Skip to content

docs: #781 align testing.md and the README with shipped behavior - #790

Merged
jasonssdev merged 1 commit into
mainfrom
docs/781-testing-md-readme-claims
Aug 19, 2026
Merged

docs: #781 align testing.md and the README with shipped behavior#790
jasonssdev merged 1 commit into
mainfrom
docs/781-testing-md-readme-claims

Conversation

@jasonssdev

Copy link
Copy Markdown
Owner

What

Three prose corrections from the 0.2.7 manual walkthrough — in every case the code is right and only the description changes; no behavior is touched.

  1. docs/testing.md §2 — documents the second numbered picker (Installed embedding models:) that init asks after the chat model, including the sticky-model note: changing the embedding model later forces a full corpus re-embed on the next reindex. A tester following the guide hit a prompt the guide said did not exist, on the one setting with a real switching cost.
  2. docs/testing.md §7.2 — the merge/unmerge round-trip check no longer expects an empty whole-tree diff. unmerge deliberately appends its own audit line to bundle/log.md (its plan discloses it), so the instruction now checks the concept documents for byte-parity (git diff … -- ':!bundle/log.md') and expects exactly the one appended line in the log.
  3. README.mdnext does not always recommend an action: the duplicate tier counts the identical-title family only (the duplicates: status and next only see exact-title groups, so near-match backlog reads as clean #593 recall bound, by design), so a near-match-only backlog correctly yields no ranked action and status carries the standing disclosure. Reworded to the guarantee the engine actually makes.

All three claims were verified against the code before rewording: src/openkos/cli/main.py:470/:1470 (embedding picker + sticky note), src/openkos/cli/main.py:10097 (unmerge plan line), src/openkos/cli/next_action.py:748 (recall bound).

Review

Gentle AI lineage review-30b2a7560c44ef1f, docs-only candidate, low risk, zero lenses, approved.

Closes #781

Three places where the prose described a system the code does not have; in
all three the code is right and only the description changes.

- testing.md par.2: init asks a SECOND numbered picker for the embedding
  model, with the sticky-model note (a later change forces a full corpus
  re-embed on the next reindex). The guide documented only the chat picker.
- testing.md par.7.2: the merge and unmerge round-trip diff is not empty --
  unmerge appends its own audit line to bundle/log.md and its plan
  discloses it. Byte-parity is a claim about the concept documents, so the
  check now excludes bundle/log.md from the empty-diff expectation and
  expects the one appended line in it.
- README: 'next always tells you the one thing worth doing' overstated the
  #593 recall bound by design -- the duplicate tier counts the
  identical-title family only, and a near-match-only backlog correctly
  yields no ranked action. Reworded to the honest guarantee.

Closes #781
@jasonssdev
jasonssdev merged commit 1e1ff62 into main Aug 19, 2026
6 checks passed
@jasonssdev
jasonssdev deleted the docs/781-testing-md-readme-claims branch August 19, 2026 12:44
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.

docs: three places where testing.md and the README describe behavior the code does not have

1 participant