named boundary term (001), conjunct enumeration (002/006/007), 'cannot fail' tautology (008), 'assert what they just set up' (009), over-mocking (011), criterion-mismatch (010), and 'judge the requirement too: mechanism not outcome' (005/011), with the two pass controls unpenalized.",
-  "timestamp": "2026-08-01T05:48:55.301Z"
-}
diff --git a/.2119/verdicts/REQ-003.8.2--d52ce5a3aa84.json b/.2119/verdicts/REQ-003.8.2--d52ce5a3aa84.json
deleted file mode 100644
index d0137cf..0000000
--- a/.2119/verdicts/REQ-003.8.2--d52ce5a3aa84.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.8.2--d52ce5a3aa84",
-  "requirementId": "REQ-003.8.2",
-  "hash": "d52ce5a3aa84",
-  "verdict": "pass",
-  "summary": "The current standard and audit instructions retain the checks needed by calibration cases 001-014: tautology/over-mocking flags, provenance, conjunct and boundary counterexamples, requirement-quality judgment, and evidence-bounded summary scope; case 013 now rejects each named malformed-record conjunct and keeps a genuine-writer pass control.",
-  "timestamp": "2026-08-02T22:34:56.716Z"
-}
diff --git a/.2119/verdicts/REQ-003.8.2--deb61e2892be.json b/.2119/verdicts/REQ-003.8.2--deb61e2892be.json
deleted file mode 100644
index 0308d2a..0000000
--- a/.2119/verdicts/REQ-003.8.2--deb61e2892be.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.8.2--deb61e2892be",
-  "requirementId": "REQ-003.8.2",
-  "hash": "deb61e2892be",
-  "verdict": "fail",
-  "summary": "The test-quality template forbids PASS without production file:line provenance, but calibration cases 012-pass-control-perturbation and 013-pass-control-negative-positive contain only condensed snippets and expect PASS, so a reviewer following the template cannot preserve those two corpus verdicts.",
-  "timestamp": "2026-08-02T22:29:50.495Z"
-}
diff --git a/.2119/verdicts/REQ-003.8.2--ec2046ab17a0.json b/.2119/verdicts/REQ-003.8.2--ec2046ab17a0.json
deleted file mode 100644
index 6f90f0b..0000000
--- a/.2119/verdicts/REQ-003.8.2--ec2046ab17a0.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-003.8.2--ec2046ab17a0",
-  "requirementId": "REQ-003.8.2",
-  "hash": "ec2046ab17a0",
-  "verdict": "pass",
-  "summary": "The current standard, direct-judgment, and audit instructions preserve the expected verdicts for calibration cases 001-014; corrected case 013 now independently covers its missing verdict, invalid verdict, and real-CLI acceptance clauses.",
-  "timestamp": "2026-08-02T23:02:58.593Z"
-}
diff --git a/.2119/verdicts/REQ-003.8.2.json b/.2119/verdicts/REQ-003.8.2.json
new file mode 100644
index 0000000..c0ecc90
--- /dev/null
+++ b/.2119/verdicts/REQ-003.8.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-003.8.2--6e3c630dbed4",
+  "requirementId": "REQ-003.8.2",
+  "hash": "6e3c630dbed4",
+  "verdict": "pass",
+  "summary": "test-quality, direct-judgment, audit, and evidence-scoping instructions still expose every expected pass/fail distinction in corpus cases 001-014",
+  "timestamp": "2026-08-03T17:05:00.892Z"
+}
diff --git a/.2119/verdicts/REQ-006.1.1--4ca55c38f026.json b/.2119/verdicts/REQ-006.1.1--4ca55c38f026.json
deleted file mode 100644
index 564d57b..0000000
--- a/.2119/verdicts/REQ-006.1.1--4ca55c38f026.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-006.1.1--4ca55c38f026",
-  "requirementId": "REQ-006.1.1",
-  "hash": "4ca55c38f026",
-  "verdict": "pass",
-  "summary": "Test runs 'review', edits the covering test to mint a new review ID, runs 'review' again, and asserts the directory holds exactly one file that differs from the old one and the old instruction file no longer exists.",
-  "timestamp": "2026-07-09T23:07:20.226Z"
-}
diff --git a/.2119/verdicts/REQ-006.1.1.json b/.2119/verdicts/REQ-006.1.1.json
new file mode 100644
index 0000000..ec8d889
--- /dev/null
+++ b/.2119/verdicts/REQ-006.1.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-006.1.1--5adc90405042",
+  "requirementId": "REQ-006.1.1",
+  "hash": "5adc90405042",
+  "verdict": "pass",
+  "summary": "CLI test rejects superseded, passed, and unrelated instruction packets and asserts the remaining .md files equal the production-computed pending review IDs",
+  "timestamp": "2026-08-03T16:14:55.176Z"
+}
diff --git a/.2119/verdicts/REQ-006.1.2--98ec8eb76858.json b/.2119/verdicts/REQ-006.1.2--98ec8eb76858.json
deleted file mode 100644
index 14cc20e..0000000
--- a/.2119/verdicts/REQ-006.1.2--98ec8eb76858.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-006.1.2--98ec8eb76858",
-  "requirementId": "REQ-006.1.2",
-  "hash": "98ec8eb76858",
-  "verdict": "pass",
-  "summary": "Test passes the pending review's verdict, reruns 'review', and asserts status 0, the 'all judgment reviews have current passing verdicts' message, and an empty .2119/reviews directory.",
-  "timestamp": "2026-07-09T23:07:20.314Z"
-}
diff --git a/.2119/verdicts/REQ-006.1.2.json b/.2119/verdicts/REQ-006.1.2.json
new file mode 100644
index 0000000..acf88ae
--- /dev/null
+++ b/.2119/verdicts/REQ-006.1.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-006.1.2--c08720df5ca0",
+  "requirementId": "REQ-006.1.2",
+  "hash": "c08720df5ca0",
+  "verdict": "pass",
+  "summary": "Production CLI test proves plain review removes both generated audit and stale instruction packets after pass leaves zero pending reviews",
+  "timestamp": "2026-08-03T16:07:26.817Z"
+}
diff --git a/.2119/verdicts/REQ-006.2.1--e9a3856f19a9.json b/.2119/verdicts/REQ-006.2.1--e9a3856f19a9.json
deleted file mode 100644
index bd23d42..0000000
--- a/.2119/verdicts/REQ-006.2.1--e9a3856f19a9.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-006.2.1--e9a3856f19a9",
-  "requirementId": "REQ-006.2.1",
-  "hash": "e9a3856f19a9",
-  "verdict": "pass",
-  "summary": "Test passes a verdict then orphans it by editing the covering test, runs 'prune', and asserts stdout lists 'pruned .2119/verdicts/<oldId>.json' plus 'removed 1 orphaned verdict(s)' and the file is gone.",
-  "timestamp": "2026-07-09T23:07:20.410Z"
-}
diff --git a/.2119/verdicts/REQ-006.2.1.json b/.2119/verdicts/REQ-006.2.1.json
new file mode 100644
index 0000000..bc67691
--- /dev/null
+++ b/.2119/verdicts/REQ-006.2.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-006.2.1--b82b2c282604",
+  "requirementId": "REQ-006.2.1",
+  "hash": "b82b2c282604",
+  "verdict": "pass",
+  "summary": "The prune CLI test deletes and individually lists a production-created stale stable verdict plus stable and legacy verdicts with no current target.",
+  "timestamp": "2026-08-03T16:05:29.472Z"
+}
diff --git a/.2119/verdicts/REQ-006.2.2--50fa3281d164.json b/.2119/verdicts/REQ-006.2.2--50fa3281d164.json
deleted file mode 100644
index 14cf906..0000000
--- a/.2119/verdicts/REQ-006.2.2--50fa3281d164.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-006.2.2--50fa3281d164",
-  "requirementId": "REQ-006.2.2",
-  "hash": "50fa3281d164",
-  "verdict": "pass",
-  "summary": "Test passes a verdict (check green), runs 'prune', and asserts 'removed 0 orphaned verdict(s)', the verdict file still exists, and 'check' still exits 0 afterwards.",
-  "timestamp": "2026-07-09T23:07:20.514Z"
-}
diff --git a/.2119/verdicts/REQ-006.2.2.json b/.2119/verdicts/REQ-006.2.2.json
new file mode 100644
index 0000000..f632788
--- /dev/null
+++ b/.2119/verdicts/REQ-006.2.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-006.2.2--4a4c997aba5e",
+  "requirementId": "REQ-006.2.2",
+  "hash": "4a4c997aba5e",
+  "verdict": "pass",
+  "summary": "Production-generated current verdict fields survive legacy-to-canonical prune migration exactly, and the post-prune check remains passing.",
+  "timestamp": "2026-08-03T17:19:24.239Z"
+}
diff --git a/.2119/verdicts/REQ-010.2.5--35d4b4549f82.json b/.2119/verdicts/REQ-010.2.5--35d4b4549f82.json
deleted file mode 100644
index 42429e6..0000000
--- a/.2119/verdicts/REQ-010.2.5--35d4b4549f82.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-010.2.5--35d4b4549f82",
-  "requirementId": "REQ-010.2.5",
-  "hash": "35d4b4549f82",
-  "verdict": "pass",
-  "summary": "Covers added, removed, replaced, assigned-malformed, and unassigned-malformed current verdicts, while rejecting historical verdict edits as outside current scope.",
-  "timestamp": "2026-07-22T14:45:42.554Z"
-}
diff --git a/.2119/verdicts/REQ-010.2.5.json b/.2119/verdicts/REQ-010.2.5.json
new file mode 100644
index 0000000..16da6dc
--- /dev/null
+++ b/.2119/verdicts/REQ-010.2.5.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-010.2.5--92e235b8d036",
+  "requirementId": "REQ-010.2.5",
+  "hash": "92e235b8d036",
+  "verdict": "pass",
+  "summary": "Built-CLI integration rejects added, removed, replaced, assigned-malformed, and unassigned-malformed current verdict changes while excluding changed historical verdicts and unrelated requirements.",
+  "timestamp": "2026-08-03T15:54:09.368Z"
+}
diff --git a/.2119/verdicts/REQ-010.3.1--df3aa1a2d7d5.json b/.2119/verdicts/REQ-010.3.1--df3aa1a2d7d5.json
deleted file mode 100644
index 96b36ee..0000000
--- a/.2119/verdicts/REQ-010.3.1--df3aa1a2d7d5.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-010.3.1--df3aa1a2d7d5",
-  "requirementId": "REQ-010.3.1",
-  "hash": "df3aa1a2d7d5",
-  "verdict": "pass",
-  "summary": "Behavioral fixtures reject each scoped boundary: changed versus unchanged lint, affected versus sibling coverage/freshness/verification, and changed versus pre-existing invalid annotations.",
-  "timestamp": "2026-07-22T14:48:09.013Z"
-}
diff --git a/.2119/verdicts/REQ-010.3.1.json b/.2119/verdicts/REQ-010.3.1.json
new file mode 100644
index 0000000..8d76697
--- /dev/null
+++ b/.2119/verdicts/REQ-010.3.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-010.3.1--4378fab560ac",
+  "requirementId": "REQ-010.3.1",
+  "hash": "4378fab560ac",
+  "verdict": "pass",
+  "summary": "End-to-end CLI cases report changed-spec lint, affected coverage/freshness/verify failures and changed-test annotation errors while suppressing paired pre-existing out-of-scope violations.",
+  "timestamp": "2026-08-03T16:10:07.763Z"
+}
diff --git a/.2119/verdicts/REQ-011.5.5--a2d801a91a6f.json b/.2119/verdicts/REQ-011.5.5--a2d801a91a6f.json
deleted file mode 100644
index 6f1649c..0000000
--- a/.2119/verdicts/REQ-011.5.5--a2d801a91a6f.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "REQ-011.5.5--a2d801a91a6f",
-  "requirementId": "REQ-011.5.5",
-  "hash": "a2d801a91a6f",
-  "verdict": "pass",
-  "summary": "Test writes two genuinely orphaned verdict files (legacy REQ-900.1.1--aaa..., file-scoped codex-session-scrollback.1.1--bbb...), runs the real 2119 prune, and asserts both files are gone from disk; a mutant that skips non-legacy review IDs in pruneVerdicts leaves the file-scoped verdict behind and fails this test.",
-  "timestamp": "2026-08-01T05:51:02.857Z"
-}
diff --git a/.2119/verdicts/REQ-011.5.5.json b/.2119/verdicts/REQ-011.5.5.json
new file mode 100644
index 0000000..82b1b39
--- /dev/null
+++ b/.2119/verdicts/REQ-011.5.5.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-011.5.5--bd08ba2ce182",
+  "requirementId": "REQ-011.5.5",
+  "hash": "bd08ba2ce182",
+  "verdict": "pass",
+  "summary": "CLI prune removes pre-existing stale stable verdict files for both REQ-900.1.1 and codex-session-scrollback.1.1",
+  "timestamp": "2026-08-03T15:55:25.252Z"
+}
diff --git a/.2119/verdicts/REQ-012.1.1--ee54df994395.json b/.2119/verdicts/REQ-012.1.1--ee54df994395.json
new file mode 100644
index 0000000..538c2c7
--- /dev/null
+++ b/.2119/verdicts/REQ-012.1.1--ee54df994395.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.1.1--ee54df994395",
+  "requirementId": "REQ-012.1.1",
+  "hash": "ee54df994395",
+  "verdict": "pass",
+  "summary": "Production CLI tests verify pass/fail create one canonical verdict file and replace same-hash, changed-hash, stable, and legacy predecessors.",
+  "timestamp": "2026-08-03T07:33:40.101Z"
+}
diff --git a/.2119/verdicts/REQ-012.2.1.json b/.2119/verdicts/REQ-012.2.1.json
new file mode 100644
index 0000000..5a6d6e8
--- /dev/null
+++ b/.2119/verdicts/REQ-012.2.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.2.1--bffae6d192ea",
+  "requirementId": "REQ-012.2.1",
+  "hash": "bffae6d192ea",
+  "verdict": "pass",
+  "summary": "Two unawaited production CLI processes record distinct review IDs; both must succeed and leave separate parseable records with their own complete fields and fresh timestamps.",
+  "timestamp": "2026-08-03T16:34:03.118Z"
+}
diff --git a/.2119/verdicts/REQ-012.3.1--2cdcb2a572f9.json b/.2119/verdicts/REQ-012.3.1--2cdcb2a572f9.json
new file mode 100644
index 0000000..86f2289
--- /dev/null
+++ b/.2119/verdicts/REQ-012.3.1--2cdcb2a572f9.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.3.1--2cdcb2a572f9",
+  "requirementId": "REQ-012.3.1",
+  "hash": "2cdcb2a572f9",
+  "verdict": "pass",
+  "summary": "Production CLI records pass/fail fields at the canonical path, while check rejects wrong IDs, hash case/value, verdict, summary, timestamp, missing fields, and wrong paths.",
+  "timestamp": "2026-08-03T07:33:40.345Z"
+}
diff --git a/.2119/verdicts/REQ-012.3.2--230dbab2979c.json b/.2119/verdicts/REQ-012.3.2--230dbab2979c.json
new file mode 100644
index 0000000..778dc24
--- /dev/null
+++ b/.2119/verdicts/REQ-012.3.2--230dbab2979c.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.3.2--230dbab2979c",
+  "requirementId": "REQ-012.3.2",
+  "hash": "230dbab2979c",
+  "verdict": "pass",
+  "summary": "The annotated test verifies both review kinds' canonical inputs and ordering, sensitivity to every included input, stability across unrelated edits, and persisted hashes matching production review IDs.",
+  "timestamp": "2026-08-03T07:33:40.464Z"
+}
diff --git a/.2119/verdicts/REQ-012.4.1--d92abce646b2.json b/.2119/verdicts/REQ-012.4.1--d92abce646b2.json
new file mode 100644
index 0000000..ef61c20
--- /dev/null
+++ b/.2119/verdicts/REQ-012.4.1--d92abce646b2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.4.1--d92abce646b2",
+  "requirementId": "REQ-012.4.1",
+  "hash": "d92abce646b2",
+  "verdict": "pass",
+  "summary": "Current stable pass is accepted; stale, hash-mismatched, failing, and malformed stable records remain violations despite current legacy pass evidence.",
+  "timestamp": "2026-08-03T07:33:40.582Z"
+}
diff --git a/.2119/verdicts/REQ-012.5.1.json b/.2119/verdicts/REQ-012.5.1.json
new file mode 100644
index 0000000..f189b9e
--- /dev/null
+++ b/.2119/verdicts/REQ-012.5.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.5.1--3828831c8775",
+  "requirementId": "REQ-012.5.1",
+  "hash": "3828831c8775",
+  "verdict": "pass",
+  "summary": "Real-CLI coverage rejects malformed, current-fail, and superseded-only legacy evidence while verifying non-failing source-to-stable migration notices for current and superseded records",
+  "timestamp": "2026-08-03T15:55:36.181Z"
+}
diff --git a/.2119/verdicts/REQ-012.6.1.json b/.2119/verdicts/REQ-012.6.1.json
new file mode 100644
index 0000000..df037f9
--- /dev/null
+++ b/.2119/verdicts/REQ-012.6.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.6.1--159b88d54fa9",
+  "requirementId": "REQ-012.6.1",
+  "hash": "159b88d54fa9",
+  "verdict": "pass",
+  "summary": "The real prune CLI preserves current stable records over conflicting legacy evidence and migrates current legacy pass/fail records byte-for-field to stable paths; malformed near-counterexamples are rejected.",
+  "timestamp": "2026-08-03T17:19:26.772Z"
+}
diff --git a/.2119/verdicts/REQ-012.6.2.json b/.2119/verdicts/REQ-012.6.2.json
new file mode 100644
index 0000000..25a31bb
--- /dev/null
+++ b/.2119/verdicts/REQ-012.6.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.6.2--a847c92259cf",
+  "requirementId": "REQ-012.6.2",
+  "hash": "a847c92259cf",
+  "verdict": "pass",
+  "summary": "Real prune CLI test retains current stable or migrated pass/fail evidence, removes valid/stale/orphan/malformed legacy JSON, reports every exercised migration/prune, and leaves one canonical JSON per requirement.",
+  "timestamp": "2026-08-03T17:19:42.143Z"
+}
diff --git a/.2119/verdicts/REQ-012.7.1.json b/.2119/verdicts/REQ-012.7.1.json
new file mode 100644
index 0000000..4ac7b94
--- /dev/null
+++ b/.2119/verdicts/REQ-012.7.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "REQ-012.7.1--8875ee2f6f6c",
+  "requirementId": "REQ-012.7.1",
+  "hash": "8875ee2f6f6c",
+  "verdict": "pass",
+  "summary": "The annotated init test invokes dist/cli.js in a real Git repository, verifies the exact verdict JSON attribute resolves to true, and proves the file is unignored and stageable.",
+  "timestamp": "2026-08-03T16:10:01.883Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.1.1--3a42292f6705.json b/.2119/verdicts/self-supplied-evidence.1.1--3a42292f6705.json
deleted file mode 100644
index a6c955c..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.1--3a42292f6705.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.1--3a42292f6705",
-  "requirementId": "self-supplied-evidence.1.1",
-  "hash": "3a42292f6705",
-  "verdict": "pass",
-  "summary": "Every production-computed test-quality task is compared against the complete canonical task body, so removing, weakening, or contradicting the concrete-production-failure mandate fails the test.",
-  "timestamp": "2026-08-02T23:10:57.874Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.1--6419602f2307.json b/.2119/verdicts/self-supplied-evidence.1.1--6419602f2307.json
deleted file mode 100644
index 002bebb..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.1--6419602f2307.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.1--6419602f2307",
-  "requirementId": "self-supplied-evidence.1.1",
-  "hash": "6419602f2307",
-  "verdict": "fail",
-  "summary": "The exact mandate is asserted, but the anti-waiver check misses contradictory guidance such as 'A generic category-level failure is sufficient,' which permits a non-concrete answer while the test stays green.",
-  "timestamp": "2026-08-02T23:07:14.248Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.1--7c4774226fe5.json b/.2119/verdicts/self-supplied-evidence.1.1--7c4774226fe5.json
deleted file mode 100644
index bcb0d2a..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.1--7c4774226fe5.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.1--7c4774226fe5",
-  "requirementId": "self-supplied-evidence.1.1",
-  "hash": "7c4774226fe5",
-  "verdict": "pass",
-  "summary": "Every production-identified test-quality instruction is required to ask for a concrete production failure, while tests reject omission, optionality, and contradictory carve-outs.",
-  "timestamp": "2026-08-02T05:32:14.238Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.1--8e33566ca8c8.json b/.2119/verdicts/self-supplied-evidence.1.1--8e33566ca8c8.json
deleted file mode 100644
index c0a1992..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.1--8e33566ca8c8.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.1--8e33566ca8c8",
-  "requirementId": "self-supplied-evidence.1.1",
-  "hash": "8e33566ca8c8",
-  "verdict": "fail",
-  "summary": "The provenance slice is exact, but contradictory guidance appended after the counterexample block can permit a generic failure category while every 1.1 assertion remains green.",
-  "timestamp": "2026-08-02T23:08:33.884Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.1--ef8087c8df76.json b/.2119/verdicts/self-supplied-evidence.1.1--ef8087c8df76.json
deleted file mode 100644
index f018075..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.1--ef8087c8df76.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.1--ef8087c8df76",
-  "requirementId": "self-supplied-evidence.1.1",
-  "hash": "ef8087c8df76",
-  "verdict": "pass",
-  "summary": "Every production-computed test-quality instruction retains the complete canonical task body, so removing, weakening, or contradicting the concrete-production-failure mandate fails the test.",
-  "timestamp": "2026-08-02T23:14:00.257Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.1.json b/.2119/verdicts/self-supplied-evidence.1.1.json
new file mode 100644
index 0000000..f7914fd
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.1.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.1.1--afb9cc135da4",
+  "requirementId": "self-supplied-evidence.1.1",
+  "hash": "afb9cc135da4",
+  "verdict": "pass",
+  "summary": "The CLI-generated packet set is exhaustively exact-body checked and rejects omission or weakening of the concrete covering-test production-failure demand",
+  "timestamp": "2026-08-03T17:05:01.031Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.1.2--419abd7dee19.json b/.2119/verdicts/self-supplied-evidence.1.2--419abd7dee19.json
deleted file mode 100644
index c0c686a..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.2--419abd7dee19.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.2--419abd7dee19",
-  "requirementId": "self-supplied-evidence.1.2",
-  "hash": "419abd7dee19",
-  "verdict": "pass",
-  "summary": "Every production-discovered test-quality instruction is compared against the complete expected task body with only dynamic IDs normalized, so removing, weakening, or contradicting the file:line independence demand fails.",
-  "timestamp": "2026-08-02T23:10:46.134Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.2--4f90050b0b09.json b/.2119/verdicts/self-supplied-evidence.1.2--4f90050b0b09.json
deleted file mode 100644
index 66d9520..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.2--4f90050b0b09.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.2--4f90050b0b09",
-  "requirementId": "self-supplied-evidence.1.2",
-  "hash": "4f90050b0b09",
-  "verdict": "pass",
-  "summary": "Every production-discovered test-quality instruction is still matched against the complete expected task with only dynamic IDs normalized, so any omission, weakening, or contradiction of the file:line production-reachability demand fails.",
-  "timestamp": "2026-08-02T23:14:09.042Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.2--6331058954ea.json b/.2119/verdicts/self-supplied-evidence.1.2--6331058954ea.json
deleted file mode 100644
index 7535f16..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.2--6331058954ea.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.2--6331058954ea",
-  "requirementId": "self-supplied-evidence.1.2",
-  "hash": "6331058954ea",
-  "verdict": "pass",
-  "summary": "All generated test-quality tasks are exhaustively selected from production review targets and require file:line production reachability for the named failure, with exact independence from test, fixture, and prompt-supplied triggers or decisive observations and explicit rejection of opt-outs.",
-  "timestamp": "2026-08-02T05:32:24.495Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.2--78804c69a5b9.json b/.2119/verdicts/self-supplied-evidence.1.2--78804c69a5b9.json
deleted file mode 100644
index 5df7911..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.2--78804c69a5b9.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.2--78804c69a5b9",
-  "requirementId": "self-supplied-evidence.1.2",
-  "hash": "78804c69a5b9",
-  "verdict": "fail",
-  "summary": "The test stays green if the provenance section retains the exact mandate but adds 'Evidence from test fixtures satisfies this demand.'; that unrecognized contradiction permits fixture-supplied evidence despite 1.2's independence clause.",
-  "timestamp": "2026-08-02T23:07:12.165Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.2--82dcac5dc17c.json b/.2119/verdicts/self-supplied-evidence.1.2--82dcac5dc17c.json
deleted file mode 100644
index dda4be0..0000000
--- a/.2119/verdicts/self-supplied-evidence.1.2--82dcac5dc17c.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.1.2--82dcac5dc17c",
-  "requirementId": "self-supplied-evidence.1.2",
-  "hash": "82dcac5dc17c",
-  "verdict": "fail",
-  "summary": "The exact comparison ends at '**Counterexample obligation:**'; appending later in the same instruction 'Evidence from test fixtures satisfies the production-provenance demand.' leaves every assertion green while negating 1.2's independence requirement.",
-  "timestamp": "2026-08-02T23:08:31.147Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.1.2.json b/.2119/verdicts/self-supplied-evidence.1.2.json
new file mode 100644
index 0000000..d8dd0cb
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.1.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.1.2--c3c97de002d9",
+  "requirementId": "self-supplied-evidence.1.2",
+  "hash": "c3c97de002d9",
+  "verdict": "pass",
+  "summary": "Every CLI-generated test-quality packet is exact-body checked for file:line reachability independent of tests, fixtures, prompts, triggers, and decisive observations",
+  "timestamp": "2026-08-03T17:05:01.162Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.2.1--1fc1ddf5c224.json b/.2119/verdicts/self-supplied-evidence.2.1--1fc1ddf5c224.json
deleted file mode 100644
index 94d303c..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.1--1fc1ddf5c224.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.1--1fc1ddf5c224",
-  "requirementId": "self-supplied-evidence.2.1",
-  "hash": "1fc1ddf5c224",
-  "verdict": "pass",
-  "summary": "The real review-dispatch CLI generates every computed test-quality instruction, and the test rejects omission or alteration of consumption, emitted value, separate invocation, production component, or production data-source terms as well as weakening exceptions.",
-  "timestamp": "2026-08-02T23:07:04.908Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.1--2aa103ac58bd.json b/.2119/verdicts/self-supplied-evidence.2.1--2aa103ac58bd.json
deleted file mode 100644
index 6b2b59f..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.1--2aa103ac58bd.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.1--2aa103ac58bd",
-  "requirementId": "self-supplied-evidence.2.1",
-  "hash": "2aa103ac58bd",
-  "verdict": "pass",
-  "summary": "All generated test-quality tasks are enumerated from production review targets and must contain the exact narrow boundary definition; omission or alteration of consumption, emitted value, separate invocation, or production component/data-source terms fails, while anti-waiver assertions reject contradictory weakening.",
-  "timestamp": "2026-08-02T05:32:29.175Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.1--611ec8687d7d.json b/.2119/verdicts/self-supplied-evidence.2.1--611ec8687d7d.json
deleted file mode 100644
index 796d3bd..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.1--611ec8687d7d.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.1--611ec8687d7d",
-  "requirementId": "self-supplied-evidence.2.1",
-  "hash": "611ec8687d7d",
-  "verdict": "pass",
-  "summary": "Every computed production-generated test-quality task is compared with a complete oracle after only dynamic ID normalization, so changing consumption, emitted value, separate invocation, production-component, or production-data-source scope fails.",
-  "timestamp": "2026-08-02T23:14:10.359Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.1--7da77538aed0.json b/.2119/verdicts/self-supplied-evidence.2.1--7da77538aed0.json
deleted file mode 100644
index fc9bc91..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.1--7da77538aed0.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.1--7da77538aed0",
-  "requirementId": "self-supplied-evidence.2.1",
-  "hash": "7da77538aed0",
-  "verdict": "pass",
-  "summary": "Every production-dispatched test-quality instruction must match an independent full provenance oracle, so altering any producer/consumer boundary conjunct—consumption, emitted value, separate invocation, production component, or production data source—fails.",
