Repository navigation
Edits polish: residuals from PR #11's adversarial review #12
Description
Discovered from #5 / PR #11 (three review rounds). None corrupt silently in a plausible flow; all graded ride-along. Collected so they are not re-derived:
- Optional span ids (additive
SourceSpan.id) — the real cure for the one residual mis-identification case: a locked span whose start moves while a same-(source, role)sibling assumes its exact old start gets its value written onto the impostor. Content-anchored matching cannot beat identity-swapping content; ids can. Fits the Evidence layer: persist analysis into a lacing store #4 evidence-layer timeline (store-side identity adoption wants them anyway). - Merge after an id rename resurrects the old id: renaming a step severs the merge key, so a later regeneration re-introduces the old id alongside the renamed step (visible, not silent). Document, or map renames via the lock record (
/idlock'swasis exactly the old→new mapping). Lock.at/Lock.byaccept any string (emptyby, junk timestamps) — validate.Lock.wasis an unconstrained wire channel: a float set via attrs lands as a float literal in an otherwise no-floats document. Same class as the pre-existingattrs/confidenceholes; consider a document-wide float scan invalidate_document.Lock.pathserialises snake_case into a camelCase wire document (input side accepts both; output is snake-only). A wire consumer resolving lock paths against the wire needs the camel spelling or a documented rule.- Renaming a step id onto an existing id is accepted and only flagged by
validate_document— consider refusing at edit time. - Selection boost fires on fact presence, not usability: boundaries + N names + N matching chapters picks
explicit(unnamed spans + mismatch flag) where chapters would have named everything. Honestly flagged today; revisit scoring when a fourth segmenter lands. - One-time
"locks": []re-serialisation churn on pre-0.0.3 documents (accepted).
Activity
thorwhalen commentedon Sep 8, 2026
Landed 4 of 7 in #13 (or the PR opened from branch fix/issue-12-edit-protection-residuals — see PR link below): Lock.by/Lock.at validation, document-wide float scan in validate_document, refuse renaming a step id onto an existing id, and merge_regenerated no longer resurrects the pre-rename id.
Declining the remaining 3, each with a one-line reason (details in docs/07 §6.6/§7):
SourceSpan.idfield on the wire); out of scope for this residuals pass. Best paired with Evidence layer: persist analysis into a lacing store #4's evidence layer, where store-side identity adoption wants ids anyway.Lock.pathstays snake_case (matching whatapply_editsalready writes to every existing document); a wire consumer converts each field segment withto_camelbefore indexing. Changing the recorded spelling would itself be an on-disk format change for every already-written lock — out of scope here.Fix 4 of 7 edit-protection residuals from PR #11's review (#22)
thorwhalen commentedon Sep 22, 2026
Closing: all 7 residuals from PR #11's review are accounted for — 4 fixed and merged in PR #22 (b310851), and the remaining 3 (optional span ids, Lock.path casing, selection-boost scoring) explicitly declined with reasons in the comment above. Nothing actionable remains here; optional span ids are noted as best paired with #4's evidence layer if picked up later.