core-labs/core
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

docs(adr-0085): content style pass v2 brief — 3sg + plural agreement (#75)

Browse source 
Follow-up brief for the cheaper dev agent.  Scoped tight: ~18 concrete
row-by-row edits across two patterns the v1 pass (PR #73) deferred or
missed.

Pattern A — 3sg present-tense agreement after relative pronouns
  what/who/that/which followed by a bare-form verb where the implied
  subject is singular.  10 candidates identified up front by repo scan:
    - en_core_causation_v1/effect       (with judgment note)
    - en_core_cognition_v1/beginning, creation, definition, evidence,
      light, reason, symbol
    - en_core_meta_v1/example, mind

  Brief explicitly clarifies that bare verbs after modals (`can`,
  `may`, `should`) are CORRECT and must NOT be changed — flagging
  the "who can know and do" case the v1 agent did right.

Pattern B — plural agreement after quantifier / preposition
  between/among/of/two/three/many + count-noun-in-singular.  5
  confident edits + 3 borderline cases with judgment guidance:
    - en_core_attitude_v1/broad
    - en_core_cognition_v1/context, order, style
    - en_core_spatial_v1/between
  Borderline guidance distinguishes count vs mass nouns: `reason`
  in "group of reason" is count (apply fix); `reason` in "because
  of reason" is mass (leave alone).

Same hard rules as v1 brief:
  - no code edits
  - definitional_atoms / predicates_invited / pos / lemma /
    definition_version must not change
  - Greek/Hebrew packs and primitives pack untouched
  - closure verifier must exit 0
  - cognition eval must stay byte-identical
  - draft PR, human review before merge

Estimated effort: ~18 one-character edits.  Whole pass should take an
order of magnitude less time than v1 because the candidate rows are
enumerated in the brief itself rather than discovered via heuristic
scan.

Why a v2 brief rather than amending v1's PR:
  Plural patterns were not on v1's explicit pattern list (the v1
  brief only named verb agreement, missing articles, missing
  infinitives, missing copulas).  3sg agreement was named but
  required a different tool than the agent had at hand.  Scoping
  v2 to exactly the rows known to need fixing is cheaper than
  re-running the v1 heuristic scan with an expanded ruleset.

Provenance tag for the v2 pass:
  adr-0085-style-v2:reviewed:2026-05-22
This commit is contained in:
Shay 2026-05-20 16:23:14 -07:00 committed by GitHub
parent 0cf54a009d
commit 886fc4553f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 143 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

143
docs/handoff/ADR-0085-content-style-pass-v2-brief.md Normal file
View file

@ -0,0 +1,143 @@
# ADR-0085 Content Style Pass v2 — Brief
**Audience:** Same cheaper dev agent that did PR #73 (v1 fluency pass). You can re-read [ADR-0085-content-style-pass-brief.md](./ADR-0085-content-style-pass-brief.md) for context but most of it still applies — this v2 narrows scope to the two patterns v1 deferred or missed.
**Mission:** Apply targeted 3sg-agreement and plural-after-quantifier fixes to the existing gloss entries. ~18 candidate entries identified up front. No other patterns in scope.
**Estimated effort:** very small — ~18 one-character edits.
---
## What v1 covered (already on main as PR #73)
- Article insertion (`Light is a visible medium...` ← `Light is visible medium...`)
- Infinitive insertion (`To recall means to get memory...` ← `To recall means get memory...`)
- Copula insertion (`Something is bad when it is different...` ← `... when it different`)
All shipped. Closure verifier exits 0, cognition eval byte-identical. The composer surfaces now read well except for two remaining patterns.
## What v2 covers
### Pattern A — 3sg present-tense agreement after relative pronouns
Glosses with `what {VERB}`, `who {VERB}`, `that {VERB}`, or `which {VERB}` where `{VERB}` is in bare form and the implied subject is singular. The verb needs an `s`.
| pack | lemma | before | after |
|---|---|---|---|
| en_core_causation_v1 | effect | `"what cause make"` | `"what causes make"` (and `make` → `makes` too — see note) |
| en_core_cognition_v1 | beginning | `"start before what follow"` | `"start before what follows"` |
| en_core_cognition_v1 | creation | `"start of what make"` | `"start of what makes"` |
| en_core_cognition_v1 | definition | `"word that explain meaning"` | `"word that explains meaning"` |
| en_core_cognition_v1 | evidence | `"what support truth"` | `"what supports truth"` |
| en_core_cognition_v1 | light | `"a visible medium that reveal truth"` | `"a visible medium that reveals truth"` |
| en_core_cognition_v1 | reason | `"what explain why"` | `"what explains why"` |
| en_core_cognition_v1 | symbol | `"image that mean more"` | `"image that means more"` |
| en_core_meta_v1 | example | `"one case that make a point visible"` | `"one case that makes a point visible"` |
| en_core_meta_v1 | mind | `"part that think and know"` | `"part that thinks and knows"` |
**Note on `effect`** — both verbs need the `s`: `"what causes makes"` reads wrong (looks like two finite verbs). The actual fix is `"what a cause makes"` (insert article) — that converts the second verb to make the gloss grammatical. Worth your judgment: either rewrite OR accept the slight awkwardness OR leave `effect` for a future pass.
**Modal verbs are NOT 3sg.** Glosses like `"who can know and do"` (en_core_cognition_v1/person) are CORRECT — `can` is a modal, the bare verbs after it (`know`, `do`) take no `s`. Do NOT change those.
### Pattern B — Plural agreement after quantifier / preposition
Glosses with `between {N}`, `among {N}`, `of {N}`, `two {N}`, `three {N}`, `many {N}`, etc., where `{N}` is a count noun in singular form. The noun needs an `s`.
| pack | lemma | before | after |
|---|---|---|---|
| en_core_attitude_v1 | broad | `"is general in many part"` | `"is general in many parts"` |
| en_core_cognition_v1 | context | `"relation between word"` | `"relation between words"` |
| en_core_cognition_v1 | order | `"relation of part to part"` | `"relation of parts to parts"` |
| en_core_cognition_v1 | style | `"way of voice and word"` | `"way of voice and words"` (note: `voice` is mass, leave it) |
| en_core_spatial_v1 | between | `"in relation to two place"` | `"in relation to two places"` |
**Three borderline cases** — apply your judgment, default to leave-alone if uncertain:
| pack | lemma | gloss | judgment |
|---|---|---|---|
| en_core_attitude_v1 | factual | `"is of fact and truth"` | Both `fact` and `truth` are mass-noun-ish in this context. **Probably leave.** |
| en_core_cognition_v1 | therefore | `"then because of reason"` | `reason` here is mass-noun-ish ("reasoning"). **Probably leave.** |
| en_core_meta_v1 | argument | `"group of reason for a point"` | `reason` here is count ("group of reasons"). **Apply fix → `"group of reasons for a point"`.** |
---
## What v2 does NOT do
- No edits beyond the patterns above. Other awkward surfaces (if any) are out of scope.
- No edits to `definitional_atoms`, `predicates_invited`, `pos`, `lemma`, `definition_version` — same rules as v1.
- No edits to Greek/Hebrew packs.
- No edits to the primitives pack.
- No new fluency patterns invented in this pass.
## Phase 0 — Baseline (mandatory)
```bash
PYTHONPATH=. python3 -m core.cli eval cognition > baseline.json
```
## Phase 1 — Apply Pattern A edits (10 entries)
For each row in the Pattern A table above, edit the gloss in place. Append `"adr-0085-style-v2:reviewed:2026-05-22"` to `provenance_ids`.
## Phase 2 — Apply Pattern B edits (5 confident + 1 borderline-accepted entry = 6 total)
For each row in the Pattern B table where the column says "→ ...", edit the gloss in place. Append the same provenance tag.
## Phase 3 — Refresh checksums
Same per-pack checksum-refresh shell snippet as v1 (in [ADR-0085-content-style-pass-brief.md](./ADR-0085-content-style-pass-brief.md) Phase 3).
## Phase 4 — Verify
All four must pass:
```bash
PYTHONPATH=. python3 scripts/verify_definitional_closure.py
# Expected: 0 unresolved / 0 mismatches
PYTHONPATH=. python3 -m pytest tests/test_adr_0084_integration_closure.py -q
# Expected: 30 passed
PYTHONPATH=. python3 -m core.cli eval cognition > /tmp/eval_post.json
diff /tmp/eval_post.json baseline.json
# Expected: empty diff
PYTHONPATH=. python3 -m core.cli test --suite smoke -q
PYTHONPATH=. python3 -m core.cli test --suite packs -q
```
If `core eval cognition` drifts byte-from-byte from baseline.json, **STOP** and roll back the edit that caused it.
---
## Why this v2 exists separately from a single bigger v1
v1 covered the patterns that could be reliably detected via simple substring search (missing articles, missing infinitives, missing copulas). 3sg-agreement and plural-after-quantifier need a small POS-aware classifier that wasn't worth building for the v1 scope. v2 ships the *concrete row-by-row edit list* up front so no classifier is required — you just apply the table.
---
## Deliverables
1. Edits on branch `feat/adr-0085-content-style-pass-v2`.
2. Draft PR titled `feat(packs): ADR-0085 content style pass v2 (3sg + plural agreement)` with this report:
```
## v2 fluency summary
- pattern A (3sg): N of 10 applied, M skipped (list which + why)
- pattern B (plural): N of 5+1 applied, M skipped
- closure verifier: exit 0
- integration test: 30/30
- cognition eval: byte-identical to baseline
- packs lane: pass
## sample 10 surfaces post-v2
(framed via the composer's POS table — same format as v1's report)
```
3. Mark **Draft**. Human reviews before merge.
## Hard rules
Same as v1. Do not touch code. Do not change atoms / predicates / pos / lemma / definition_version. Cognition eval must stay byte-identical. Closure verifier must exit 0.