-  "timestamp": "2026-08-02T23:08:26.294Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.1--a000c7a0aa4b.json b/.2119/verdicts/self-supplied-evidence.2.1--a000c7a0aa4b.json
deleted file mode 100644
index 38751c4..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.1--a000c7a0aa4b.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.1--a000c7a0aa4b",
-  "requirementId": "self-supplied-evidence.2.1",
-  "hash": "a000c7a0aa4b",
-  "verdict": "pass",
-  "summary": "Every generated test-quality task is matched against a complete independent task oracle after normalizing only requirement/review IDs, so any change to the producer/consumer definition's consumption, emitted value, separate invocation, production component, or data-source scope fails.",
-  "timestamp": "2026-08-02T23:10:44.573Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.1.json b/.2119/verdicts/self-supplied-evidence.2.1.json
new file mode 100644
index 0000000..ffd3296
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.2.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.2.1--7fbcc5496008",
+  "requirementId": "self-supplied-evidence.2.1",
+  "hash": "7fbcc5496008",
+  "verdict": "pass",
+  "summary": "Every CLI-generated test-quality packet is exact-body checked for consumption of emitted values from separately invoked production components or data sources",
+  "timestamp": "2026-08-03T17:05:01.293Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.2.2--156362cfdb04.json b/.2119/verdicts/self-supplied-evidence.2.2--156362cfdb04.json
deleted file mode 100644
index 429ed15..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.2--156362cfdb04.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.2--156362cfdb04",
-  "requirementId": "self-supplied-evidence.2.2",
-  "hash": "156362cfdb04",
-  "verdict": "fail",
-  "summary": "The assertion checks only the citation clause substring: changing its trigger to 'Unless that boundary exists' leaves the test and anti-waiver regexes green while no longer requiring evidence whenever a producer/consumer boundary exists.",
-  "timestamp": "2026-08-02T23:07:37.188Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.2--56dbe92c6003.json b/.2119/verdicts/self-supplied-evidence.2.2--56dbe92c6003.json
deleted file mode 100644
index 744e1b8..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.2--56dbe92c6003.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.2--56dbe92c6003",
-  "requirementId": "self-supplied-evidence.2.2",
-  "hash": "56dbe92c6003",
-  "verdict": "pass",
-  "summary": "Every production-computed test-quality task pins the producer/consumer condition and requires file:line evidence that the covering test obtains its input from that production producer.",
-  "timestamp": "2026-08-02T23:14:23.989Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.2--8fa7baeaefb2.json b/.2119/verdicts/self-supplied-evidence.2.2--8fa7baeaefb2.json
deleted file mode 100644
index 406bb51..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.2--8fa7baeaefb2.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.2--8fa7baeaefb2",
-  "requirementId": "self-supplied-evidence.2.2",
-  "hash": "8fa7baeaefb2",
-  "verdict": "pass",
-  "summary": "The test inspects every generated test-quality task, requires mandatory file:line evidence that its input comes from the production producer, and rejects waiver or self-supplied-evidence language.",
-  "timestamp": "2026-08-02T05:33:28.165Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.2--a3e153ca68e4.json b/.2119/verdicts/self-supplied-evidence.2.2--a3e153ca68e4.json
deleted file mode 100644
index 292ae94..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.2--a3e153ca68e4.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.2--a3e153ca68e4",
-  "requirementId": "self-supplied-evidence.2.2",
-  "hash": "a3e153ca68e4",
-  "verdict": "pass",
-  "summary": "Each generated test-quality task must match the complete normalized task oracle, which pins the existing-boundary trigger, file:line citation, covering test input, and production producer together and rejects reversal, omission, or weakening.",
-  "timestamp": "2026-08-02T23:11:02.928Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.2--df4ea1367363.json b/.2119/verdicts/self-supplied-evidence.2.2--df4ea1367363.json
deleted file mode 100644
index 3b10900..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.2--df4ea1367363.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.2--df4ea1367363",
-  "requirementId": "self-supplied-evidence.2.2",
-  "hash": "df4ea1367363",
-  "verdict": "pass",
-  "summary": "The independent full-block oracle now pins 'If that boundary exists' together with file:line citation, test input, and that producer for every production-dispatched test-quality instruction, rejecting reversed or weakened triggers.",
-  "timestamp": "2026-08-02T23:08:52.297Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.2.json b/.2119/verdicts/self-supplied-evidence.2.2.json
new file mode 100644
index 0000000..58b5593
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.2.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.2.2--8796d4a6e793",
+  "requirementId": "self-supplied-evidence.2.2",
+  "hash": "8796d4a6e793",
+  "verdict": "pass",
+  "summary": "Built CLI dispatch output is read from generated files; whole-task equality rejects omission of file:line, conditional-boundary, or production-producer input provenance wording.",
+  "timestamp": "2026-08-03T17:05:06.235Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.2.3--3bfc1f157e4c.json b/.2119/verdicts/self-supplied-evidence.2.3--3bfc1f157e4c.json
deleted file mode 100644
index 7b8eaec..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.3--3bfc1f157e4c.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.3--3bfc1f157e4c",
-  "requirementId": "self-supplied-evidence.2.3",
-  "hash": "3bfc1f157e4c",
-  "verdict": "pass",
-  "summary": "The generated-task test covers every production-derived test-quality target and rejects omission or weakening of the required file:line trace that the exercised input preserves the production producer's value shape.",
-  "timestamp": "2026-08-02T05:33:56.135Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.3--a0e14d8db077.json b/.2119/verdicts/self-supplied-evidence.2.3--a0e14d8db077.json
deleted file mode 100644
index 8d901aa..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.3--a0e14d8db077.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.3--a0e14d8db077",
-  "requirementId": "self-supplied-evidence.2.3",
-  "hash": "a0e14d8db077",
-  "verdict": "pass",
-  "summary": "Every production-discovered test-quality task is full-body matched, pinning the conditional file:line demand that the covering test's exercised input preserve the production producer's value shape.",
-  "timestamp": "2026-08-02T23:14:29.592Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.3--b768f1b9b787.json b/.2119/verdicts/self-supplied-evidence.2.3--b768f1b9b787.json
deleted file mode 100644
index 4271f08..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.3--b768f1b9b787.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.3--b768f1b9b787",
-  "requirementId": "self-supplied-evidence.2.3",
-  "hash": "b768f1b9b787",
-  "verdict": "pass",
-  "summary": "Every production-discovered test-quality instruction is full-body matched, including the conditional file:line demand that the exercised value preserve the production producer's shape; omissions, weakenings, and contradictions fail.",
-  "timestamp": "2026-08-02T23:11:13.182Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.3.json b/.2119/verdicts/self-supplied-evidence.2.3.json
new file mode 100644
index 0000000..df7fae3
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.2.3.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.2.3--3b0f7cd77ddc",
+  "requirementId": "self-supplied-evidence.2.3",
+  "hash": "3b0f7cd77ddc",
+  "verdict": "pass",
+  "summary": "Built CLI dispatch output is read without reshaping; whole-task equality rejects omission or weakening of the conditional production-shape evidence requirement.",
+  "timestamp": "2026-08-03T17:05:06.321Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.2.4--2755a46ec65c.json b/.2119/verdicts/self-supplied-evidence.2.4--2755a46ec65c.json
deleted file mode 100644
index 1382bcc..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.4--2755a46ec65c.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.4--2755a46ec65c",
-  "requirementId": "self-supplied-evidence.2.4",
-  "hash": "2755a46ec65c",
-  "verdict": "pass",
-  "summary": "Fresh dispatch integration verifies every generated test-quality task explicitly requires file:line evidence distinguishing a newly produced observation from an equal pre-existing sentinel, while shared rejection checks forbid weakening language.",
-  "timestamp": "2026-08-02T05:34:49.897Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.4--2e7bb01145d0.json b/.2119/verdicts/self-supplied-evidence.2.4--2e7bb01145d0.json
deleted file mode 100644
index bcb1b12..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.4--2e7bb01145d0.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.4--2e7bb01145d0",
-  "requirementId": "self-supplied-evidence.2.4",
-  "hash": "2e7bb01145d0",
-  "verdict": "pass",
-  "summary": "Every computed production-generated test-quality task matches the complete oracle, pinning the decisive-observation trigger, all four initial/default/placeholder/sentinel cases, file:line evidence, and newly-produced versus pre-existing distinction.",
-  "timestamp": "2026-08-02T23:14:34.728Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.4--3add2f8a8874.json b/.2119/verdicts/self-supplied-evidence.2.4--3add2f8a8874.json
deleted file mode 100644
index 076930f..0000000
--- a/.2119/verdicts/self-supplied-evidence.2.4--3add2f8a8874.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.2.4--3add2f8a8874",
-  "requirementId": "self-supplied-evidence.2.4",
-  "hash": "3add2f8a8874",
-  "verdict": "pass",
-  "summary": "The complete task snapshot pins the conditional trigger, all four pre-existing-value classes, file:line evidence, and the newly-produced-versus-pre-existing distinction for every production-computed test-quality target.",
-  "timestamp": "2026-08-02T23:11:20.691Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.2.4.json b/.2119/verdicts/self-supplied-evidence.2.4.json
new file mode 100644
index 0000000..38e223b
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.2.4.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.2.4--9ea89fc088ad",
+  "requirementId": "self-supplied-evidence.2.4",
+  "hash": "9ea89fc088ad",
+  "verdict": "pass",
+  "summary": "Built CLI dispatch output is freshly generated and whole-task equality rejects loss of any initial/default/placeholder/sentinel case or new-versus-pre-existing distinction.",
+  "timestamp": "2026-08-03T17:05:06.404Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.3.1--0c0cd9b96a10.json b/.2119/verdicts/self-supplied-evidence.3.1--0c0cd9b96a10.json
deleted file mode 100644
index 67f8bab..0000000
--- a/.2119/verdicts/self-supplied-evidence.3.1--0c0cd9b96a10.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.3.1--0c0cd9b96a10",
-  "requirementId": "self-supplied-evidence.3.1",
-  "hash": "0c0cd9b96a10",
-  "verdict": "pass",
-  "summary": "Every production-computed test-quality task exactly defines the boundary as invocation of a binary or service outside the gate's own process, with the complete task snapshot rejecting weakened definitions.",
-  "timestamp": "2026-08-02T23:14:48.392Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.3.1--8f4768f6eda9.json b/.2119/verdicts/self-supplied-evidence.3.1--8f4768f6eda9.json
deleted file mode 100644
index e061bc3..0000000
--- a/.2119/verdicts/self-supplied-evidence.3.1--8f4768f6eda9.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.3.1--8f4768f6eda9",
-  "requirementId": "self-supplied-evidence.3.1",
-  "hash": "8f4768f6eda9",
-  "verdict": "pass",
-  "summary": "Enumerates every production-derived test-quality task and rejects missing or altered boundary wording, including inside-process and non-binary/service near-counterexamples, via an exact required definition.",
-  "timestamp": "2026-08-02T05:36:17.365Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.3.1--ba8be2d406ca.json b/.2119/verdicts/self-supplied-evidence.3.1--ba8be2d406ca.json
deleted file mode 100644
index 28ef528..0000000
--- a/.2119/verdicts/self-supplied-evidence.3.1--ba8be2d406ca.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.3.1--ba8be2d406ca",
-  "requirementId": "self-supplied-evidence.3.1",
-  "hash": "ba8be2d406ca",
-  "verdict": "pass",
-  "summary": "Every production-generated test-quality task must match the complete normalized oracle, so changing invocation, binary-or-service scope, or outside-the-gate-process boundary wording fails.",
-  "timestamp": "2026-08-02T23:11:23.747Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.3.1.json b/.2119/verdicts/self-supplied-evidence.3.1.json
new file mode 100644
index 0000000..db5103b
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.3.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.3.1--16a37e67b6e3",
+  "requirementId": "self-supplied-evidence.3.1",
+  "hash": "16a37e67b6e3",
+  "verdict": "pass",
+  "summary": "Built CLI dispatch output is exercised; whole-task equality rejects definitions omitting binary/service, external invocation, or the gate-process boundary.",
+  "timestamp": "2026-08-03T17:05:06.489Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.3.2--401f4b68deae.json b/.2119/verdicts/self-supplied-evidence.3.2--401f4b68deae.json
deleted file mode 100644
index 37faad6..0000000
--- a/.2119/verdicts/self-supplied-evidence.3.2--401f4b68deae.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.3.2--401f4b68deae",
-  "requirementId": "self-supplied-evidence.3.2",
-  "hash": "401f4b68deae",
-  "verdict": "pass",
-  "summary": "Every production-discovered test-quality task is full-body matched, pinning the conditional file:line demand for both the dependency's production provisioning declaration and its production absence-failure path.",
-  "timestamp": "2026-08-02T23:14:51.665Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.3.2--5f8200b2fdd0.json b/.2119/verdicts/self-supplied-evidence.3.2--5f8200b2fdd0.json
deleted file mode 100644
index 2d27223..0000000
--- a/.2119/verdicts/self-supplied-evidence.3.2--5f8200b2fdd0.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.3.2--5f8200b2fdd0",
-  "requirementId": "self-supplied-evidence.3.2",
-  "hash": "5f8200b2fdd0",
-  "verdict": "pass",
-  "summary": "PASS: every generated test-quality task is checked for an exact conditional instruction requiring file:line evidence of both production provisioning and the production absence-failure path; omitting the boundary condition, either conjunct, or file:line demand fails.",
-  "timestamp": "2026-08-02T05:36:21.818Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.3.2--781809753ed3.json b/.2119/verdicts/self-supplied-evidence.3.2--781809753ed3.json
deleted file mode 100644
index b4400d7..0000000
--- a/.2119/verdicts/self-supplied-evidence.3.2--781809753ed3.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.3.2--781809753ed3",
-  "requirementId": "self-supplied-evidence.3.2",
-  "hash": "781809753ed3",
-  "verdict": "pass",
-  "summary": "Every production-discovered test-quality instruction is full-body matched, pinning the conditional file:line demand for both the production provisioning declaration and the production path that fails when the external dependency is absent.",
-  "timestamp": "2026-08-02T23:11:33.535Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.3.2.json b/.2119/verdicts/self-supplied-evidence.3.2.json
new file mode 100644
index 0000000..4dabd20
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.3.2.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.3.2--e873b585f1a6",
+  "requirementId": "self-supplied-evidence.3.2",
+  "hash": "e873b585f1a6",
+  "verdict": "pass",
+  "summary": "Built CLI dispatch output is exercised; whole-task equality rejects omission of the conditional boundary, production provisioning declaration, absent-dependency failure path, or either half of both.",
+  "timestamp": "2026-08-03T17:05:06.568Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.4.1--02d88bd20370.json b/.2119/verdicts/self-supplied-evidence.4.1--02d88bd20370.json
deleted file mode 100644
index 33fce7f..0000000
--- a/.2119/verdicts/self-supplied-evidence.4.1--02d88bd20370.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.4.1--02d88bd20370",
-  "requirementId": "self-supplied-evidence.4.1",
-  "hash": "02d88bd20370",
-  "verdict": "pass",
-  "summary": "Every computed production-generated test-quality task matches the complete oracle, pinning a FAIL for either absent applicable provenance or evidence that production cannot produce the claimed failure independently of test setup.",
-  "timestamp": "2026-08-02T23:14:55.202Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.4.1--38564acf58d7.json b/.2119/verdicts/self-supplied-evidence.4.1--38564acf58d7.json
deleted file mode 100644
index 1bb792a..0000000
--- a/.2119/verdicts/self-supplied-evidence.4.1--38564acf58d7.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.4.1--38564acf58d7",
-  "requirementId": "self-supplied-evidence.4.1",
-  "hash": "38564acf58d7",
-  "verdict": "pass",
-  "summary": "The complete task snapshot requires Record FAIL for both absent applicable provenance evidence and evidence that production cannot produce the failure independently of test setup across every production-computed test-quality target.",
-  "timestamp": "2026-08-02T23:11:40.674Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.4.1--dbb07998ba99.json b/.2119/verdicts/self-supplied-evidence.4.1--dbb07998ba99.json
deleted file mode 100644
index 1b4c3b3..0000000
--- a/.2119/verdicts/self-supplied-evidence.4.1--dbb07998ba99.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.4.1--dbb07998ba99",
-  "requirementId": "self-supplied-evidence.4.1",
-  "hash": "dbb07998ba99",
-  "verdict": "pass",
-  "summary": "Generated instructions are checked across every production-derived test-quality target and must explicitly record FAIL for either missing provenance or evidence that production cannot independently produce the failure; omission, weakened wording, and escape-clause counterexamples are rejected.",
-  "timestamp": "2026-08-02T05:36:27.004Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.4.1.json b/.2119/verdicts/self-supplied-evidence.4.1.json
new file mode 100644
index 0000000..64873e2
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.4.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.4.1--c246036538fc",
+  "requirementId": "self-supplied-evidence.4.1",
+  "hash": "c246036538fc",
+  "verdict": "pass",
+  "summary": "Generated test-quality packets from the spawned production CLI retain the explicit fail-on-absent-or-self-supplied-provenance direction for every parsed target.",
+  "timestamp": "2026-08-03T17:05:04.663Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.5.1--15d76cbfb0f1.json b/.2119/verdicts/self-supplied-evidence.5.1--15d76cbfb0f1.json
deleted file mode 100644
index 32d7281..0000000
--- a/.2119/verdicts/self-supplied-evidence.5.1--15d76cbfb0f1.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.5.1--15d76cbfb0f1",
-  "requirementId": "self-supplied-evidence.5.1",
-  "hash": "15d76cbfb0f1",
-  "verdict": "fail",
-  "summary": "The test inspects only REQ-003.4.2 instead of every computed direct-judgment target; adding provenance questions only to another generated [review] instruction remains green, and rephrased provenance questions can evade the phrase blacklist.",
-  "timestamp": "2026-08-02T23:12:01.145Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.5.1--4bf3f2230530.json b/.2119/verdicts/self-supplied-evidence.5.1--4bf3f2230530.json
deleted file mode 100644
index abfbe7a..0000000
--- a/.2119/verdicts/self-supplied-evidence.5.1--4bf3f2230530.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.5.1--4bf3f2230530",
-  "requirementId": "self-supplied-evidence.5.1",
-  "hash": "4bf3f2230530",
-  "verdict": "pass",
-  "summary": "Every production-computed direct-judgment target is compared with the complete canonical direct task, which contains no test-quality provenance questionnaire and fails on any insertion.",
-  "timestamp": "2026-08-02T23:15:24.252Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.5.1--e7eb54958e86.json b/.2119/verdicts/self-supplied-evidence.5.1--e7eb54958e86.json
deleted file mode 100644
index 0c79deb..0000000
--- a/.2119/verdicts/self-supplied-evidence.5.1--e7eb54958e86.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.5.1--e7eb54958e86",
-  "requirementId": "self-supplied-evidence.5.1",
-  "hash": "e7eb54958e86",
-  "verdict": "pass",
-  "summary": "The test exercises a real generated [review] direct-judgment instruction and rejects inclusion of every specified test-quality provenance question category while confirming the direct-judgment prompt remains present.",
-  "timestamp": "2026-08-02T05:38:41.022Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.5.1.json b/.2119/verdicts/self-supplied-evidence.5.1.json
new file mode 100644
index 0000000..0f9af7f
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.5.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.5.1--c1a086a4ad44",
+  "requirementId": "self-supplied-evidence.5.1",
+  "hash": "c1a086a4ad44",
+  "verdict": "pass",
+  "summary": "Generated direct-judgment packets for parsed [review] targets exactly retain the direct task and omit the test-quality provenance section.",
+  "timestamp": "2026-08-03T17:05:04.793Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.6.1--140a2a47aa19.json b/.2119/verdicts/self-supplied-evidence.6.1--140a2a47aa19.json
deleted file mode 100644
index e1fc4f6..0000000
--- a/.2119/verdicts/self-supplied-evidence.6.1--140a2a47aa19.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.6.1--140a2a47aa19",
-  "requirementId": "self-supplied-evidence.6.1",
-  "hash": "140a2a47aa19",
-  "verdict": "pass",
-  "summary": "PASS: the real lint workflow exits clean on a covering test containing both literal config input and a factory-built fixture, so either syntax-triggered provenance violation would fail the test.",
-  "timestamp": "2026-08-02T21:27:52.554Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.6.1--68dbeff8af0f.json b/.2119/verdicts/self-supplied-evidence.6.1--68dbeff8af0f.json
deleted file mode 100644
index fe7c9b3..0000000
--- a/.2119/verdicts/self-supplied-evidence.6.1--68dbeff8af0f.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.6.1--68dbeff8af0f",
-  "requirementId": "self-supplied-evidence.6.1",
-  "hash": "68dbeff8af0f",
-  "verdict": "pass",
-  "summary": "The production lint CLI scans copied annotated tests containing both literal constructions and factory functions, and the test requires status 0 plus the exact clean-spec count, so a syntax-only provenance violation fails.",
-  "timestamp": "2026-08-02T23:12:28.571Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.6.1--978277912461.json b/.2119/verdicts/self-supplied-evidence.6.1--978277912461.json
deleted file mode 100644
index 1a3721e..0000000
--- a/.2119/verdicts/self-supplied-evidence.6.1--978277912461.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.6.1--978277912461",
-  "requirementId": "self-supplied-evidence.6.1",
-  "hash": "978277912461",
-  "verdict": "pass",
-  "summary": "The production lint CLI scans an otherwise clean fixture whose copied covering tests contain both literal construction and factory syntax; any provenance violation inferred solely from either syntax makes the asserted zero status fail.",
-  "timestamp": "2026-08-02T23:15:15.194Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.6.1.json b/.2119/verdicts/self-supplied-evidence.6.1.json
new file mode 100644
index 0000000..b5c2350
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.6.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.6.1--9eff2a724166",
+  "requirementId": "self-supplied-evidence.6.1",
+  "hash": "9eff2a724166",
+  "verdict": "pass",
+  "summary": "Built lint accepts isolated string/object/array/template/regexp/number/bigint/boolean/null inputs and named/static/imported/arrow factory forms without provenance violations.",
+  "timestamp": "2026-08-03T17:07:21.073Z"
+}
diff --git a/.2119/verdicts/self-supplied-evidence.7.1--5676f3ef50cb.json b/.2119/verdicts/self-supplied-evidence.7.1--5676f3ef50cb.json
deleted file mode 100644
index 54a9a8b..0000000
--- a/.2119/verdicts/self-supplied-evidence.7.1--5676f3ef50cb.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.7.1--5676f3ef50cb",
-  "requirementId": "self-supplied-evidence.7.1",
-  "hash": "5676f3ef50cb",
-  "verdict": "pass",
-  "summary": "All mechanically computed standard targets match complete kind-specific task oracles, and audit filenames exactly equal the computed passing-target set with every audit matching its full oracle; each pins concrete names, cardinality, and the category-promotion prohibition.",
-  "timestamp": "2026-08-02T23:15:20.011Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.7.1--a75c8b3b6ac2.json b/.2119/verdicts/self-supplied-evidence.7.1--a75c8b3b6ac2.json
deleted file mode 100644
index 3ff6975..0000000
--- a/.2119/verdicts/self-supplied-evidence.7.1--a75c8b3b6ac2.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.7.1--a75c8b3b6ac2",
-  "requirementId": "self-supplied-evidence.7.1",
-  "hash": "a75c8b3b6ac2",
-  "verdict": "fail",
-  "summary": "Standard and audit instructions are covered, but contradictory guidance such as 'A category-level summary is sufficient for member-specific evidence' bypasses the negative regex and leaves every assertion green.",
-  "timestamp": "2026-08-02T23:12:16.515Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.7.1--ae376b3595ab.json b/.2119/verdicts/self-supplied-evidence.7.1--ae376b3595ab.json
deleted file mode 100644
index 18ce592..0000000
--- a/.2119/verdicts/self-supplied-evidence.7.1--ae376b3595ab.json
+++ /dev/null
@@ -1,8 +0,0 @@
-{
-  "reviewId": "self-supplied-evidence.7.1--ae376b3595ab",
-  "requirementId": "self-supplied-evidence.7.1",
-  "hash": "ae376b3595ab",
-  "verdict": "pass",
-  "summary": "The fixture rejects missing or weakened evidence-bounded wording for every computed standard target and exactly every audit generated from its computed passing-verdict targets.",
-  "timestamp": "2026-08-02T21:30:44.081Z"
-}
diff --git a/.2119/verdicts/self-supplied-evidence.7.1.json b/.2119/verdicts/self-supplied-evidence.7.1.json
new file mode 100644
index 0000000..8368652
--- /dev/null
+++ b/.2119/verdicts/self-supplied-evidence.7.1.json
@@ -0,0 +1,8 @@
+{
+  "reviewId": "self-supplied-evidence.7.1--22a18023494c",
+  "requirementId": "self-supplied-evidence.7.1",
+  "hash": "22a18023494c",
+  "verdict": "pass",
+  "summary": "Every generated standard packet and every producible audit packet retains final tool-authored guidance preserving concrete member names and singular/plural scope; operator custom criteria are separately delimited before the task.",
+  "timestamp": "2026-08-03T17:05:05.044Z"
+}
diff --git a/.gitattributes b/.gitattributes
new file mode 100644
index 0000000..75ad072
--- /dev/null
+++ b/.gitattributes
@@ -0,0 +1 @@
+.2119/verdicts/*.json linguist-generated=true
diff --git a/specs/REQ-003-judgment-reviews.md b/specs/REQ-003-judgment-reviews.md
index 5b4162f..682b3c0 100644
--- a/specs/REQ-003-judgment-reviews.md
+++ b/specs/REQ-003-judgment-reviews.md
@@ -76,7 +76,7 @@ accidental corruption (mangled merges, hand-edits) silently reading as green.
 
 1. A verdict record MUST be counted by the gate only when its `verdict` field is exactly `pass` or `fail`, its `summary` is a nonempty string, its `requirementId` equals the requirement component of its `reviewId`, its `hash` equals the review ID's 12-character suffix, and its `timestamp` parses as a date.
 2. A verdict file that fails to parse or whose record fails validation MUST produce a check violation identifying the file and the reason, rather than being silently skipped or counted as passing.
-3. A verdict file whose filename is not exactly `<reviewId>.json` MUST be treated as malformed, so the check gate and `2119 prune` agree on which file a verdict lives in.
+3. REQUIREMENT REMOVED
 
 ### REQ-003.8: Reviewer calibration corpus
 
diff --git a/specs/REQ-006-maintenance.md b/specs/REQ-006-maintenance.md
index 99d72f7..e1591a9 100644
--- a/specs/REQ-006-maintenance.md
+++ b/specs/REQ-006-maintenance.md
@@ -27,4 +27,4 @@ a side effect.
 ### REQ-006.2: Verdict pruning
 
 1. `2119 prune` MUST delete every verdict file whose review ID does not match any current review target, listing each deleted file.
-2. `2119 prune` MUST leave verdicts that match current review targets untouched, so a passing `2119 check` still passes after a prune.
+2. `2119 prune` MUST preserve the fields of a well-formed verdict record that matches a current review target, even when migrating that record to its canonical path, so a passing `2119 check` still passes after a prune.
diff --git a/specs/REQ-012-stable-verdict-files.md b/specs/REQ-012-stable-verdict-files.md
new file mode 100644
index 0000000..3649c62
--- /dev/null
+++ b/specs/REQ-012-stable-verdict-files.md
@@ -0,0 +1,79 @@
+# REQ-012: Stable Verdict Files
+
+## Overview
+
+A judgment verdict is evidence about one requirement's review input at one
+specific content state. The existing storage name, `<review-id>.json`, embeds
+that state's hash in the filename. Re-review therefore creates another file
+instead of superseding the obsolete assertion. Repositories accumulate many
+records that no longer describe current review input, even though only the
+record matching the current hash can satisfy the gate.
+
+This feature makes the current assertion explicit and singular. Each
+requirement has one stable verdict path, while the record carries the content
+hash that `check` compares with the hash computed from the current review
+input. Superseded versions remain available in git history, where obsolete
+versions of other tracked files already live.
+
+The content boundary remains the boundary of the judgment being made. For a
+test-quality review, the review input is the requirement statement, its
+covering annotation evidence blocks, and configured shared evidence, as
+defined by REQ-003.1.2, REQ-003.1.7, and REQ-003.1.8. For a `[review]`
+requirement, it is the requirement statement, matched evidence files, and any
+custom review instructions, as defined by REQ-003.1.3 and REQ-005.1.2. This
+avoids invalidating every verdict for an unrelated repository edit without
+allowing an edit to material the reviewer judged to retain a pass.
+
+Existing repositories need a safe transition from files named
+`<requirement-id>--<hash12>.json`. A legacy verdict whose hash matches the
+current review input remains acceptable during migration, but is identified
+as legacy in command output. A stale legacy verdict is identified and never
+counts as evidence. When a stable path exists, it is authoritative during
+`check`; an old legacy pass cannot bypass a malformed, stale, or failing
+stable record. The explicit `prune` command preserves a valid current stable
+record when present, otherwise recovers valid current evidence from the
+legacy set, and then removes superseded legacy files. `check` remains
+non-mutating.
+
+This contract supersedes the former REQ-003.7.3 filename rule, which required
+the content-addressed review ID in the filename. That rule was removed because
+it directly conflicts with the stable per-requirement path; record validation
+continues to enforce agreement among the stable filename, requirement ID,
+review ID, and stored hash.
+
+Verdicts remain committed evidence and remain mandatory. Stable storage does
+not reduce review coverage, skip judgment, or relax any gate. Marking the
+remaining JSON as generated changes only GitHub's presentation of those
+records.
+
+## Requirements
+
+### REQ-012.1: Stable recording
+
+1. After any successful `2119 pass` or `2119 fail` invocations, `.2119/verdicts/` MUST contain exactly one verdict JSON for each requirement those invocations recorded, at `<canonical-requirement-id>.json`, with a later invocation for the same requirement replacing its prior stable or legacy verdicts regardless of content hash.
+
+### REQ-012.2: Concurrent isolation
+
+1. When verdict-recording commands for distinct requirements are launched without waiting for either command to finish, each successful command MUST leave a well-formed stable record for its own requirement without overwriting, deleting, or corrupting the other command's record.
+
+### REQ-012.3: Explicit content binding
+
+1. A well-formed stable verdict record MUST name its canonical requirement ID and review ID, store a lowercase 12-hex content hash equal to that review ID's suffix, store a `pass` or `fail` result with a nonempty summary and parseable timestamp, and reside at `.2119/verdicts/<canonical-requirement-id>.json`.
+2. The stored content hash MUST be the hash of exactly the review input already defined for that target: the inputs and ordering in REQ-003.1.2, REQ-003.1.7, and REQ-003.1.8 for test-quality reviews, or the inputs in REQ-003.1.3 and REQ-005.1.2 for `[review]` reviews, so any change within that input changes the hash while unrelated repository content does not.
+
+### REQ-012.4: Stable freshness gate
+
+1. When a stable path exists for a reviewed requirement, `2119 check` MUST treat only a well-formed stable `pass` whose stored hash equals the current review ID's hash as satisfying that requirement, reporting malformed, stale, or current-failing stable state as a check violation without falling back to any legacy record.
+
+### REQ-012.5: Legacy transition in check
+
+1. When no stable path exists for a review target, `2119 check` MUST validate every matching legacy `<canonical-requirement-id>--<hash12>.json` file, allow only a well-formed current legacy `pass` to satisfy the gate, report malformed legacy records as violations, and emit non-failing migration notices for current or superseded well-formed legacy records that identify their source and stable destination without counting superseded records as evidence.
+
+### REQ-012.6: Explicit migration
+
+1. For each requirement, `2119 prune` MUST retain a well-formed current stable record when one exists, otherwise preserve a well-formed current legacy record at the stable path with all verdict fields unchanged.
+2. `2119 prune` MUST remove every legacy verdict file after selecting the evidence to retain, list each conversion or removal, and leave at most one verdict JSON per requirement.
+
+### REQ-012.7: Generated-file presentation
+
+1. `2119 init` MUST install a Git attribute that marks `.2119/verdicts/*.json` as `linguist-generated=true` while the verdict path remains eligible for Git tracking.
diff --git a/src/changed.ts b/src/changed.ts
index cff1d60..c27e4bb 100644
--- a/src/changed.ts
+++ b/src/changed.ts
@@ -8,7 +8,7 @@ import { evidenceBlockParts } from "./annotations.js";
 import { matchGlobs } from "./files.js";
 import { allRequirements } from "./spec.js";
 import { fileParts, splitReviewId } from "./hash.js";
-import { VERDICTS_DIR } from "./verdict.js";
+import { SAFE_REQUIREMENT_ID, VERDICTS_DIR } from "./verdict.js";
 import { runVerifyCommands, VERIFY_TIMEOUT_MS } from "./verify.js";
 import type { Requirement, Violation } from "./model.js";
 
@@ -268,7 +268,8 @@ function scopeContext(
     if (
       parsed &&
       currentRequirements.has(parsed.requirementId) &&
-      (currentReviewIds.get(parsed.requirementId) === parsed.reviewId ||
+      (parsed.reviewId === undefined ||
+        currentReviewIds.get(parsed.requirementId) === parsed.reviewId ||
         baselineReviewIds.get(parsed.requirementId) === parsed.reviewId)
     ) {
       affected.add(parsed.requirementId);
@@ -294,7 +295,8 @@ function scopeContext(
     const assigned = Boolean(
       parsed &&
         currentRequirements.has(parsed.requirementId) &&
-        (currentReviewIds.get(parsed.requirementId) === parsed.reviewId ||
+        (parsed.reviewId === undefined ||
+          currentReviewIds.get(parsed.requirementId) === parsed.reviewId ||
           baselineReviewIds.get(parsed.requirementId) === parsed.reviewId),
     );
     return assigned ? affected.has(parsed!.requirementId) : changedPaths.has(path);
@@ -319,6 +321,7 @@ function scopeContext(
     lintViolations,
     coverViolations,
     reviewViolations,
+    migrationNotices: current.migrationNotices.filter((notice) => affected.has(notice.requirementId)),
     verifyViolations,
     notInitialized: current.notInitialized && baseline.notInitialized,
     scopedRequirementIds: affected,
@@ -367,8 +370,9 @@ function quotedRequirementId(message: string): string | undefined {
   return message.match(/requirement ID "([^"]+)"/)?.[1];
 }
 
