Skip to content

dataset(code-review): BC apps negative-feedback FP gold answers (7 entries)#745

Draft
gggdttt wants to merge 5 commits into
mainfrom
bench/tooltip-gold-answer
Draft

dataset(code-review): BC apps negative-feedback FP gold answers (7 entries)#745
gggdttt wants to merge 5 commits into
mainfrom
bench/tooltip-gold-answer

Conversation

@gggdttt

@gggdttt gggdttt commented Jul 20, 2026

Copy link
Copy Markdown
Collaborator

What

Single PR for the BC apps negative-feedback batch-1 gold answers — four code-review synthetic entries in dataset/codereview.jsonl that lock in BCQuality precision guards (PRs #112 / #117) as offline regressions. Every entry is new-files-only with expected_comments: [], so any finding the agent emits on them costs precision.

Entries

instance_id area Guard exercised From FP cluster
synthetic__style-018 style ui/bound-page-field-inherits-source-field-tooltip (#117) PR 9315
synthetic__upgrade-clean-05 upgrade upgrade/unreleased-schema-change-needs-no-upgrade-path (#112) PR 8167
synthetic__perf-clean-01 performance performance/page-effective-filter-may-live-outside-the-diff (#112) PR 8124
synthetic__error-clean-01 error-handling error-handling/page-boolean-triggers-default-to-true (#112) PR 8419
  • style-018 — a bound page field whose tooltip is inherited from the source-table field is flagged as "missing tooltip"; that is a false positive. (Also carries one required tooltip-less finding to check recall.)
  • upgrade-clean-05 — a brand-new, never-released app (app.json 1.0.0.0, no released baseline) adds persisted tables + enum. New schema in an unreleased app has no shipped data to preserve → a migration / data-loss / obsoletion finding is a false positive.
  • perf-clean-01 — a warehouse activities cue field uses DrillDownPageId to a list page whose SourceTableView already filters the result set (Status = Open), matching the cue FlowField. The effective filter lives on the drill-down page, outside the changed hunk → an "unfiltered / full-table drill-down" finding is a false positive.
  • error-clean-01 — a card page's OnInsertRecord/OnModifyRecord triggers do work but omit an explicit exit(true). Page Boolean record triggers return true by default → a "returns false, blocking all inserts / discards all edits" finding is a false positive.

Validation

  • Real loader accepts the file: CodeReviewEntry.load(...) → 104 entries, all four ids present.
  • Each patch git apply --check clean against base 70fd0246a0a4dbc72cb183ca719106722c03be4d.
  • No explanatory comments in the synthetic AL (intent encoded in code shape only).

Tracked in ADO 639657. Supersedes and consolidates the earlier per-guard PRs (#746, #747), which can be closed in favor of this one.

…_style-018)

Encodes the tooltip-inheritance boundary as a code-review gold answer. A page binds three Vendor fields with no inline ToolTip: 'No.' and 'Name' inherit their source-field tooltips and must NOT be flagged (false-positive guard), while the field bound to a same-diff tableextension field ('Sample Note') that itself has no ToolTip has nothing to inherit and MUST be flagged medium. Guards against both over-flagging inherited fields and silently deferring the genuinely-missing case to compiler AA0218, whose severity is per-app configurable and often disabled.
…old answers

Consolidates the BC apps negative-feedback batch-1 gold answers into this single PR (per request: one PR for all gold answers):

synthetic__upgrade-clean-05 (upgrade): brand-new never-released app adding persisted schema needs no upgrade path; a migration/data-loss finding is a FP. From FP cluster PR 8167.

synthetic__perf-clean-01 (performance): cue DrillDownPageId to a page whose SourceTableView filter matches the FlowField; effective filter lives outside the changed hunk, so an 'unfiltered drill-down' finding is a FP. From FP cluster PR 8124.

synthetic__error-clean-01 (error-handling): page OnInsertRecord/OnModifyRecord omit exit(true); Boolean page record triggers default to true, so a 'returns false, blocking inserts' finding is a FP. From FP cluster PR 8419.
@gggdttt gggdttt changed the title dataset(code-review): add tooltip-inheritance gold answer (synthetic__style-018) dataset(code-review): BC apps negative-feedback batch-1 gold answers (4 entries) Jul 21, 2026
3 negative examples matching new BCQuality guards (PR 8290/8584/8617)
@gggdttt gggdttt changed the title dataset(code-review): BC apps negative-feedback batch-1 gold answers (4 entries) dataset(code-review): BC apps negative-feedback FP gold answers (7 entries) Jul 21, 2026
wenjiefan added 2 commits July 21, 2026 10:40
Local engine run flagged the FlowField SIFT gap (AA0232) and missing tooltips (AA0218) as real defects; a negative example must be otherwise-clean so the only tested construct (CalcFields in OnAfterGetCurrRecord) is what the guard suppresses. Re-run confirms 0 findings.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant