Add vero.interpret: analyse what optimizer agents changed - #86

Open

varunursekar wants to merge 16 commits into

mainfrom

worktree-interpret-module

Commits on Aug 3, 2026

Add interpret module: harbor adapter and deterministic edit decomposition

varunursekar
and
claude
committed
9df50ea
Add interpret CLI: extract, edits, symbols

varunursekar
and
claude
committed
39e1a5f
Add interpret taxonomy derived from the corpus, not chosen in advance

varunursekar
and
claude
committed
b6976e3
Fix two silent edit-dropping bugs found by auditing zero-edit candidates

varunursekar
and
claude
committed
7901b56
Add labelling: async client, facet labeler, resumable cache

varunursekar
and
claude
committed
9fd0e32
Add analysis: prevalence, rarefaction, Jaccard-vs-null, and HTML figures

varunursekar
and
claude
committed
464b036
Derive fix provenance from the seed tree instead of asking the model

varunursekar
and
claude
committed
b68ea95
Put method, vocabulary and the error inventory above the figures

varunursekar
and
claude
committed
0b951a7
Drop the audit-rate sampling: the audit was already complete

varunursekar
and
claude
committed
95208a4
Rubrics, surrounding code, history, and a 5x larger diff budget

varunursekar
and
claude
committed
743cc2e
Add print figures in the Scale brand style, vendored for reproducibility

varunursekar
and
claude
committed
a9891a3
Add figure specs: semantics in the repo, styling in the style module

varunursekar
and
claude
committed
7d7d653
One directory per figure

varunursekar
and
claude
committed
edb6be2
Title-case display names, kept out of the taxonomy

varunursekar
and
claude
committed
a10136b
Add ordering spread to the rarefaction figure, as small multiples

varunursekar
and
claude
committed
2789fd3
Neutralise brand references in the vendored style module, and lock the extra

varunursekar
and
claude
committed
49d44fb

Navigation Menu

Uh oh!

Add vero.interpret: analyse what optimizer agents changed - #86

Add vero.interpret: analyse what optimizer agents changed#86
varunursekar wants to merge 16 commits into
mainfrom
worktree-interpret-module

Commits on Aug 3, 2026

Add interpret module: harbor adapter and deterministic edit decomposition

Add interpret CLI: extract, edits, symbols

Add interpret taxonomy derived from the corpus, not chosen in advance

Fix two silent edit-dropping bugs found by auditing zero-edit candidates

Add labelling: async client, facet labeler, resumable cache

Add analysis: prevalence, rarefaction, Jaccard-vs-null, and HTML figures

Derive fix provenance from the seed tree instead of asking the model

Put method, vocabulary and the error inventory above the figures

Drop the audit-rate sampling: the audit was already complete

Rubrics, surrounding code, history, and a 5x larger diff budget

Add print figures in the Scale brand style, vendored for reproducibility

Add figure specs: semantics in the repo, styling in the style module

One directory per figure

Title-case display names, kept out of the taxonomy

Add ordering spread to the rarefaction figure, as small multiples

Neutralise brand references in the vendored style module, and lock the extra