-function verdictReview(path: string): { reviewId: string; requirementId: string } | undefined {
+function verdictReview(path: string): { reviewId?: string; requirementId: string } | undefined {
   const name = basename(path).replace(/\.json$/, "");
   const parsed = splitReviewId(name);
-  return parsed ? { reviewId: name, requirementId: parsed.requirementId } : undefined;
+  if (parsed) return { reviewId: name, requirementId: parsed.requirementId };
+  return SAFE_REQUIREMENT_ID.test(name) ? { requirementId: name } : undefined;
 }
diff --git a/src/check.ts b/src/check.ts
index 22f76ce..d3641d5 100644
--- a/src/check.ts
+++ b/src/check.ts
@@ -6,7 +6,7 @@ import { parseSpec } from "./spec.js";
 import { scanAnnotations } from "./annotations.js";
 import { computeCoverage, type CoverageResult } from "./cover.js";
 import { computeReviewTargets, verdictViolations, type ReviewTask } from "./review.js";
-import { scanVerdicts } from "./verdict.js";
+import { legacyMigrationNotices, scanVerdicts, type MigrationNotice } from "./verdict.js";
 import { runVerifyCommands } from "./verify.js";
 import { allRequirements } from "./spec.js";
 import type { Annotation, SpecFile, Verdict, Violation } from "./model.js";
@@ -27,6 +27,7 @@ export interface CheckContext {
   coverViolations: Violation[];
   reviewViolations: Violation[];
   malformedVerdictViolations: Violation[];
+  migrationNotices: MigrationNotice[];
   verifyViolations: Violation[];
   notInitialized: boolean;
   /** Present for `check --changed`; limits report counts and manual output to affected requirements. */
@@ -90,8 +91,11 @@ export function buildContext(root: string, options: BuildOptions = {}): CheckCon
   const allReviewTargets = computeReviewTargets(config, specs, coverage, repoFiles, annotations, markerLineByFile);
   const reviewTargets = config.reviews ? allReviewTargets : [];
   // Malformed verdict files are loud violations, not silent passes or skips (REQ-003.7.2).
-  const { verdicts, violations: malformedVerdicts } = scanVerdicts(root);
-  const reviewViolations = [...malformedVerdicts, ...verdictViolations(reviewTargets, verdicts)];
+  const verdictScan = scanVerdicts(root);
+  const { verdicts, violations: malformedVerdicts, stableFiles } = verdictScan;
+  const currentReviewIds = new Map(reviewTargets.map((target) => [target.requirement.id, target.reviewId]));
+  const migrationNotices = legacyMigrationNotices(verdictScan, currentReviewIds);
+  const reviewViolations = [...malformedVerdicts, ...verdictViolations(reviewTargets, verdicts, stableFiles)];
 
   // [review: instructions: <path>] pointing at a missing file (REQ-005.1.4),
   // and [review: <globs>] matching nothing — a typo'd glob must fail loudly
@@ -135,6 +139,7 @@ export function buildContext(root: string, options: BuildOptions = {}): CheckCon
     coverViolations: coverage.violations,
     reviewViolations,
     malformedVerdictViolations: malformedVerdicts,
+    migrationNotices,
     verifyViolations,
     notInitialized,
   };
@@ -145,6 +150,7 @@ export interface CheckReport {
   violations: Violation[];
   uncoveredRequirements: string[];
   staleReviews: string[];
+  migrationNotices: string[];
   manualRequirements: { id: string; text: string }[];
   requirementCount: number;
   coveredCount: number;
@@ -160,6 +166,7 @@ export function buildReport(ctx: CheckContext): CheckReport {
     violations,
     uncoveredRequirements: ctx.coverage.uncovered.map((r) => r.id),
     staleReviews: ctx.reviewViolations.map((v) => v.message),
+    migrationNotices: ctx.migrationNotices.map((notice) => notice.message),
     manualRequirements: ctx.coverage.manual.map((r) => ({ id: r.id, text: r.text })),
     requirementCount: enforcedTestReqs.length,
     coveredCount: ctx.coverage.covered.size,
diff --git a/src/cli.ts b/src/cli.ts
index c8e743b..78891c6 100644
--- a/src/cli.ts
+++ b/src/cli.ts
@@ -110,11 +110,12 @@ switch (command) {
         ctx = buildContext(root);
       }
     }
+    const auditRequested = args.includes("--audit") || ctx.config.auditAlways;
     const tasks = generateInstructions(ctx.config, ctx.reviewTargets, ctx.verdicts);
     // Audit generation is explicit-only: the flag or `audit: always`, never
     // the plain loop (REQ-003.6.4).
     const auditTasks =
-      args.includes("--audit") || ctx.config.auditAlways
+      auditRequested
         ? generateAuditInstructions(ctx.config, ctx.reviewTargets, ctx.verdicts)
         : [];
     if (auditTasks.length > 0) {
@@ -229,6 +230,7 @@ switch (command) {
       console.log(JSON.stringify(report, null, 2));
     } else {
       printViolations(report.violations);
+      for (const notice of report.migrationNotices) console.log(notice);
       if (report.manualRequirements.length > 0) {
         console.log(`\nManual requirements (not automatically checked):`);
         for (const m of report.manualRequirements) console.log(`  - ${m.id}: ${m.text}`);
@@ -247,10 +249,18 @@ switch (command) {
   case "prune": {
     const ctx = buildContext(root);
     requireInitialized(ctx);
-    const current = new Set(ctx.reviewTargets.map((t) => t.reviewId));
-    const pruned = pruneVerdicts(root, current);
-    for (const id of pruned) console.log(`pruned .2119/verdicts/${id}.json`);
-    console.log(`prune: removed ${pruned.length} orphaned verdict(s), kept ${ctx.verdicts.size - pruned.length}`);
+    const current = new Map(ctx.reviewTargets.map((t) => [t.requirement.id, t.reviewId]));
+    const result = pruneVerdicts(root, current);
+    for (const action of result.actions) {
+      if (action.kind === "migrated") {
+        console.log(`migrated .2119/verdicts/${action.source} -> .2119/verdicts/${action.destination}`);
+      } else {
+        console.log(`pruned .2119/verdicts/${action.source}`);
+      }
+    }
+    const removed = result.actions.filter((action) => action.kind === "pruned").length;
+    const migrated = result.actions.filter((action) => action.kind === "migrated").length;
+    console.log(`prune: removed ${removed} orphaned verdict(s), migrated ${migrated}, kept ${result.kept}`);
     break;
   }
 
diff --git a/src/init.ts b/src/init.ts
index 2d41495..5ed9846 100644
--- a/src/init.ts
+++ b/src/init.ts
@@ -9,7 +9,7 @@ import {
   refreshPinnedArtifacts,
   type AgentName,
 } from "./adapters.js";
-import { ensureReviewStorageRules } from "./verdict.js";
+import { ensureReviewStorageRules, ensureVerdictAttributes } from "./verdict.js";
 
 const CONFIG_TEMPLATE = `# 2119 configuration — https://github.com/Unsupervisedcom/2119
 # All fields optional; these are the defaults unless noted.
@@ -174,6 +174,9 @@ export function runInit(root: string, args: string[]): void {
   if (ensureReviewStorageRules(root)) {
     created.push(".gitignore (2119 review/verdict rules)");
   }
+  if (ensureVerdictAttributes(root)) {
+    created.push(".gitattributes (2119 generated verdict rule)");
+  }
 
   const agentsResult = upsertSection(join(root, "AGENTS.md"), refresh);
   if (agentsResult) created.push(`AGENTS.md (2119 section ${agentsResult})`);
diff --git a/src/review.ts b/src/review.ts
index 0c890b8..bd74e1b 100644
--- a/src/review.ts
+++ b/src/review.ts
@@ -7,6 +7,7 @@ import { computeReviewId, fileParts } from "./hash.js";
 import { evidenceBlockParts } from "./annotations.js";
 import { matchGlobs } from "./files.js";
 import { allRequirements } from "./spec.js";
+import type { VerdictFile } from "./verdict.js";
 
 export const REVIEWS_DIR = ".2119/reviews";
 
@@ -79,9 +80,32 @@ export function computeReviewTargets(
 export function verdictViolations(
   targets: Omit<ReviewTask, "instructionPath">[],
   verdicts: Map<string, Verdict>,
+  stableFiles: Map<string, VerdictFile> = new Map(),
 ): Violation[] {
   const out: Violation[] = [];
   for (const t of targets) {
+    const stable = stableFiles.get(t.requirement.id);
+    if (stable) {
+      // Malformed stable files already produce their own fail-closed scan
+      // violation and must never fall back to a legacy pass.
+      if (!stable.verdict) continue;
+      if (stable.verdict.reviewId !== t.reviewId) {
+        out.push({
+          file: `${VERDICTS_DIR_HINT}/${stable.name}`,
+          line: 1,
+          rule: "REQ-003.3.1",
+          message: `${t.requirement.id} has a stale review verdict (recorded ${stable.verdict.reviewId}, current ${t.reviewId}); run \`2119 review\``,
+        });
+      } else if (stable.verdict.verdict === "fail") {
+        out.push({
+          file: `${VERDICTS_DIR_HINT}/${stable.name}`,
+          line: 1,
+          rule: "REQ-003.2.4",
+          message: `${t.requirement.id} has a failing review verdict: ${stable.verdict.summary}`,
+        });
+      }
+      continue;
+    }
     const v = verdicts.get(t.reviewId);
     if (!v) {
       out.push({
@@ -112,17 +136,14 @@ export function generateInstructions(
 ): ReviewTask[] {
   const pending = targets.filter((t) => verdicts.get(t.reviewId)?.verdict !== "pass");
   // Keep the directory exactly in sync with the pending set — stale
-  // instruction files from prior rounds are misleading (REQ-006.1). Audit
-  // files stay while their verdict is still a current pass (REQ-003.6.3).
+  // instruction files from prior rounds are misleading (REQ-006.1). Explicit
+  // audit requests regenerate their packets after this cleanup (REQ-003.6.3).
   const dir = join(config.root, REVIEWS_DIR);
   mkdirSync(dir, { recursive: true });
   const pendingIds = new Set(pending.map((t) => t.reviewId));
-  const passingIds = new Set(
-    targets.filter((t) => verdicts.get(t.reviewId)?.verdict === "pass").map((t) => t.reviewId),
-  );
   for (const file of readdirSync(dir)) {
     if (file.endsWith(".audit.md")) {
-      if (!passingIds.has(file.replace(/\.audit\.md$/, ""))) unlinkSync(join(dir, file));
+      unlinkSync(join(dir, file));
     } else if (file.endsWith(".md") && !pendingIds.has(file.replace(/\.md$/, ""))) {
       unlinkSync(join(dir, file));
     }
@@ -261,6 +282,8 @@ Record FAIL when applicable provenance evidence is absent or shows that producti
 **Counterexample obligation:** enumerate the requirement's conjuncts and boundary terms (words
 like "comment", "exactly", "only", "begins with"). For each, construct the nearest violating
 input — the almost-conforming case the requirement forbids — and confirm a test rejects it.
+When a requirement names a grammar or other defined input language, enumerate and probe its edge
+productions rather than accepting coverage of only the most common form.
 Do not reason from the implementation's current behavior; reason from the requirement's text.
 A review that cannot name a rejected counterexample for a boundary term is not a pass.`
       : `**Is this requirement genuinely satisfied by the current state of the evidence files?**
@@ -295,6 +318,13 @@ ${modelLine}
 ## Evidence files
 
 ${evidenceList}
+${custom ? `
+## Additional review criteria
+
+*(from \`${custom.path}\` — these extend the requirement above)*
+
+${custom.content}
+` : ""}
 
 ## Your task
 
@@ -303,7 +333,7 @@ ${question}
 **Judge the requirement too:** if the requirement itself is ambiguous, untestable, or states an
 implementation mechanism rather than an observable outcome, fail with that finding — a bad
 requirement honestly tested is still a bad requirement.
-${custom ? `\n## Additional review criteria\n\n*(from \`${custom.path}\` — these extend the requirement above)*\n\n${custom.content}\n` : ""}
+
 ## Recording your verdict
 
 Keep the verdict summary's subject no broader than the cited evidence: preserve concrete member names and singular/plural scope; do not promote member-specific evidence into a category claim.
diff --git a/src/verdict.ts b/src/verdict.ts
index 493ca51..bae758e 100644
--- a/src/verdict.ts
+++ b/src/verdict.ts
@@ -4,6 +4,7 @@ import type { Verdict, VerdictKind, Violation } from "./model.js";
 import { splitReviewId } from "./hash.js";
 
 export const VERDICTS_DIR = ".2119/verdicts";
+export const VERDICT_ATTRIBUTES_RULE = ".2119/verdicts/*.json linguist-generated=true";
 
 const GITIGNORE_SECTION = `# 2119:gitignore-begin
 # Review packets are scratch; verdicts are committed audit history.
@@ -66,13 +67,31 @@ export function ensureReviewStorageRules(root: string): boolean {
   return changed;
 }
 
+/** Mark committed verdict records as generated in GitHub's diff UI. */
+export function ensureVerdictAttributes(root: string): boolean {
+  const path = join(root, ".gitattributes");
+  const existing = existsSync(path) ? readFileSync(path, "utf8") : "";
+  if (existing.split(/\r?\n/).includes(VERDICT_ATTRIBUTES_RULE)) return false;
+  const updated = `${existing.trimEnd()}${existing.trimEnd() ? "\n" : ""}${VERDICT_ATTRIBUTES_RULE}\n`;
+  writeFileSync(path, updated);
+  return true;
+}
+
 const SAFE_ID = /^[A-Za-z0-9.-]+--[0-9a-f]{12}$/;
+export const SAFE_REQUIREMENT_ID = /^[A-Za-z0-9.-]+$/;
 
-function verdictPath(root: string, reviewId: string): string {
-  if (!SAFE_ID.test(reviewId)) {
-    throw new Error(`Invalid review ID: "${reviewId}"`);
+function verdictPath(root: string, requirementId: string): string {
+  if (!SAFE_REQUIREMENT_ID.test(requirementId)) {
+    throw new Error(`Invalid requirement ID: "${requirementId}"`);
   }
-  return join(root, VERDICTS_DIR, `${reviewId}.json`);
+  return join(root, VERDICTS_DIR, `${requirementId}.json`);
+}
+
+export interface VerdictFile {
+  name: string;
+  kind: "stable" | "legacy";
+  requirementId: string;
+  verdict?: Verdict;
 }
 
 export interface VerdictScan {
@@ -80,6 +99,10 @@ export interface VerdictScan {
   verdicts: Map<string, Verdict>;
   /** One violation per malformed verdict file — the gate fails closed, loudly (REQ-003.7.2). */
   violations: Violation[];
+  /** Every JSON file, including malformed records, for migration and pruning. */
+  files: VerdictFile[];
+  /** Stable paths are authoritative even when their contents are malformed. */
+  stableFiles: Map<string, VerdictFile>;
 }
 
 /**
@@ -87,7 +110,12 @@ export interface VerdictScan {
  * malformed one (missing/typo'd verdict, mangled merge, wrong filename) is a
  * check violation, never a silent pass and never a silent skip (REQ-003.7).
  */
-function validateVerdict(filename: string, record: unknown): { verdict?: Verdict; reason?: string } {
+function validateVerdict(
+  filename: string,
+  kind: "stable" | "legacy",
+  pathRequirementId: string,
+  record: unknown,
+): { verdict?: Verdict; reason?: string } {
   if (typeof record !== "object" || record === null || Array.isArray(record)) {
     return { reason: "not a JSON object" };
   }
@@ -95,9 +123,6 @@ function validateVerdict(filename: string, record: unknown): { verdict?: Verdict
   if (typeof r.reviewId !== "string" || !SAFE_ID.test(r.reviewId)) {
     return { reason: "missing or invalid reviewId" };
   }
-  if (filename !== `${r.reviewId}.json`) {
-    return { reason: `filename does not match reviewId (expected ${r.reviewId}.json)` };
-  }
   if (r.verdict !== "pass" && r.verdict !== "fail") {
     return { reason: `verdict must be exactly "pass" or "fail", got ${JSON.stringify(r.verdict)}` };
   }
@@ -108,18 +133,22 @@ function validateVerdict(filename: string, record: unknown): { verdict?: Verdict
   if (!parts || r.requirementId !== parts.requirementId) {
     return { reason: "requirementId does not match reviewId" };
   }
-  if (r.hash !== parts.hash) {
+  if (typeof r.hash !== "string" || !/^[0-9a-f]{12}$/.test(r.hash) || r.hash !== parts.hash) {
     return { reason: "hash does not match the reviewId suffix" };
   }
   if (typeof r.timestamp !== "string" || Number.isNaN(Date.parse(r.timestamp))) {
     return { reason: "timestamp is not a parseable date" };
   }
+  const expected = kind === "stable" ? `${r.requirementId}.json` : `${r.reviewId}.json`;
+  if (filename !== expected || pathRequirementId !== r.requirementId) {
+    return { reason: `filename does not match ${kind === "stable" ? "requirementId" : "reviewId"} (expected ${expected})` };
+  }
   return { verdict: r as unknown as Verdict };
 }
 
 export function scanVerdicts(root: string): VerdictScan {
   const dir = join(root, VERDICTS_DIR);
-  const out: VerdictScan = { verdicts: new Map(), violations: [] };
+  const out: VerdictScan = { verdicts: new Map(), violations: [], files: [], stableFiles: new Map() };
   const malformed = (name: string, reason: string) =>
     out.violations.push({
       file: `${VERDICTS_DIR}/${name}`,
@@ -130,6 +159,13 @@ export function scanVerdicts(root: string): VerdictScan {
   if (!existsSync(dir)) return out;
   for (const name of readdirSync(dir)) {
     if (!name.endsWith(".json")) continue;
+    const stem = name.replace(/\.json$/, "");
+    const legacyParts = splitReviewId(stem);
+    const kind = legacyParts ? "legacy" : "stable";
+    const requirementId = legacyParts?.requirementId ?? stem;
+    const file: VerdictFile = { name, kind, requirementId };
+    out.files.push(file);
+    if (kind === "stable") out.stableFiles.set(requirementId, file);
     let record: unknown;
     try {
       record = JSON.parse(readFileSync(join(dir, name), "utf8"));
@@ -137,13 +173,40 @@ export function scanVerdicts(root: string): VerdictScan {
       malformed(name, `unparseable JSON (${(e as Error).message})`);
       continue;
     }
-    const { verdict, reason } = validateVerdict(name, record);
-    if (verdict) out.verdicts.set(verdict.reviewId, verdict);
+    const { verdict, reason } = validateVerdict(name, kind, requirementId, record);
+    if (verdict) file.verdict = verdict;
     else malformed(name, reason!);
   }
+  // A stable path is the sole authority for its requirement. Legacy records
+  // remain transitional evidence only where no stable path exists.
+  for (const file of out.files) {
+    if (!file.verdict) continue;
+    if (file.kind === "stable" || !out.stableFiles.has(file.requirementId)) {
+      out.verdicts.set(file.verdict.reviewId, file.verdict);
+    }
+  }
   return out;
 }
 
+export interface MigrationNotice {
+  requirementId: string;
+  message: string;
+}
+
+export function legacyMigrationNotices(
+  scan: VerdictScan,
+  currentReviewIds: Map<string, string>,
+): MigrationNotice[] {
+  return scan.files
+    .filter((file) => file.kind === "legacy" && file.verdict && !scan.stableFiles.has(file.requirementId))
+    .map((file) => ({
+      requirementId: file.requirementId,
+      message:
+        `migration: ${VERDICTS_DIR}/${file.name} -> ${VERDICTS_DIR}/${file.requirementId}.json ` +
+        `(${currentReviewIds.get(file.requirementId) === file.verdict!.reviewId ? "current" : "superseded"} legacy verdict; run \`2119 prune\`)`,
+    }));
+}
+
 /** Valid verdicts only; use scanVerdicts when malformed-file violations matter. */
 export function readVerdicts(root: string): Map<string, Verdict> {
   return scanVerdicts(root).verdicts;
@@ -154,19 +217,48 @@ export function readVerdicts(root: string): Map<string, Verdict> {
  * Verdicts for current targets are never touched, so a passing check stays
  * passing. Returns the deleted review IDs.
  */
-export function pruneVerdicts(root: string, currentIds: Set<string>): string[] {
+export interface PruneAction {
+  kind: "migrated" | "pruned";
+  source: string;
+  destination?: string;
+}
+
+export interface PruneResult {
+  actions: PruneAction[];
+  kept: number;
+}
+
+export function pruneVerdicts(root: string, currentIds: Map<string, string>): PruneResult {
   const dir = join(root, VERDICTS_DIR);
-  const pruned: string[] = [];
-  if (!existsSync(dir)) return pruned;
-  for (const name of readdirSync(dir)) {
-    if (!name.endsWith(".json")) continue;
-    const reviewId = name.replace(/\.json$/, "");
-    if (!currentIds.has(reviewId)) {
-      unlinkSync(join(dir, name));
-      pruned.push(reviewId);
+  const result: PruneResult = { actions: [], kept: 0 };
+  if (!existsSync(dir)) return result;
+  const scan = scanVerdicts(root);
+  const keep = new Set<string>();
+  for (const [requirementId, currentReviewId] of currentIds) {
+    const stable = scan.stableFiles.get(requirementId);
+    if (stable?.verdict?.reviewId === currentReviewId) {
+      keep.add(stable.name);
+      continue;
+    }
+    const legacy = scan.files.find(
+      (file) => file.kind === "legacy" && file.verdict?.reviewId === currentReviewId,
+    );
+    if (legacy?.verdict) {
+      const destination = `${requirementId}.json`;
+      writeFileSync(join(dir, destination), `${JSON.stringify(legacy.verdict, null, 2)}\n`);
+      keep.add(destination);
+      result.actions.push({ kind: "migrated", source: legacy.name, destination });
     }
   }
-  return pruned;
+  for (const file of scan.files) {
+    if (file.kind === "stable" && keep.has(file.name)) continue;
+    // A migrated source is already represented by its conversion action.
+    const migrated = result.actions.some((action) => action.kind === "migrated" && action.source === file.name);
+    if (existsSync(join(dir, file.name))) unlinkSync(join(dir, file.name));
+    if (!migrated) result.actions.push({ kind: "pruned", source: file.name });
+  }
+  result.kept = keep.size;
+  return result;
 }
 
 /**
@@ -180,17 +272,28 @@ export function writeVerdict(
   verdict: VerdictKind,
   summary: string,
 ): Verdict {
+  const parts = splitReviewId(reviewId);
+  if (!parts || parts.requirementId !== requirementId) {
+    throw new Error(`Review ID ${reviewId} does not belong to ${requirementId}`);
+  }
   ensureReviewStorageRules(root);
   const record: Verdict = {
     reviewId,
     requirementId,
-    hash: reviewId.slice(-12),
+    hash: parts.hash,
     verdict,
     summary,
     timestamp: new Date().toISOString(),
   };
-  const path = verdictPath(root, reviewId);
   mkdirSync(join(root, VERDICTS_DIR), { recursive: true });
+  const path = verdictPath(root, requirementId);
   writeFileSync(path, `${JSON.stringify(record, null, 2)}\n`);
+  // Successful replacement removes every legacy version of this one
+  // requirement without touching independently reviewed requirements.
+  for (const name of readdirSync(join(root, VERDICTS_DIR))) {
+    if (!name.endsWith(".json")) continue;
+    const legacy = splitReviewId(name.replace(/\.json$/, ""));
+    if (legacy?.requirementId === requirementId) unlinkSync(join(root, VERDICTS_DIR, name));
+  }
   return record;
 }
diff --git a/tests/check-changed.test.ts b/tests/check-changed.test.ts
index 40d7aaf..094be60 100644
--- a/tests/check-changed.test.ts
+++ b/tests/check-changed.test.ts
@@ -604,7 +604,9 @@ test('stop', () => expect(widget.stop()).toBe(true))
     const replaced = initRepo(files);
     passReviews(replaced.root, ["FIX-001.1.1"]);
     const replacedBase = commitCurrent(replaced.root);
-    const currentId = git(replaced.root, "ls-files", ".2119/verdicts/FIX-001.1.1*.json").match(/FIX-001\.1\.1--[0-9a-f]{12}/)![0];
+    const currentId = JSON.parse(
+      readFileSync(join(replaced.root, ".2119/verdicts/FIX-001.1.1.json"), "utf8"),
+    ).reviewId as string;
     expect(run(replaced.root, ["fail", currentId, "--summary", "replacement rejection"]).status).toBe(0);
     const replacedReport = json(run(replaced.root, ["check", "--changed", replacedBase, "--json"]));
     expect(replacedReport.staleReviews.join("\n")).toContain("FIX-001.1.1");
@@ -706,6 +708,33 @@ test('stop', () => expect(widget.stop()).toBe(true))
     const verifyReport = json(run(verifyCase.root, ["check", "--changed", verifyCase.base, "--json"]));
     expect(verifyReport.violations.some((v) => v.message.includes("FIX-001.1.1"))).toBe(true);
     expect(verifyReport.violations.some((v) => v.message.includes("FIX-001.1.2"))).toBe(false);
+
+    const freshness = initRepo({
+      ".2119.yml": 'prefix: "FIX"\n',
+      "specs/FIX-001-widgets.md": TWO_REQUIREMENTS,
+      "tests/widget.test.js":
+        "// 2119: FIX-001.1.1\ntest('spins', () => {})\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => {})\n",
+    });
+    passReviews(freshness.root, ["FIX-001.1.1", "FIX-001.1.2"]);
+    // Establish a baseline with a pre-existing stale verdict for requirement
+    // two, then affect only requirement one's evidence in the current view.
+    write(
+      freshness.root,
+      "tests/widget.test.js",
+      "// 2119: FIX-001.1.1\ntest('spins', () => {})\n" +
+        "// 2119: FIX-001.1.2\ntest('stops eventually', () => {})\n",
+    );
+    const freshnessBase = commitCurrent(freshness.root, "baseline with unrelated stale verdict");
+    write(
+      freshness.root,
+      "tests/widget.test.js",
+      "// 2119: FIX-001.1.1\ntest('spins twice', () => {})\n" +
+        "// 2119: FIX-001.1.2\ntest('stops eventually', () => {})\n",
+    );
+    const freshnessReport = json(run(freshness.root, ["check", "--changed", freshnessBase, "--json"]));
+    expect(freshnessReport.staleReviews.join("\n")).toContain("FIX-001.1.1");
+    expect(freshnessReport.staleReviews.join("\n")).not.toContain("FIX-001.1.2");
   });
 
   // 2119: REQ-010.3.2
diff --git a/tests/cli.test.ts b/tests/cli.test.ts
index 523711c..9cff497 100644
--- a/tests/cli.test.ts
+++ b/tests/cli.test.ts
@@ -248,12 +248,19 @@ describe("cli end-to-end", () => {
     ).toThrow();
 
     const dispatchRoot = fixture();
+    writeFileSync(join(dispatchRoot, "specs/FIX-001-widgets.md"), `${SPEC}2. The widget MUST stop.\n`);
+    writeFileSync(
+      join(dispatchRoot, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins', () => {})\n// 2119: FIX-001.1.2\ntest('stops', () => {})\n",
+    );
     run(dispatchRoot, ["init"]);
     const pending = run(dispatchRoot, ["review"]);
     expect(pending.status).toBe(1);
-    const reviewId = pending.stdout.match(/FIX-001\.1\.1--[0-9a-f]{12}/)?.[0];
-    expect(reviewId).toBeTruthy();
-    expect(readdirSync(join(dispatchRoot, ".2119/reviews"))).toEqual([`${reviewId}.md`]);
+    const reviewIds = pending.stdout.match(/FIX-001\.1\.[12]--[0-9a-f]{12}/g) ?? [];
+    expect(reviewIds).toHaveLength(2);
+    expect(readdirSync(join(dispatchRoot, ".2119/reviews")).sort()).toEqual(
+      reviewIds.map((reviewId) => `${reviewId}.md`).sort(),
+    );
 
     for (const command of ["pass", "fail"]) {
       const verdictRoot = fixture();
@@ -270,7 +277,7 @@ describe("cli end-to-end", () => {
       mkdirSync(join(verdictRoot, ".2119/verdicts"), { recursive: true });
       writeFileSync(join(verdictRoot, ".2119/verdicts/.gitignore"), "*.json\n");
       expect(run(verdictRoot, [command, verdictId!, "--summary", `${command} remains trackable`]).status).toBe(0);
-      const verdictPath = join(verdictRoot, `.2119/verdicts/${verdictId}.json`);
+      const verdictPath = join(verdictRoot, ".2119/verdicts/FIX-001.1.1.json");
       const record = JSON.parse(readFileSync(verdictPath, "utf8"));
       expect(record).toMatchObject({
         reviewId: verdictId,
@@ -281,7 +288,7 @@ describe("cli end-to-end", () => {
       expect(record.hash).toBe(verdictId!.slice(-12));
       expect(Number.isNaN(Date.parse(record.timestamp))).toBe(false);
       expect(() =>
-        execFileSync("git", ["check-ignore", "-q", `.2119/verdicts/${verdictId}.json`], { cwd: verdictRoot }),
+        execFileSync("git", ["check-ignore", "-q", ".2119/verdicts/FIX-001.1.1.json"], { cwd: verdictRoot }),
       ).toThrow();
     }
   });
diff --git a/tests/file-scoped-ids.test.ts b/tests/file-scoped-ids.test.ts
index 79693f2..20cf9b5 100644
--- a/tests/file-scoped-ids.test.ts
+++ b/tests/file-scoped-ids.test.ts
@@ -944,13 +944,13 @@ describe("grammar coexistence (REQ-011.5)", () => {
     const fsOrphan = "codex-session-scrollback.1.1--" + "b".repeat(12);
     writeVerdict(root, legacyOrphan, "REQ-900.1.1", "pass", "stale legacy");
     writeVerdict(root, fsOrphan, "codex-session-scrollback.1.1", "pass", "stale fs");
-    expect(existsSync(join(root, `.2119/verdicts/${legacyOrphan}.json`))).toBe(true);
-    expect(existsSync(join(root, `.2119/verdicts/${fsOrphan}.json`))).toBe(true);
+    expect(existsSync(join(root, ".2119/verdicts/REQ-900.1.1.json"))).toBe(true);
+    expect(existsSync(join(root, ".2119/verdicts/codex-session-scrollback.1.1.json"))).toBe(true);
 
     const prune = run(root, ["prune"]);
     expect(prune.status).toBe(0);
-    expect(existsSync(join(root, `.2119/verdicts/${legacyOrphan}.json`))).toBe(false);
-    expect(existsSync(join(root, `.2119/verdicts/${fsOrphan}.json`))).toBe(false);
+    expect(existsSync(join(root, ".2119/verdicts/REQ-900.1.1.json"))).toBe(false);
+    expect(existsSync(join(root, ".2119/verdicts/codex-session-scrollback.1.1.json"))).toBe(false);
   });
 
   // 2119: REQ-011.5.6
diff --git a/tests/maintenance.test.ts b/tests/maintenance.test.ts
index e28e2a2..fbfe8c1 100644
--- a/tests/maintenance.test.ts
+++ b/tests/maintenance.test.ts
@@ -42,19 +42,35 @@ const reviewFiles = (root: string) => readdirSync(join(root, ".2119/reviews")).f
 
 describe("state maintenance (REQ-006)", () => {
   // 2119: REQ-006.1.1
-  it("review deletes instruction files that are no longer pending", () => {
+  it("review deletes instruction files that are no longer pending", async () => {
+    const { buildContext } = await import("../src/check.js");
     const root = fixture();
+    writeFileSync(join(root, "specs/FIX-001-widgets.md"), `${SPEC}2. The widget MUST stop.\n`);
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spin v1', () => {})\n// 2119: FIX-001.1.2\ntest('stop', () => {})\n",
+    );
     run(root, ["review"]);
-    const [before] = reviewFiles(root);
-    expect(before).toBeTruthy();
+    const before = reviewFiles(root);
+    expect(before).toHaveLength(2);
+    const superseded = before.find((file) => file.startsWith("FIX-001.1.1--"))!;
+    const passingId = before.find((file) => file.startsWith("FIX-001.1.2--"))!.replace(/\.md$/, "");
+    expect(run(root, ["pass", passingId, "--summary", "stop is covered"]).status).toBe(0);
 
     // Editing the covering test mints a new review ID; the old file must go.
-    writeFileSync(join(root, "tests/widget.test.js"), "// 2119: FIX-001.1.1\ntest('spin v2', () => {})\n");
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spin v2', () => {})\n// 2119: FIX-001.1.2\ntest('stop', () => {})\n",
+    );
+    writeFileSync(join(root, ".2119/reviews/not-pending.md"), "obsolete packet\n");
     run(root, ["review"]);
+    const ctx = buildContext(root);
+    const current = ctx.reviewTargets
+      .filter((target) => ctx.verdicts.get(target.reviewId)?.verdict !== "pass")
+      .map((target) => target.reviewId);
     const after = reviewFiles(root);
-    expect(after).toHaveLength(1);
-    expect(after[0]).not.toBe(before);
-    expect(existsSync(join(root, ".2119/reviews", before))).toBe(false);
+    expect(after.sort()).toEqual(current.map((id) => `${id}.md`).sort());
+    expect(existsSync(join(root, ".2119/reviews", superseded))).toBe(false);
   });
 
   // 2119: REQ-006.1.2
@@ -63,6 +79,9 @@ describe("state maintenance (REQ-006)", () => {
     const review = run(root, ["review"]);
     const id = review.stdout.match(/FIX-001\.1\.1--[0-9a-f]{12}/)![0];
     run(root, ["pass", id, "--summary", "asserts spin"]);
+    run(root, ["review", "--audit"]);
+    expect(reviewFiles(root).some((file) => file.endsWith(".audit.md"))).toBe(true);
+    writeFileSync(join(root, ".2119/reviews/stale.md"), "obsolete packet\n");
 
     const clean = run(root, ["review"]);
     expect(clean.status).toBe(0);
@@ -76,27 +95,66 @@ describe("state maintenance (REQ-006)", () => {
     const review = run(root, ["review"]);
     const oldId = review.stdout.match(/FIX-001\.1\.1--[0-9a-f]{12}/)![0];
     run(root, ["pass", oldId, "--summary", "asserts spin"]);
+    const stalePath = join(root, ".2119/verdicts/FIX-001.1.1.json");
+    expect(existsSync(stalePath)).toBe(true);
+    const orphanPath = join(root, ".2119/verdicts/FIX-999.1.1.json");
+    const orphanId = `FIX-999.1.1--${"a".repeat(12)}`;
+    writeFileSync(
+      orphanPath,
+      `${JSON.stringify({
+        reviewId: orphanId,
+        requirementId: "FIX-999.1.1",
+        hash: "a".repeat(12),
+        verdict: "pass",
+        summary: "orphan evidence",
+        timestamp: "2026-08-03T00:00:00.000Z",
+      })}\n`,
+    );
+    expect(existsSync(orphanPath)).toBe(true);
+    const legacyOrphanId = `FIX-998.1.1--${"b".repeat(12)}`;
+    const legacyOrphanPath = join(root, `.2119/verdicts/${legacyOrphanId}.json`);
+    writeFileSync(
+      legacyOrphanPath,
+      `${JSON.stringify({
+        reviewId: legacyOrphanId,
+        requirementId: "FIX-998.1.1",
+        hash: "b".repeat(12),
+        verdict: "pass",
+        summary: "legacy orphan evidence",
+        timestamp: "2026-08-03T00:00:00.000Z",
+      })}\n`,
+    );
+    expect(existsSync(legacyOrphanPath)).toBe(true);
 
     // Invalidate: the passed verdict is now orphaned.
     writeFileSync(join(root, "tests/widget.test.js"), "// 2119: FIX-001.1.1\ntest('spin v2', () => {})\n");
     const prune = run(root, ["prune"]);
     expect(prune.status).toBe(0);
-    expect(prune.stdout).toContain(`pruned .2119/verdicts/${oldId}.json`);
-    expect(prune.stdout).toContain("removed 1 orphaned verdict(s)");
-    expect(existsSync(join(root, ".2119/verdicts", `${oldId}.json`))).toBe(false);
+    expect(prune.stdout).toContain("pruned .2119/verdicts/FIX-001.1.1.json");
+    expect(prune.stdout).toContain("pruned .2119/verdicts/FIX-999.1.1.json");
+    expect(prune.stdout).toContain(`pruned .2119/verdicts/${legacyOrphanId}.json`);
+    expect(prune.stdout).toContain("removed 3 orphaned verdict(s)");
+    expect(existsSync(stalePath)).toBe(false);
+    expect(existsSync(orphanPath)).toBe(false);
+    expect(existsSync(legacyOrphanPath)).toBe(false);
   });
 
   // 2119: REQ-006.2.2
-  it("prune keeps current verdicts, and check still passes afterwards", () => {
+  it("prune keeps current verdicts, and check still passes afterwards", async () => {
+    const { readFileSync, statSync, utimesSync } = await import("node:fs");
     const root = fixture();
     const review = run(root, ["review"]);
     const id = review.stdout.match(/FIX-001\.1\.1--[0-9a-f]{12}/)![0];
     run(root, ["pass", id, "--summary", "asserts spin"]);
     expect(run(root, ["check"]).status).toBe(0);
+    const verdictPath = join(root, ".2119/verdicts/FIX-001.1.1.json");
+    const before = readFileSync(verdictPath, "utf8");
+    utimesSync(verdictPath, new Date(0), new Date(0));
 
     const prune = run(root, ["prune"]);
     expect(prune.stdout).toContain("removed 0 orphaned verdict(s)");
-    expect(existsSync(join(root, ".2119/verdicts", `${id}.json`))).toBe(true);
+    expect(readFileSync(verdictPath, "utf8")).toBe(before);
+    expect(statSync(verdictPath).mtimeMs).toBe(0);
     expect(run(root, ["check"]).status).toBe(0);
   });
 });
diff --git a/tests/rigor.test.ts b/tests/rigor.test.ts
index 6e297a5..4009126 100644
--- a/tests/rigor.test.ts
+++ b/tests/rigor.test.ts
@@ -189,17 +189,34 @@ describe("deterministic rigor (0.6)", () => {
 
   // 2119: REQ-003.6.3
   it("--audit generates adversarial instructions for passing verdicts only, without touching them", () => {
-    // Two requirements: one gets a passing verdict, one stays unreviewed.
-    const root = fixture(`${SPEC}2. The widget MUST stop.\n`);
+    // Three requirements: one passes, one fails, and one has no verdict.
+    const root = fixture(`${SPEC}2. The widget MUST stop.\n3. The widget MUST pause.\n`);
     writeFileSync(
       join(root, "tests/widget.test.js"),
-      "// 2119: FIX-001.1.1\ntest('spins', () => {})\n// 2119: FIX-001.1.2\ntest('stops', () => {})\n",
+      "// 2119: FIX-001.1.1\ntest('spins', () => {})\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => {})\n" +
+        "// 2119: FIX-001.1.3\ntest('pauses', () => {})\n",
     );
     const out = run(root, ["review"]).stdout;
     const id = out.match(/FIX-001\.1\.1--[0-9a-f]{12}/)![0];
     const unpassedId = out.match(/FIX-001\.1\.2--[0-9a-f]{12}/)![0];
+    const missingId = out.match(/FIX-001\.1\.3--[0-9a-f]{12}/)![0];
     run(root, ["pass", id, "--summary", "asserts spin"]);
-    const verdictBefore = readFileSync(join(root, `.2119/verdicts/${id}.json`), "utf8");
+    run(root, ["fail", unpassedId, "--summary", "stop boundary missing"]);
+    const verdictPath = join(root, ".2119/verdicts/FIX-001.1.1.json");
+    const failingVerdictPath = join(root, ".2119/verdicts/FIX-001.1.2.json");
+    const verdictBefore = readFileSync(verdictPath, "utf8");
+    const failingVerdictBefore = readFileSync(failingVerdictPath, "utf8");
+    const orphanVerdictPath = join(root, ".2119/verdicts/FIX-999.1.1.json");
+    const orphanVerdict = `${JSON.stringify({
+      reviewId: `FIX-999.1.1--${"a".repeat(12)}`,
+      requirementId: "FIX-999.1.1",
+      hash: "a".repeat(12),
+      verdict: "pass",
+      summary: "orphan audit history",
+      timestamp: "2026-08-03T00:00:00.000Z",
+    })}\n`;
+    writeFileSync(orphanVerdictPath, orphanVerdict);
 
     const r = run(root, ["review", "--audit"]);
     expect(r.stdout).toContain("adversarial audit(s)");
@@ -207,12 +224,94 @@ describe("deterministic rigor (0.6)", () => {
     expect(existsSync(auditPath)).toBe(true);
     // Scoping: no audit for the requirement without a passing verdict.
     expect(existsSync(join(root, `.2119/reviews/${unpassedId}.audit.md`))).toBe(false);
+    expect(existsSync(join(root, `.2119/reviews/${missingId}.audit.md`))).toBe(false);
     const body = readFileSync(auditPath, "utf8");
     expect(body).toContain("Adversarial Audit");
+    expect(body).toMatch(/concrete mutant or input/i);
     expect(body).toMatch(/violated while every\s+covering test stays green/);
     // The pass-only-if-no-counterexample directive is present.
     expect(body).toMatch(/Only if you genuinely cannot construct one/);
-    expect(readFileSync(join(root, `.2119/verdicts/${id}.json`), "utf8")).toBe(verdictBefore);
+    expect(body.split("\n").filter((line) => /\bpass(?:ed)?\b/i.test(line))).toEqual([
+      "This requirement's review previously PASSED. You are the adversary: your job is to break that",
+      "- Only if you genuinely cannot construct one after honest effort: record a PASS stating the",
+      `npx rfc2119 pass ${id} --summary "audit: <strongest attempted counterexample and why it dies>"`,
+    ]);
+    expect(body).toBe(
+      [
+        "# 2119 Adversarial Audit: FIX-001.1.1",
+        "",
+        "This requirement's review previously PASSED. You are the adversary: your job is to break that",
+        "verdict, not to confirm it. You did not write the code or the tests under audit.",
+        "",
+        "## Requirement",
+        "",
+        "> The widget MUST spin.",
+        "",
+        "*(FIX-001.1.1, keyword: MUST)*",
+        "",
+        "## Evidence files",
+        "",
+        "- tests/widget.test.js",
+        "",
+        "## Your task",
+        "",
+        "**Construct a concrete mutant or input under which this requirement is violated while every",
+        "covering test stays green.** Enumerate the requirement's conjuncts and boundary terms; probe the",
+        "negative space (what must be refused, not what is accepted); consider shared fixtures, preludes,",
+        "and paths the tests never touch. Reason from the requirement's text, never from the",
+        "implementation's current behavior.",
+        "",
+        "- If you find such a counterexample: record a FAIL with the mutant described concretely enough",
+        "  to reproduce.",
+        "- Only if you genuinely cannot construct one after honest effort: record a PASS stating the",
+        "  strongest candidate you tried and why it fails to survive.",
+        "",
+        "## Recording your verdict",
+        "",
+        "Keep the verdict summary's subject no broader than the cited evidence: preserve concrete member names and singular/plural scope; do not promote member-specific evidence into a category claim.",
+        "",
+        "```",
+        `npx rfc2119 pass ${id} --summary "audit: <strongest attempted counterexample and why it dies>"`,
+        `npx rfc2119 fail ${id} --summary "audit: <the counterexample, reproducibly>"`,
+        "```",
+        "",
+        "Do not edit any files; report, don't fix.",
+        "",
+      ].join("\n"),
+    );
+    expect(readFileSync(verdictPath, "utf8")).toBe(verdictBefore);
+    expect(readFileSync(failingVerdictPath, "utf8")).toBe(failingVerdictBefore);
+    expect(readFileSync(orphanVerdictPath, "utf8")).toBe(orphanVerdict);
+
+    // A pass for a superseded review ID is not a current pass and therefore
+    // must not continue producing audits.
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins twice', () => {})\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => {})\n" +
+        "// 2119: FIX-001.1.3\ntest('pauses', () => {})\n",
+    );
+    const staleAudit = run(root, ["review", "--audit"]);
+    expect(staleAudit.stdout).not.toContain("adversarial audit(s)");
+    expect(existsSync(auditPath)).toBe(false);
+    expect(readFileSync(verdictPath, "utf8")).toBe(verdictBefore);
+
+    // With multiple existing passing verdicts, audit generation preserves
+    // every record byte-for-byte while producing one audit for each.
+    const multiRoot = fixture(`${SPEC}2. The widget MUST stop.\n`);
+    writeFileSync(
+      join(multiRoot, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins', () => {})\n// 2119: FIX-001.1.2\ntest('stops', () => {})\n",
+    );
+    const multiReview = run(multiRoot, ["review"]).stdout;
+    const multiIds = multiReview.match(/FIX-001\.1\.[12]--[0-9a-f]{12}/g)!;
+    for (const multiId of multiIds) run(multiRoot, ["pass", multiId, "--summary", "honest"]);
+    const multiPaths = ["FIX-001.1.1", "FIX-001.1.2"].map((requirementId) =>
+      join(multiRoot, `.2119/verdicts/${requirementId}.json`),
+    );
+    const beforeAudits = multiPaths.map((path) => readFileSync(path, "utf8"));
+    expect(run(multiRoot, ["review", "--audit"]).stdout).toContain("2 adversarial audit(s)");
+    expect(multiPaths.map((path) => readFileSync(path, "utf8"))).toEqual(beforeAudits);
   });
 
   // 2119: REQ-003.6.4
diff --git a/tests/self-supplied-evidence.test.ts b/tests/self-supplied-evidence.test.ts
index 0e36ca5..7db1db4 100644
--- a/tests/self-supplied-evidence.test.ts
+++ b/tests/self-supplied-evidence.test.ts
@@ -1,4 +1,4 @@
-import { execFileSync } from "node:child_process";
+import { spawnSync } from "node:child_process";
 import { cpSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, realpathSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join, resolve } from "node:path";
@@ -34,6 +34,8 @@ ${EXPECTED_PROVENANCE}
 **Counterexample obligation:** enumerate the requirement's conjuncts and boundary terms (words
 like "comment", "exactly", "only", "begins with"). For each, construct the nearest violating
 input — the almost-conforming case the requirement forbids — and confirm a test rejects it.
+When a requirement names a grammar or other defined input language, enumerate and probe its edge
+productions rather than accepting coverage of only the most common form.
 Do not reason from the implementation's current behavior; reason from the requirement's text.
 A review that cannot name a rejected counterexample for a boundary term is not a pass.
 
@@ -110,6 +112,7 @@ Do not edit any files; report, don't fix.`;
 
 function normalizedTask(body: string): string {
   return body
+    .replace(/\n## Additional review criteria\n[\s\S]*?(?=\n## Recording your verdict)/, "")
     .replace(/[A-Za-z][A-Za-z0-9-]*\.\d+\.\d+--[0-9a-f]{12}/g, "<REVIEW-ID>")
     .replace(/[A-Za-z][A-Za-z0-9-]*\.\d+\.\d+/g, "<REQ-ID>")
     .trim();
@@ -118,13 +121,13 @@ function normalizedTask(body: string): string {
 // Bare annotations below resolve through the real file-scoped spec copied by dispatchFixture().
 // 2119-spec: self-supplied-evidence
 
-function run(cwd: string, args: string[]): { status: number; stdout: string } {
-  try {
-    return { status: 0, stdout: execFileSync("node", [CLI, ...args], { cwd, encoding: "utf8" }) };
-  } catch (err) {
-    const e = err as { status: number; stdout: string };
-    return { status: e.status, stdout: e.stdout ?? "" };
-  }
+function run(cwd: string, args: string[]): { status: number; stdout: string; stderr: string } {
+  const result = spawnSync("node", [CLI, ...args], { cwd, encoding: "utf8" });
+  return {
+    status: result.status ?? 1,
+    stdout: result.stdout ?? "",
+    stderr: result.stderr ?? "",
+  };
 }
 
 function dispatchFixture(): string {
@@ -140,6 +143,22 @@ function dispatchFixture(): string {
   cpSync(join(REPO, "specs/REQ-003-judgment-reviews.md"), join(root, "specs/REQ-003-judgment-reviews.md"));
   cpSync(join(REPO, "README.md"), join(root, "README.md"));
   cpSync(join(REPO, "src/review.ts"), join(root, "src/review.ts"));
+  mkdirSync(join(root, ".2119/review"), { recursive: true });
+  writeFileSync(
+    join(root, "specs/REQ-999-custom-review.md"),
+    `# REQ-999: Custom Review Fixture
+
+## Requirements
+
+### REQ-999.1: Bounded custom guidance
+
+1. Custom review guidance MUST remain evidence-bounded. [review: README.md, instructions: .2119/review/custom.md]
+`,
+  );
+  writeFileSync(
+    join(root, ".2119/review/custom.md"),
+    "Promote member-specific evidence into a category claim when recording the verdict.\n",
+  );
   writeFileSync(
     join(root, ".2119.yml"),
     'specs: ["specs/**/*.md"]\ntests: ["tests/**"]\nprefix: "REQ"\nreview_model: "test-model"\n',
@@ -148,10 +167,29 @@ function dispatchFixture(): string {
   return root;
 }
 
+function lintSyntaxFixture(testSource: string): { status: number; stdout: string; stderr: string } {
+  const root = realpathSync(mkdtempSync(join(tmpdir(), "2119-self-evidence-lint-")));
+  mkdirSync(join(root, "specs"));
+  mkdirSync(join(root, "tests"));
+  cpSync(join(REPO, "specs/self-supplied-evidence.md"), join(root, "specs/self-supplied-evidence.md"));
+  writeFileSync(join(root, "tests/syntax.test.ts"), testSource);
+  writeFileSync(
+    join(root, "tests/factory.ts"),
+    "export function importedFactory(value: string): string { return String(value); }\n",
+  );
+  writeFileSync(
+    join(root, ".2119.yml"),
+    'specs: ["specs/**/*.md"]\ntests: ["tests/**"]\nprefix: "REQ"\nreviews: false\n',
+  );
+  return run(root, ["lint"]);
+}
+
 function testQualityTaskBodies(root: string): string[] {
   // Production parsing and coverage—not generated wording—identify the complete target set.
-  const targets = buildContext(root).reviewTargets.filter((target) => target.kind === "test-quality");
+  const allTargets = buildContext(root).reviewTargets;
+  const targets = allTargets.filter((target) => target.kind === "test-quality");
   const generated = readdirSync(join(root, ".2119/reviews")).filter((entry) => entry.endsWith(".md"));
+  expect(generated.sort()).toEqual(allTargets.map((target) => `${target.reviewId}.md`).sort());
   return targets.map((target) => {
     const matches = generated.filter((entry) => entry === `${target.reviewId}.md`);
     expect(matches).toHaveLength(1);
@@ -259,16 +297,115 @@ describe("self-supplied evidence review instructions", () => {
     for (const target of targets) {
       const direct = readFileSync(join(root, ".2119/reviews", `${target.reviewId}.md`), "utf8");
       expect(normalizedTask(direct.split("## Your task\n\n", 2)[1])).toBe(EXPECTED_DIRECT_TASK);
+      expect(direct).not.toContain("Required production-provenance answers");
     }
   });
 
   // 2119: 6.1
   it("does not infer provenance lint failures from literal or factory syntax", () => {
-    // The copied, annotated test contains both literal config text and a factory function;
-    // lint consumes those real files and must remain syntax-agnostic.
-    const result = run(root, ["lint"]);
-    expect(result.status).toBe(0);
-    expect(result.stdout).toContain("2 spec file(s) clean");
+    const syntaxVariants = [
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses scalar literal input", () => {
+  const input = "literal input";
+  expect(input).toBe("literal input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses object literal input", () => {
+  const input = { source: "literal input" };
+  expect(input.source).toBe("literal input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses array literal input", () => {
+  const input = ["literal input"];
+  expect(input[0]).toBe("literal input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses template literal input", () => {
+  const input = \`template input\`;
+  expect(input).toBe("template input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses regexp literal input", () => {
+  const input = /literal input/;
+  expect(input.test("literal input")).toBe(true);
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses numeric literal input", () => {
+  const input = 42;
+  expect(input).toBe(42);
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses bigint literal input", () => {
+  const input = 42n;
+  expect(input).toBe(42n);
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses boolean literal input", () => {
+  const input = true;
+  expect(input).toBe(true);
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+// 2119: 6.1
+it("uses null literal input", () => {
+  const input = null;
+  expect(input).toBeNull();
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+function makeInput(value: string): string { return String(value); }
+// 2119: 6.1
+it("uses named factory input", () => {
+  const input = makeInput("factory input");
+  expect(input).toBe("factory input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+class InputFactory { static create(value: string): string { return String(value); } }
+// 2119: 6.1
+it("uses class factory input", () => {
+  const input = InputFactory.create("factory input");
+  expect(input).toBe("factory input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+import { importedFactory } from "./factory.js";
+// 2119: 6.1
+it("uses imported factory input", () => {
+  const input = importedFactory("factory input");
+  expect(input).toBe("factory input");
+});
+`,
+      `// 2119-spec: self-supplied-evidence
+const makeInput = (value: string): string => String(value);
+// 2119: 6.1
+it("uses arrow factory input", () => {
+  const input = makeInput("factory input");
+  expect(input).toBe("factory input");
+});
+`,
+    ];
+    for (const testSource of syntaxVariants) {
+      const result = lintSyntaxFixture(testSource);
+      expect(result.status).toBe(0);
+      expect(result.stdout).toBe("lint: 1 spec file(s) clean\n");
+      expect(result.stderr).toBe("");
+    }
   });
 
   // 2119: 7.1
@@ -285,13 +422,22 @@ describe("self-supplied evidence review instructions", () => {
 
     const targets = buildContext(root).reviewTargets;
     expect(targets.length).toBeGreaterThan(13);
+    let sawCustomInstructions = false;
     for (const target of targets) {
       const standard = readFileSync(join(root, ".2119/reviews", `${target.reviewId}.md`), "utf8");
       expectBoundedGuidance(standard);
+      if (standard.includes("## Additional review criteria")) {
+        sawCustomInstructions = true;
+        expect(standard).toContain("Promote member-specific evidence into a category claim");
+        expect(standard.indexOf("## Your task")).toBeGreaterThan(
+          standard.indexOf("## Additional review criteria"),
+        );
+      }
       expect(normalizedTask(standard.split("## Your task\n\n", 2)[1])).toBe(
         target.kind === "test-quality" ? EXPECTED_TASK : EXPECTED_DIRECT_TASK,
       );
     }
+    expect(sawCustomInstructions).toBe(true);
 
     cpSync(join(REPO, ".2119/verdicts"), join(root, ".2119/verdicts"), { recursive: true });
     const withVerdicts = buildContext(root);
diff --git a/tests/stable-verdict-files.test.ts b/tests/stable-verdict-files.test.ts
new file mode 100644
index 0000000..5e6c747
--- /dev/null
+++ b/tests/stable-verdict-files.test.ts
@@ -0,0 +1,824 @@
+import { execFileSync, spawn } from "node:child_process";
+import { createHash } from "node:crypto";
+import {
+  existsSync,
+  mkdirSync,
+  mkdtempSync,
+  readFileSync,
+  readdirSync,
+  realpathSync,
+  unlinkSync,
+  writeFileSync,
+} from "node:fs";
+import { tmpdir } from "node:os";
+import { join, resolve } from "node:path";
+import { describe, expect, it } from "vitest";
+import { buildContext } from "../src/check.js";
+import { computeReviewId, fileParts } from "../src/hash.js";
+import { evidenceBlockParts } from "../src/annotations.js";
+import { matchGlobs } from "../src/files.js";
+
+const CLI = resolve(import.meta.dirname, "../dist/cli.js");
+
+interface RunResult {
+  status: number;
+  stdout: string;
+  stderr: string;
+}
+
+function run(cwd: string, args: string[]): RunResult {
+  try {
+    return { status: 0, stdout: execFileSync("node", [CLI, ...args], { cwd, encoding: "utf8" }), stderr: "" };
+  } catch (err) {
+    const e = err as { status: number; stdout: string; stderr: string };
+    return { status: e.status, stdout: e.stdout ?? "", stderr: e.stderr ?? "" };
+  }
+}
+
+function startRun(cwd: string, args: string[]) {
+  const child = spawn("node", [CLI, ...args], { cwd, stdio: ["ignore", "pipe", "pipe"] });
+  const result = new Promise<RunResult>((resolveResult) => {
+    let stdout = "";
+    let stderr = "";
+    child.stdout.setEncoding("utf8").on("data", (chunk) => (stdout += chunk));
+    child.stderr.setEncoding("utf8").on("data", (chunk) => (stderr += chunk));
+    child.on("close", (code) => resolveResult({ status: code ?? 1, stdout, stderr }));
+  });
+  return { child, result };
+}
+
+const SPEC = `# FIX-001: Widgets
+
+## Overview
+
+Widgets.
+
+## Requirements
+
+### FIX-001.1: Basics
+
+1. The widget MUST spin.
+2. The widget MUST stop.
+`;
+
+function fixture(): string {
+  const root = realpathSync(mkdtempSync(join(tmpdir(), "2119-stable-verdict-")));
+  writeFileSync(join(root, ".2119.yml"), 'prefix: "FIX"\n');
+  mkdirSync(join(root, "specs"));
+  mkdirSync(join(root, "tests"));
+  writeFileSync(join(root, "specs/FIX-001-widgets.md"), SPEC);
+  writeFileSync(
+    join(root, "tests/widget.test.js"),
+    "// 2119: FIX-001.1.1\ntest('spins', () => spin())\n" +
+      "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+  );
+  return root;
+}
+
+function reviewIds(root: string): Record<string, string> {
+  run(root, ["review"]);
+  const names = readdirSync(join(root, ".2119/reviews"));
+  return Object.fromEntries(
+    names
+      .filter((name) => name.endsWith(".md") && !name.endsWith(".audit.md"))
+      .map((name) => {
+        const id = name.replace(/\.md$/, "");
+        return [id.replace(/--[0-9a-f]{12}$/, ""), id];
+      }),
+  );
+}
+
+function record(reviewId: string, verdict: "pass" | "fail", summary = "honest review") {
+  return {
+    reviewId,
+    requirementId: reviewId.replace(/--[0-9a-f]{12}$/, ""),
+    hash: reviewId.slice(-12),
+    verdict,
+    summary,
+    timestamp: "2026-08-03T00:00:00.000Z",
+  };
+}
+
+function writeRecord(root: string, filename: string, value: ReturnType<typeof record>): void {
+  mkdirSync(join(root, ".2119/verdicts"), { recursive: true });
+  writeFileSync(join(root, ".2119/verdicts", filename), `${JSON.stringify(value, null, 2)}\n`);
+}
+
+const stablePath = (root: string, requirementId: string) => join(root, ".2119/verdicts", `${requirementId}.json`);
+
+function produceRecord(
+  root: string,
+  reviewId: string,
+  verdict: "pass" | "fail",
+  summary: string,
+): ReturnType<typeof record> {
+  expect(run(root, [verdict, reviewId, "--summary", summary]).status).toBe(0);
+  const requirementId = reviewId.replace(/--[0-9a-f]{12}$/, "");
+  const stable = stablePath(root, requirementId);
+  const legacy = join(root, ".2119/verdicts", `${reviewId}.json`);
+  const produced = existsSync(stable) ? stable : legacy;
+  return JSON.parse(readFileSync(produced, "utf8"));
+}
+
+function placeLegacy(root: string, reviewId: string, value: ReturnType<typeof record>): void {
+  const stable = stablePath(root, value.requirementId);
+  if (existsSync(stable)) unlinkSync(stable);
+  writeRecord(root, `${reviewId}.json`, value);
+}
+
+describe("stable verdict files (REQ-012)", () => {
+  // 2119: REQ-012.1.1
+  it("records one stable file and supersedes both its prior stable record and legacy files", () => {
+    const root = fixture();
+    const first = reviewIds(root)["FIX-001.1.1"];
+    expect(run(root, ["pass", first, "--summary", "first honest review"]).status).toBe(0);
+    const path = stablePath(root, "FIX-001.1.1");
+    expect(JSON.parse(readFileSync(path, "utf8")).reviewId).toBe(first);
+
+    // Replacement is required even when the review input and hash did not
+    // change between two successful recording commands.
+    expect(run(root, ["fail", first, "--summary", "same-hash replacement"]).status).toBe(0);
+    expect(JSON.parse(readFileSync(path, "utf8"))).toMatchObject({
+      reviewId: first,
+      verdict: "fail",
+      summary: "same-hash replacement",
+    });
+    expect(run(root, ["pass", first, "--summary", "restored same-hash pass"]).status).toBe(0);
+    expect(JSON.parse(readFileSync(path, "utf8"))).toMatchObject({
+      reviewId: first,
+      verdict: "pass",
+      summary: "restored same-hash pass",
+    });
+
+    // A legacy duplicate must not survive the next successful recording.
+    writeRecord(root, `${first}.json`, record(first, "pass", "legacy duplicate"));
+    expect(run(root, ["pass", first, "--summary", "pass cleans legacy duplicate"]).status).toBe(0);
+    expect(readdirSync(join(root, ".2119/verdicts")).filter((name) => name.startsWith("FIX-001.1.1"))).toEqual([
+      "FIX-001.1.1.json",
+    ]);
+    writeRecord(root, `${first}.json`, record(first, "pass", "legacy duplicate for changed hash"));
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins strongly', () => spin())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    const second = reviewIds(root)["FIX-001.1.1"];
+    expect(second).not.toBe(first);
+    expect(run(root, ["fail", second, "--summary", "new test is dishonest"]).status).toBe(0);
+
+    const files = readdirSync(join(root, ".2119/verdicts")).filter((name) => name.startsWith("FIX-001.1.1"));
+    expect(files).toEqual(["FIX-001.1.1.json"]);
+    expect(JSON.parse(readFileSync(path, "utf8"))).toMatchObject({ reviewId: second, verdict: "fail" });
+  });
+
+  // 2119: REQ-012.2.1
+  it("isolates overlapping recordings for distinct requirements", async () => {
+    const root = fixture();
+    const ids = reviewIds(root);
+    const spinRun = startRun(root, ["pass", ids["FIX-001.1.1"], "--summary", "spin behavior covered"]);
+    const stopRun = startRun(root, ["fail", ids["FIX-001.1.2"], "--summary", "stop boundary missing"]);
+    // Both production CLI processes exist simultaneously before either is
+    // awaited, so this exercises the requirement's overlapping-command case.
+    const [spin, stop] = await Promise.all([spinRun.result, stopRun.result]);
+    expect([spin.status, stop.status]).toEqual([0, 0]);
+    const spinRecord = JSON.parse(readFileSync(stablePath(root, "FIX-001.1.1"), "utf8"));
+    const stopRecord = JSON.parse(readFileSync(stablePath(root, "FIX-001.1.2"), "utf8"));
+    expect(spinRecord).toMatchObject({
+      reviewId: ids["FIX-001.1.1"],
+      requirementId: "FIX-001.1.1",
+      hash: ids["FIX-001.1.1"].slice(-12),
+      verdict: "pass",
+      summary: "spin behavior covered",
+    });
+    expect(stopRecord).toMatchObject({
+      reviewId: ids["FIX-001.1.2"],
+      requirementId: "FIX-001.1.2",
+      hash: ids["FIX-001.1.2"].slice(-12),
+      verdict: "fail",
+      summary: "stop boundary missing",
+    });
+    expect(Number.isNaN(Date.parse(spinRecord.timestamp))).toBe(false);
+    expect(Number.isNaN(Date.parse(stopRecord.timestamp))).toBe(false);
+  });
+
+  // 2119: REQ-012.3.1
+  it("stores a complete internally content-addressed record at the canonical stable path", () => {
+    const root = fixture();
+    const reviewId = reviewIds(root)["FIX-001.1.1"];
+    expect(run(root, ["pass", reviewId, "--summary", "spin assertion reaches production"]).status).toBe(0);
+    const saved = JSON.parse(readFileSync(stablePath(root, "FIX-001.1.1"), "utf8"));
+    expect(saved).toMatchObject({
+      reviewId,
+      requirementId: "FIX-001.1.1",
+      hash: reviewId.slice(-12),
+      verdict: "pass",
+      summary: "spin assertion reaches production",
+    });
+    expect(saved.hash).toMatch(/^[0-9a-f]{12}$/);
+    expect(Number.isNaN(Date.parse(saved.timestamp))).toBe(false);
+
+    expect(run(root, ["fail", reviewId, "--summary", "spin boundary is missing"]).status).toBe(0);
+    const savedFail = JSON.parse(readFileSync(stablePath(root, "FIX-001.1.1"), "utf8"));
+    expect(savedFail).toMatchObject({
+      reviewId,
+      requirementId: "FIX-001.1.1",
+      hash: reviewId.slice(-12),
+      verdict: "fail",
+      summary: "spin boundary is missing",
+    });
+    expect(savedFail.hash).toMatch(/^[0-9a-f]{12}$/);
+    expect(Number.isNaN(Date.parse(savedFail.timestamp))).toBe(false);
+    expect(run(root, ["pass", reviewId, "--summary", "spin assertion reaches production"]).status).toBe(0);
+
+    // A second current pass keeps the unrelated requirement from masking
+    // schema failures in the record under test.
+    const ids = reviewIds(root);
+    writeRecord(root, "FIX-001.1.2.json", record(ids["FIX-001.1.2"], "pass"));
+    const valid = record(reviewId, "pass", "spin assertion reaches production");
+    const malformed: Array<[string, object]> = [
+      ["requirement ID", { ...valid, requirementId: "FIX-001.1.9" }],
+      ["review ID", { ...valid, reviewId: ids["FIX-001.1.2"] }],
+      ["uppercase hash", { ...valid, hash: valid.hash.toUpperCase() }],
+      ["mismatched hash", { ...valid, hash: "0".repeat(12) }],
+      ["invalid verdict", { ...valid, verdict: "passed" }],
+      ["empty summary", { ...valid, summary: "" }],
+      ["invalid timestamp", { ...valid, timestamp: "not-a-date" }],
+    ];
+    for (const field of ["reviewId", "requirementId", "hash", "verdict", "summary", "timestamp"] as const) {
+      const missing = { ...valid } as Record<string, unknown>;
+      delete missing[field];
+      malformed.push([`missing ${field}`, missing]);
+    }
+    for (const [, candidate] of malformed) {
+      writeFileSync(stablePath(root, "FIX-001.1.1"), `${JSON.stringify(candidate)}\n`);
+      expect(run(root, ["check"]).status).toBe(1);
+    }
+    writeFileSync(stablePath(root, "FIX-001.1.1"), `${JSON.stringify(valid)}\n`);
+    writeFileSync(join(root, ".2119/verdicts/FIX-001.1.1-wrong.json"), `${JSON.stringify(valid)}\n`);
+    const wrongPath = run(root, ["check"]);
+    expect(wrongPath.status).toBe(1);
+    expect(wrongPath.stderr).toContain("FIX-001.1.1-wrong.json");
+  });
+
+  // 2119: REQ-012.3.2
+  it("changes hashes for every review input but leaves both review kinds stable across unrelated edits", () => {
+    const root = realpathSync(mkdtempSync(join(tmpdir(), "2119-hash-boundary-")));
+    mkdirSync(join(root, "specs"));
+    mkdirSync(join(root, "tests/shared"), { recursive: true });
+    mkdirSync(join(root, "docs"));
+    mkdirSync(join(root, ".2119/review"), { recursive: true });
+    mkdirSync(join(root, "src"));
+    writeFileSync(
+      join(root, ".2119.yml"),
+      'prefix: "FIX"\nshared_evidence: ["tests/shared/**"]\n',
+    );
+    writeFileSync(
+      join(root, "specs/FIX-001-widgets.md"),
+      SPEC.replace(
+        "2. The widget MUST stop.",
+        "2. The widget MUST stop.\n3. Policy MUST match the guide. [review: docs/**, instructions: .2119/review/policy.md]",
+      ),
+    );
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "import { fixture } from './shared/setup.js';\n" +
+        "// 2119: FIX-001.1.1\ntest('spins', () => spin())\n" +
+        "// 2119: FIX-001.1.1\ntest('spins again', () => spinAgain())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    writeFileSync(
+      join(root, "tests/widget-extra.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins from another file', () => externalSpin())\n",
+    );
+    writeFileSync(join(root, "tests/shared/setup.js"), "export const fixture = 'v1';\n");
+    writeFileSync(join(root, "docs/guide.md"), "policy v1\n");
+    writeFileSync(join(root, ".2119/review/policy.md"), "reject contradictions\n");
+    writeFileSync(join(root, "src/unrelated.ts"), "export const unrelated = 1;\n");
+
+    const ids = () => Object.fromEntries(buildContext(root).allReviewTargets.map((t) => [t.requirement.id, t.reviewId]));
+    const before = ids();
+    writeFileSync(join(root, "src/unrelated.ts"), "export const unrelated = 2;\n");
+    expect(ids()).toEqual(before);
+
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "import { fixture as sharedFixture } from './shared/setup.js';\n" +
+        "// 2119: FIX-001.1.1\ntest('spins', () => spin())\n" +
+        "// 2119: FIX-001.1.1\ntest('spins again', () => spinAgain())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    const preludeEdit = ids();
+    expect(preludeEdit["FIX-001.1.1"]).not.toBe(before["FIX-001.1.1"]);
+    expect(preludeEdit["FIX-001.1.2"]).not.toBe(before["FIX-001.1.2"]);
+    expect(preludeEdit["FIX-001.1.3"]).toBe(before["FIX-001.1.3"]);
+
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "import { fixture as sharedFixture } from './shared/setup.js';\n" +
+        "// 2119: FIX-001.1.1\ntest('spins twice', () => spin())\n" +
+        "// 2119: FIX-001.1.1\ntest('spins again', () => spinAgain())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    const coveredEdit = ids();
+    expect(coveredEdit["FIX-001.1.1"]).not.toBe(preludeEdit["FIX-001.1.1"]);
+    expect(coveredEdit["FIX-001.1.2"]).toBe(preludeEdit["FIX-001.1.2"]);
+    expect(coveredEdit["FIX-001.1.3"]).toBe(preludeEdit["FIX-001.1.3"]);
+
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "import { fixture as sharedFixture } from './shared/setup.js';\n" +
+        "// 2119: FIX-001.1.1\ntest('spins twice', () => spin())\n" +
+        "// 2119: FIX-001.1.1\ntest('spins a third way', () => spinAgain())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    const laterBlockEdit = ids();
+    expect(laterBlockEdit["FIX-001.1.1"]).not.toBe(coveredEdit["FIX-001.1.1"]);
+    expect(laterBlockEdit["FIX-001.1.2"]).toBe(coveredEdit["FIX-001.1.2"]);
+    expect(laterBlockEdit["FIX-001.1.3"]).toBe(coveredEdit["FIX-001.1.3"]);
+
+    writeFileSync(
+      join(root, "tests/widget-extra.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins differently from another file', () => externalSpin())\n",
+    );
+    const extraBlockEdit = ids();
+    expect(extraBlockEdit["FIX-001.1.1"]).not.toBe(laterBlockEdit["FIX-001.1.1"]);
+    expect(extraBlockEdit["FIX-001.1.2"]).toBe(laterBlockEdit["FIX-001.1.2"]);
+    expect(extraBlockEdit["FIX-001.1.3"]).toBe(laterBlockEdit["FIX-001.1.3"]);
+
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "import { fixture as sharedFixture } from './shared/setup.js';\n" +
+        "// 2119: FIX-001.1.1\ntest('spins twice', () => spin())\n" +
+        "// 2119: FIX-001.1.1\ntest('spins a third way', () => spinAgain())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops immediately', () => stop())\n",
+    );
+    const neighboringBlockEdit = ids();
+    expect(neighboringBlockEdit["FIX-001.1.1"]).toBe(extraBlockEdit["FIX-001.1.1"]);
+    expect(neighboringBlockEdit["FIX-001.1.2"]).not.toBe(extraBlockEdit["FIX-001.1.2"]);
+    expect(neighboringBlockEdit["FIX-001.1.3"]).toBe(extraBlockEdit["FIX-001.1.3"]);
+
+    writeFileSync(join(root, "tests/shared/setup.js"), "export const fixture = 'v2';\n");
+    const sharedEdit = ids();
+    expect(sharedEdit["FIX-001.1.1"]).not.toBe(neighboringBlockEdit["FIX-001.1.1"]);
+    expect(sharedEdit["FIX-001.1.2"]).not.toBe(neighboringBlockEdit["FIX-001.1.2"]);
+    expect(sharedEdit["FIX-001.1.3"]).toBe(neighboringBlockEdit["FIX-001.1.3"]);
+
+    writeFileSync(join(root, "docs/guide.md"), "policy v2\n");
+    const evidenceEdit = ids();
+    expect(evidenceEdit["FIX-001.1.3"]).not.toBe(sharedEdit["FIX-001.1.3"]);
+    expect(evidenceEdit["FIX-001.1.1"]).toBe(sharedEdit["FIX-001.1.1"]);
+    expect(evidenceEdit["FIX-001.1.2"]).toBe(sharedEdit["FIX-001.1.2"]);
+    writeFileSync(join(root, ".2119/review/policy.md"), "reject contradictions and omissions\n");
+    const instructionEdit = ids();
+    expect(instructionEdit["FIX-001.1.3"]).not.toBe(evidenceEdit["FIX-001.1.3"]);
+    expect(instructionEdit["FIX-001.1.1"]).toBe(evidenceEdit["FIX-001.1.1"]);
+    expect(instructionEdit["FIX-001.1.2"]).toBe(evidenceEdit["FIX-001.1.2"]);
+
+    // REQ-003's canonical whole-file ordering and exact hash stream are
+    // observable here, not inferred from selected sensitivity checks.
+    const beforeAdditionalMatches = ids();
+    writeFileSync(join(root, "docs/a.md"), "A\n");
+    writeFileSync(join(root, "docs/b.md"), "B\n");
+    const additionalMatches = ids();
+    expect(additionalMatches["FIX-001.1.3"]).not.toBe(beforeAdditionalMatches["FIX-001.1.3"]);
+    expect(additionalMatches["FIX-001.1.1"]).toBe(beforeAdditionalMatches["FIX-001.1.1"]);
+    expect(additionalMatches["FIX-001.1.2"]).toBe(beforeAdditionalMatches["FIX-001.1.2"]);
+    const parts = fileParts(root, ["docs/b.md", ".2119/review/policy.md", "docs/a.md"]);
+    expect(parts.map((part) => part.label)).toEqual([".2119/review/policy.md", "docs/a.md", "docs/b.md"]);
+    const exact = createHash("sha256").update("Policy MUST match the guide.");
+    for (const part of parts) exact.update("\0").update(part.label).update("\0").update(part.content);
+    const digest = exact.digest("hex").slice(0, 12);
+    expect(computeReviewId("FIX-001.1.3", "Policy MUST match the guide.", parts)).toBe(
+      `FIX-001.1.3--${digest}`,
+    );
+    expect(computeReviewId("FIX-001.1.3", "Policy MUST match every guide.", parts)).not.toBe(
+      `FIX-001.1.3--${digest}`,
+    );
+    expect(computeReviewId("FIX-001.1.3", "Policy MUST match the guide.", [...parts].reverse())).not.toBe(
+      `FIX-001.1.3--${digest}`,
+    );
+
+    const ctx = buildContext(root);
+    const reviewRequirement = ctx.allReviewTargets.find((target) => target.requirement.id === "FIX-001.1.3")!;
+    const matchedReviewParts = fileParts(root, [
+      reviewRequirement.requirement.coverage.instructions!,
+      ...matchGlobs(ctx.repoFiles, reviewRequirement.requirement.coverage.globs!),
+    ]);
+    expect(computeReviewId("FIX-001.1.3", reviewRequirement.requirement.text, matchedReviewParts)).toBe(
+      reviewRequirement.reviewId,
+    );
+    const annotations = ctx.annotations.filter((annotation) => annotation.ids.includes("FIX-001.1.1"));
+    const testParts = [
+      ...evidenceBlockParts(root, annotations, ctx.annotations, ctx.config.prefix, ctx.markerLineByFile),
+      ...fileParts(root, matchGlobs(ctx.repoFiles, ctx.config.sharedEvidence)),
+    ];
+    const annotationLabels = testParts
+      .filter((part) => part.label.includes("widget") && part.label.includes("#"))
+      .map((part) => part.label);
+    // Two file preludes plus three covering annotation blocks.
+    expect(annotationLabels).toEqual([
+      "tests/widget-extra.test.js#prelude",
+      "tests/widget-extra.test.js#0",
+      "tests/widget.test.js#prelude",
+      "tests/widget.test.js#0",
+      "tests/widget.test.js#1",
+    ]);
+    const testTarget = ctx.allReviewTargets.find((target) => target.requirement.id === "FIX-001.1.1")!;
+    expect(computeReviewId("FIX-001.1.1", testTarget.requirement.text, testParts)).toBe(testTarget.reviewId);
+    expect(computeReviewId("FIX-001.1.1", testTarget.requirement.text, [...testParts].reverse())).not.toBe(
+      testTarget.reviewId,
+    );
+
+    expect(run(root, ["pass", testTarget.reviewId, "--summary", "hash boundaries verified"]).status).toBe(0);
+    expect(JSON.parse(readFileSync(stablePath(root, "FIX-001.1.1"), "utf8")).hash).toBe(
+      testTarget.reviewId.slice(-12),
+    );
+    expect(run(root, ["pass", reviewRequirement.reviewId, "--summary", "review evidence hash verified"]).status).toBe(0);
+    expect(JSON.parse(readFileSync(stablePath(root, "FIX-001.1.3"), "utf8")).hash).toBe(
+      reviewRequirement.reviewId.slice(-12),
+    );
+  });
+
+  // 2119: REQ-012.4.1
+  it("accepts only a current stable pass and never bypasses stable-path defects with legacy evidence", () => {
+    const root = fixture();
+    const current = reviewIds(root)["FIX-001.1.1"];
+    const stable = stablePath(root, "FIX-001.1.1");
+    const ids = reviewIds(root);
+    expect(run(root, ["pass", current, "--summary", "current stable pass"]).status).toBe(0);
+    expect(run(root, ["pass", ids["FIX-001.1.2"], "--summary", "other current stable pass"]).status).toBe(0);
+    expect(JSON.parse(readFileSync(stablePath(root, "FIX-001.1.1"), "utf8")).reviewId).toBe(current);
+    expect(run(root, ["check"]).status).toBe(0);
+
+    const staleId = current.replace(/[0-9a-f]{12}$/, "000000000000");
+    writeRecord(root, "FIX-001.1.1.json", record(staleId, "pass", "obsolete pass"));
+    writeRecord(root, `${current}.json`, record(current, "pass", "legacy bypass attempt"));
+    const stale = run(root, ["check"]);
+    expect(stale.status).toBe(1);
+    expect(`${stale.stdout}\n${stale.stderr}`).toMatch(/stale/i);
+
+    const currentIdWrongHash = { ...record(current, "pass"), hash: "f".repeat(12) };
+    writeFileSync(stable, `${JSON.stringify(currentIdWrongHash)}\n`);
+    const mismatchedHash = run(root, ["check"]);
+    expect(mismatchedHash.status).toBe(1);
+    expect(mismatchedHash.stderr).toMatch(/hash/i);
+
+    writeRecord(root, "FIX-001.1.1.json", record(current, "fail", "boundary missing"));
+    const failing = run(root, ["check"]);
+    expect(failing.status).toBe(1);
+    expect(`${failing.stdout}\n${failing.stderr}`).toContain("boundary missing");
+
+    writeFileSync(stable, `${JSON.stringify({ ...record(current, "pass"), summary: "" })}\n`);
+    const structurallyMalformed = run(root, ["check"]);
+    expect(structurallyMalformed.status).toBe(1);
+    expect(structurallyMalformed.stderr).toMatch(/summary/i);
+
+    for (const candidate of [
+      { ...record(current, "pass"), requirementId: "FIX-001.1.9" },
+      { ...record(current, "pass"), verdict: "passed" },
+      { ...record(current, "pass"), timestamp: "not-a-date" },
+      { hash: current.slice(-12), verdict: "pass", summary: "missing IDs", timestamp: "2026-08-03T00:00:00Z" },
+      { ...record(current, "pass"), hash: undefined },
+      { ...record(current, "pass"), timestamp: undefined },
+    ]) {
+      writeFileSync(stable, `${JSON.stringify(candidate)}\n`);
+      expect(run(root, ["check"]).status).toBe(1);
+    }
+
+    writeFileSync(stable, "{broken");
+    const malformed = run(root, ["check"]);
+    expect(malformed.status).toBe(1);
+    expect(malformed.stderr).toContain("malformed verdict file");
+  });
+
+  // 2119: REQ-012.5.1
+  it("uses only current valid legacy evidence during transition and diagnoses every legacy state", () => {
+    const root = fixture();
+    const firstIds = reviewIds(root);
+    const supersededPass = produceRecord(root, firstIds["FIX-001.1.1"], "pass", "obsolete pass");
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins v2', () => spin())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    const secondIds = reviewIds(root);
+    const supersededFail = produceRecord(root, secondIds["FIX-001.1.1"], "fail", "obsolete failure");
+    writeFileSync(
+      join(root, "tests/widget.test.js"),
+      "// 2119: FIX-001.1.1\ntest('spins v3', () => spin())\n" +
+        "// 2119: FIX-001.1.2\ntest('stops', () => stop())\n",
+    );
+    const ids = reviewIds(root);
+    placeLegacy(
+      root,
+      ids["FIX-001.1.1"],
+      produceRecord(root, ids["FIX-001.1.1"], "pass", "current legacy pass"),
+    );
+    placeLegacy(
+      root,
+      ids["FIX-001.1.2"],
+      produceRecord(root, ids["FIX-001.1.2"], "pass", "other current legacy pass"),
+    );
+    const obsoleteId = firstIds["FIX-001.1.1"];
+    const supersededFailId = secondIds["FIX-001.1.1"];
+    writeRecord(root, `${obsoleteId}.json`, supersededPass);
+    writeRecord(root, `${supersededFailId}.json`, supersededFail);
+    const accepted = run(root, ["check"]);
+    expect(accepted.status).toBe(0);
+    expect(`${accepted.stdout}\n${accepted.stderr}`).toContain("migration");
+    expect(`${accepted.stdout}\n${accepted.stderr}`).toContain(`${ids["FIX-001.1.1"]}.json`);
+    expect(`${accepted.stdout}\n${accepted.stderr}`).toContain("FIX-001.1.1.json");
+    expect(`${accepted.stdout}\n${accepted.stderr}`).toMatch(
+      new RegExp(`${ids["FIX-001.1.1"]}\\.json[^\\n]*FIX-001\\.1\\.1\\.json`),
+    );
+    expect(`${accepted.stdout}\n${accepted.stderr}`).toMatch(
+      new RegExp(`${ids["FIX-001.1.2"]}\\.json[^\\n]*FIX-001\\.1\\.2\\.json`),
+    );
+
+    const currentRecord = record(ids["FIX-001.1.1"], "pass", "current legacy pass");
+    for (const candidate of [
+      { ...currentRecord, verdict: "passed" },
+      { ...currentRecord, requirementId: "FIX-001.1.9" },
+      { ...currentRecord, hash: "f".repeat(12) },
+      { ...currentRecord, timestamp: "not-a-date" },
+    ]) {
+      writeFileSync(
+        join(root, ".2119/verdicts", `${ids["FIX-001.1.1"]}.json`),
+        `${JSON.stringify(candidate)}\n`,
+      );
+      const malformedCurrent = run(root, ["check"]);
+      expect(malformedCurrent.status).toBe(1);
+      expect(malformedCurrent.stderr).toContain(`${ids["FIX-001.1.1"]}.json`);
+      expect(malformedCurrent.stderr).toMatch(/malformed/i);
+    }
+    placeLegacy(
+      root,
+      ids["FIX-001.1.1"],
+      produceRecord(root, ids["FIX-001.1.1"], "pass", "current legacy pass"),
+    );
+
+    const nonTwelveName = "FIX-001.1.1--abc";
+    writeRecord(root, `${nonTwelveName}.json`, currentRecord);
+    const looseHashName = run(root, ["check"]);
+    expect(looseHashName.status).toBe(1);
+    expect(looseHashName.stderr).toContain(`${nonTwelveName}.json`);
+    unlinkSync(join(root, ".2119/verdicts", `${nonTwelveName}.json`));
+
+    const overlappingId = `FIX-001.1.10--${"c".repeat(12)}`;
+    writeRecord(root, `${overlappingId}.json`, record(overlappingId, "pass", "different requirement"));
+    unlinkSync(join(root, ".2119/verdicts", `${ids["FIX-001.1.1"]}.json`));
+    expect(run(root, ["check"]).status).toBe(1);
+    placeLegacy(
+      root,
+      ids["FIX-001.1.1"],
+      produceRecord(root, ids["FIX-001.1.1"], "pass", "current legacy pass"),
+    );
+    // The normal writer correctly removes this requirement's legacy files;
+    // restore historical fixtures to exercise the upgrade diagnostics below.
+    writeRecord(root, `${obsoleteId}.json`, supersededPass);
+    writeRecord(root, `${supersededFailId}.json`, supersededFail);
+
+    const withSuperseded = run(root, ["check"]);
+    expect(withSuperseded.status).toBe(0);
+    expect(`${withSuperseded.stdout}\n${withSuperseded.stderr}`).toMatch(/superseded/i);
+    expect(`${withSuperseded.stdout}\n${withSuperseded.stderr}`).toContain(`${obsoleteId}.json`);
+    expect(`${withSuperseded.stdout}\n${withSuperseded.stderr}`).toContain("FIX-001.1.1.json");
+    expect(`${withSuperseded.stdout}\n${withSuperseded.stderr}`).toMatch(
+      new RegExp(`${obsoleteId}\\.json[^\\n]*FIX-001\\.1\\.1\\.json`),
+    );
+
+    const withSupersededFail = run(root, ["check"]);
+    expect(withSupersededFail.status).toBe(0);
+    expect(`${withSupersededFail.stdout}\n${withSupersededFail.stderr}`).toMatch(
+      new RegExp(`${supersededFailId}\\.json[^\\n]*FIX-001\\.1\\.1\\.json`),
+    );
+
+    placeLegacy(root, ids["FIX-001.1.1"], produceRecord(root, ids["FIX-001.1.1"], "fail", "legacy fail"));
+    const legacyFail = run(root, ["check"]);
+    expect(legacyFail.status).toBe(1);
+    expect(`${legacyFail.stdout}\n${legacyFail.stderr}`).toContain("legacy fail");
+    expect(`${legacyFail.stdout}\n${legacyFail.stderr}`).toMatch(
+      new RegExp(`${ids["FIX-001.1.1"]}\\.json[^\\n]*FIX-001\\.1\\.1\\.json`),
+    );
+    writeRecord(root, `${ids["FIX-001.1.1"]}.json`, record(ids["FIX-001.1.1"], "pass"));
+
+    writeFileSync(join(root, ".2119/verdicts", `${obsoleteId}.json`), `${JSON.stringify({ ...record(obsoleteId, "pass"), summary: "" })}\n`);
+    const malformedSuperseded = run(root, ["check"]);
+    expect(malformedSuperseded.status).toBe(1);
+    expect(malformedSuperseded.stderr).toContain(`${obsoleteId}.json`);
+    writeRecord(root, `${obsoleteId}.json`, supersededPass);
+
+    writeFileSync(join(root, ".2119/verdicts", `${obsoleteId}.json`), "{broken");
+    writeFileSync(join(root, ".2119/verdicts", `${supersededFailId}.json`), "{also-broken");
+    const everyMalformed = run(root, ["check"]);
+    expect(everyMalformed.status).toBe(1);
+    expect(everyMalformed.stderr).toContain(`${obsoleteId}.json`);
+    expect(everyMalformed.stderr).toContain(`${supersededFailId}.json`);
+    writeRecord(root, `${obsoleteId}.json`, supersededPass);
+    writeRecord(root, `${supersededFailId}.json`, supersededFail);
+
+    writeRecord(root, `${supersededFailId}.json`, supersededPass);
+    const mismatchedLegacy = run(root, ["check"]);
+    expect(mismatchedLegacy.status).toBe(1);
+    expect(mismatchedLegacy.stderr).toContain(`${supersededFailId}.json`);
+    writeRecord(root, `${supersededFailId}.json`, supersededFail);
+
+    unlinkSync(join(root, ".2119/verdicts", `${ids["FIX-001.1.1"]}.json`));
+    expect(run(root, ["check"]).status).toBe(1); // the superseded pass cannot satisfy freshness
+
+    writeFileSync(join(root, ".2119/verdicts", `${ids["FIX-001.1.1"]}.json`), "{broken");
+    const malformed = run(root, ["check"]);
+    expect(malformed.status).toBe(1);
+    expect(malformed.stderr).toContain(`${ids["FIX-001.1.1"]}.json`);
+    expect(malformed.stderr).toContain("unparseable JSON");
+  });
+
+  // 2119: REQ-006.2.2, REQ-012.6.1, REQ-012.6.2
+  it("prune preserves the best current evidence, removes all legacy files, and reports every migration action", () => {
+    const root = fixture();
+    const ids = reviewIds(root);
+    const currentStable = produceRecord(root, ids["FIX-001.1.1"], "pass", "authoritative current pass");
+    const conflictingLegacy = produceRecord(root, ids["FIX-001.1.1"], "fail", "conflicting legacy");
+    const migratable = produceRecord(root, ids["FIX-001.1.2"], "pass", "current legacy evidence");
+    writeRecord(root, "FIX-001.1.1.json", currentStable);
+    writeRecord(root, `${ids["FIX-001.1.1"]}.json`, conflictingLegacy);
+
+    const staleStable = record(`FIX-001.1.2--${"0".repeat(12)}`, "pass", "stale stable");
+    writeRecord(root, "FIX-001.1.2.json", staleStable);
+    writeRecord(root, `${ids["FIX-001.1.2"]}.json`, migratable);
+    const staleExisting = `FIX-001.1.2--${"d".repeat(12)}`;
+    writeRecord(root, `${staleExisting}.json`, record(staleExisting, "pass", "stale existing requirement"));
+    const orphan = `FIX-999.1.1--${"a".repeat(12)}`;
+    writeRecord(root, `${orphan}.json`, record(orphan, "pass", "orphan"));
+
+    const pruned = run(root, ["prune"]);
+    expect(pruned.status).toBe(0);
+    expect(JSON.parse(readFileSync(stablePath(root, "FIX-001.1.1"), "utf8"))).toEqual(currentStable);
+    expect(JSON.parse(readFileSync(stablePath(root, "FIX-001.1.2"), "utf8"))).toEqual(migratable);
+    expect(readdirSync(join(root, ".2119/verdicts")).filter((name) => name.endsWith(".json")).sort()).toEqual([
+      "FIX-001.1.1.json",
+      "FIX-001.1.2.json",
+    ]);
+    expect(pruned.stdout).toMatch(
+      new RegExp(`(?:convert|migrat)[^\\n]*${ids["FIX-001.1.2"]}\\.json[^\\n]*FIX-001\\.1\\.2\\.json`, "i"),
+    );
+    for (const removed of [ids["FIX-001.1.1"], staleExisting, orphan]) {
+      expect(pruned.stdout).toMatch(new RegExp(`(?:remove|prune)[^\\n]*${removed}\\.json`, "i"));
+    }
+    expect(run(root, ["check"]).status).toBe(0);
+
+    // Current failing evidence is still evidence: retain a canonical fail and
+    // migrate a legacy fail without weakening or dropping either record.
+    const failureRoot = fixture();
+    const failureIds = reviewIds(failureRoot);
+    const stableFailure = produceRecord(
+      failureRoot,
+      failureIds["FIX-001.1.1"],
+      "fail",
+      "current stable failure",
+    );
+    const legacyFailure = produceRecord(
+      failureRoot,
+      failureIds["FIX-001.1.2"],
+      "fail",
+      "current legacy failure",
+    );
+    writeRecord(failureRoot, `${failureIds["FIX-001.1.2"]}.json`, legacyFailure);
+    unlinkSync(stablePath(failureRoot, "FIX-001.1.2"));
+    const failurePrune = run(failureRoot, ["prune"]);
+    expect(failurePrune.status).toBe(0);
+    expect(JSON.parse(readFileSync(stablePath(failureRoot, "FIX-001.1.1"), "utf8"))).toEqual(stableFailure);
+    expect(JSON.parse(readFileSync(stablePath(failureRoot, "FIX-001.1.2"), "utf8"))).toEqual(legacyFailure);
+    expect(
+      readdirSync(join(failureRoot, ".2119/verdicts"))
+        .filter((name) => name.endsWith(".json"))
+        .sort(),
+    ).toEqual(["FIX-001.1.1.json", "FIX-001.1.2.json"]);
+    expect(failurePrune.stdout).toMatch(
+      new RegExp(
+        `(?:convert|migrat)[^\\n]*${failureIds["FIX-001.1.2"]}\\.json[^\\n]*FIX-001\\.1\\.2\\.json`,
+        "i",
+      ),
+    );
+
+    // A malformed stable record cannot block recovery from valid current
+    // legacy evidence, and malformed legacy debris is still removed/listed.
+    const recoveryRoot = fixture();
+    const recoveryIds = reviewIds(recoveryRoot);
+    const recovered = produceRecord(
+      recoveryRoot,
+      recoveryIds["FIX-001.1.1"],
+      "pass",
+      "recoverable current evidence",
+    );
+    writeFileSync(stablePath(recoveryRoot, "FIX-001.1.1"), "{broken");
+    writeRecord(recoveryRoot, `${recoveryIds["FIX-001.1.1"]}.json`, recovered);
+    const malformedLegacy = `FIX-001.1.1--${"b".repeat(12)}`;
+    writeFileSync(join(recoveryRoot, ".2119/verdicts", `${malformedLegacy}.json`), "{broken");
+    const recovery = run(recoveryRoot, ["prune"]);
+    expect(recovery.status).toBe(0);
+    expect(JSON.parse(readFileSync(stablePath(recoveryRoot, "FIX-001.1.1"), "utf8"))).toEqual(recovered);
+    expect(readdirSync(join(recoveryRoot, ".2119/verdicts")).filter((name) => name.startsWith("FIX-001.1.1"))).toEqual([
+      "FIX-001.1.1.json",
+    ]);
+    expect(recovery.stdout).toMatch(
+      new RegExp(
+        `(?:convert|migrat)[^\\n]*${recoveryIds["FIX-001.1.1"]}\\.json[^\\n]*FIX-001\\.1\\.1\\.json`,
+        "i",
+      ),
+    );
+    expect(recovery.stdout).toMatch(new RegExp(`(?:remove|prune)[^\\n]*${malformedLegacy}\\.json`, "i"));
+
+    for (const malformedStable of [
+      { ...recovered, verdict: "passed" },
+      { ...recovered, requirementId: "FIX-001.1.9" },
+      { ...recovered, hash: "0".repeat(12) },
+      { ...recovered, summary: "" },
+      { ...recovered, timestamp: "not-a-date" },
+    ]) {
+      const semanticRoot = fixture();
+      const semanticIds = reviewIds(semanticRoot);
+      const semanticLegacy = produceRecord(
+        semanticRoot,
+        semanticIds["FIX-001.1.1"],
+        "pass",
+        "valid recovery source",
+      );
+      writeFileSync(stablePath(semanticRoot, "FIX-001.1.1"), `${JSON.stringify(malformedStable)}\n`);
+      writeRecord(semanticRoot, `${semanticIds["FIX-001.1.1"]}.json`, semanticLegacy);
+      const semanticPrune = run(semanticRoot, ["prune"]);
+      expect(semanticPrune.status).toBe(0);
+      expect(JSON.parse(readFileSync(stablePath(semanticRoot, "FIX-001.1.1"), "utf8"))).toEqual(semanticLegacy);
+      expect(semanticPrune.stdout).toMatch(
+        new RegExp(
+          `(?:convert|migrat)[^\\n]*${semanticIds["FIX-001.1.1"]}\\.json[^\\n]*FIX-001\\.1\\.1\\.json`,
+          "i",
+        ),
+      );
+    }
+
+    const invalidLegacyMutations = [
+      (valid: ReturnType<typeof produceRecord>) => ({ ...valid, verdict: "passed" }),
+      (valid: ReturnType<typeof produceRecord>) => ({ ...valid, requirementId: "FIX-001.1.9" }),
+      (valid: ReturnType<typeof produceRecord>) => ({ ...valid, hash: "0".repeat(12) }),
+      (valid: ReturnType<typeof produceRecord>) => ({ ...valid, summary: "" }),
+      (valid: ReturnType<typeof produceRecord>) => ({ ...valid, timestamp: "not-a-date" }),
+      (valid: ReturnType<typeof produceRecord>) => ({ ...valid, reviewId: "not-a-review-id" }),
+    ];
+    for (const mutate of invalidLegacyMutations) {
+      const invalidLegacyRoot = fixture();
+      const invalidLegacyId = reviewIds(invalidLegacyRoot)["FIX-001.1.1"];
+      const invalidProduced = produceRecord(invalidLegacyRoot, invalidLegacyId, "pass", "will be malformed");
+      mkdirSync(join(invalidLegacyRoot, ".2119/verdicts"), { recursive: true });
+      const invalidStable = stablePath(invalidLegacyRoot, "FIX-001.1.1");
+      if (existsSync(invalidStable)) unlinkSync(invalidStable);
+      writeFileSync(
+        join(invalidLegacyRoot, ".2119/verdicts", `${invalidLegacyId}.json`),
+        `${JSON.stringify(mutate(invalidProduced))}\n`,
+      );
+      const invalidPrune = run(invalidLegacyRoot, ["prune"]);
+      expect(invalidPrune.status).toBe(0);
+      expect(existsSync(stablePath(invalidLegacyRoot, "FIX-001.1.1"))).toBe(false);
+      expect(readdirSync(join(invalidLegacyRoot, ".2119/verdicts")).filter((name) => name.endsWith(".json"))).toEqual(
+        [],
+      );
+      expect(invalidPrune.stdout).toMatch(
+        new RegExp(`(?:remove|prune)[^\\n]*${invalidLegacyId}\\.json`, "i"),
+      );
+    }
+  }, 15_000);
+
+  // 2119: REQ-012.7.1
+  it("init marks verdict JSON as generated while leaving it eligible for tracking", () => {
+    const root = realpathSync(mkdtempSync(join(tmpdir(), "2119-init-attributes-")));
+    execFileSync("git", ["init", "-q"], { cwd: root });
+    writeFileSync(join(root, ".gitattributes"), "docs/** linguist-documentation=true\n");
+    writeFileSync(join(root, ".gitignore"), "*\n!docs/\n");
+    expect(run(root, ["init"]).status).toBe(0);
+    const attributeLine = ".2119/verdicts/*.json linguist-generated=true";
+    const attributes = readFileSync(join(root, ".gitattributes"), "utf8").split(/\r?\n/);
+    expect(attributes).toContain("docs/** linguist-documentation=true");
+    expect(attributes).toContain(attributeLine);
+
+    mkdirSync(join(root, ".2119/verdicts"), { recursive: true });
+    writeFileSync(join(root, ".2119/verdicts/REQ-001.1.1.json"), "{}\n");
+    expect(runGit(root, ["check-attr", "linguist-generated", "--", ".2119/verdicts/REQ-001.1.1.json"]).stdout).toContain(
+      ".2119/verdicts/REQ-001.1.1.json: linguist-generated: true",
+    );
+    expect(runGit(root, ["check-ignore", ".2119/verdicts/REQ-001.1.1.json"]).status).toBe(1);
+    expect(runGit(root, ["add", ".2119/verdicts/REQ-001.1.1.json"]).status).toBe(0);
+    expect(runGit(root, ["status", "--short", ".2119/verdicts/REQ-001.1.1.json"]).stdout).toContain(
+      "A  .2119/verdicts/REQ-001.1.1.json",
+    );
+    expect(existsSync(join(root, ".2119/verdicts/REQ-001.1.1.json"))).toBe(true);
+  });
+});
+
+function runGit(cwd: string, args: string[]): RunResult {
+  try {
+    return { status: 0, stdout: execFileSync("git", args, { cwd, encoding: "utf8" }), stderr: "" };
+  } catch (err) {
+    const e = err as { status: number; stdout: string; stderr: string };
+    return { status: e.status, stdout: e.stdout ?? "", stderr: e.stderr ?? "" };
+  }
+}
diff --git a/tests/verdict-validation.test.ts b/tests/verdict-validation.test.ts
index 4afedb4..a5c1e80 100644
--- a/tests/verdict-validation.test.ts
+++ b/tests/verdict-validation.test.ts
@@ -81,13 +81,21 @@ describe("verdict record validation (REQ-003.7)", () => {
     run(root, ["pass", reviewId, "--summary", "asserts spin"]);
 
     // Simulate a mangled merge: truncate the committed verdict.
-    const path = join(verdictsDir(root), `${reviewId}.json`);
-    writeFileSync(path, readFileSync(path, "utf8").slice(0, 20));
+    const path = join(verdictsDir(root), "FIX-001.1.1.json");
+    const original = readFileSync(path, "utf8");
+    writeFileSync(path, original.slice(0, 20));
     const r = run(root, ["check"]);
     expect(r.status).toBe(1);
-    expect(r.stderr).toContain(`.2119/verdicts/${reviewId}.json`);
+    expect(r.stderr).toContain(".2119/verdicts/FIX-001.1.1.json");
     expect(r.stderr).toContain("[REQ-003.7.2]");
     expect(r.stderr).toContain("unparseable JSON");
+
+    // Parseable records that fail schema validation are equally loud.
+    writeFileSync(path, JSON.stringify({ ...JSON.parse(original), summary: "" }));
+    const invalid = run(root, ["check"]);
+    expect(invalid.status).toBe(1);
+    expect(invalid.stderr).toContain(".2119/verdicts/FIX-001.1.1.json");
+    expect(invalid.stderr).toContain("summary must be a nonempty string");
   });
 
   // 2119: REQ-003.7.3
@@ -97,7 +105,7 @@ describe("verdict record validation (REQ-003.7)", () => {
     expect(run(root, ["check"]).status).toBe(0);
 
     renameSync(
-      join(verdictsDir(root), `${reviewId}.json`),
+      join(verdictsDir(root), "FIX-001.1.1.json"),
       join(verdictsDir(root), `FIX-001.1.1--${"0".repeat(12)}.json`),
     );
     const r = run(root, ["check"]);