drain(#226 follow-up 2): CC log schema + populated-state alignment

docs/research/memory-reconciliation-algorithm-design-2026-04-24.md

@@ -260,22 +260,61 @@ function reconcile(facts):
 
 ### Conflict outputs
 
-Each conflict becomes a row in `docs/CONTRIBUTOR-CONFLICTS.md`
-(the file Amara's 4th ferry noted is present-with-schema-but-
-unpopulated; this design starts populating it via the
-generator).
-Row format:
+Each conflict becomes a row in `docs/CONTRIBUTOR-CONFLICTS.md`,
+which already has a populated schema (CC-001..CC-003 as of
+2026-04-23 cover the no-name-attribution-rule scope, the
+Stabilize-vs-keep-opening-frames disagreement, and the
+absent-artifact-citation discipline). This design extends
+that schema by inserting machine-generated rows from the
+reconciliation pass into the **Open** table (or into a
+dedicated autogenerated subsection within **Open** if that
+convention is later adopted) — one CC-### row per detected
+conflict, using the existing column schema:
 
 ```markdown
-### CONF-<YYYY-MM-DD>-<NNN>: <subject> / <predicate>
-- **Canonical key:** `<subject>::<predicate>::<normalized-object>`
-- **Conflicting facts:** [MF-..., MF-...]
-- **Winner (priority tiebreak):** MF-...
-- **Reason:** invariant-2 violation | broken chain | explicit disagreement
-- **Resolution:** pending | explicit-preference-recorded | escalated
-- **Resolution evidence:** <DP-NNN.yaml ref if proxy-reviewed>
+| CC-<NNN> | <YYYY-MM-DD> | <Question — invariant-2 violation
+on canonical key `<subject>::<predicate>::<normalized-object>`
+between facts MF-..., MF-...> | <Between: source agents /
+humans whose memory rows produced the conflicting facts> |
+<Positions: each fact's claim text + source_path> |
+<Resolution-so-far: pending until explicit preference
+recorded; or "auto: priority tie-break MF-... wins" when
+invariant-6 fallback applies> | <Scope: invariant-2 violation
+\| broken chain \| explicit disagreement> | <Source:
+source_path links to the contributing memory files;
+DP-NNN.yaml ref if proxy-reviewed> |
 ```
 
+The CC-### counter continues from the highest existing ID
+(e.g., next machine-generated conflict starts at CC-004).
+
+**Idempotent generator strategy.** Repeated CI runs MUST
+NOT re-append the same conflict, otherwise the **Open**
+table grows unboundedly. The generator maintains an
+explicit mapping from canonical key
+(`<subject>::<predicate>::<normalized-object>` plus the
+sorted set of contributing MF-IDs) → CC-NNN. On each run:
+
+- If the canonical key already maps to an existing CC-NNN
+  row in the **Open** table, update that row in-place
+  (refresh "Between"/"Positions" if source paths shifted;
+  preserve any human-edited "Resolution-so-far").
+- If the conflict has been moved to **Resolved** or
+  **Stale**, leave it alone — those tables are out of the
+  generator's write scope.
+- If no existing row maps, allocate the next CC-NNN and
+  insert a new row.
+
+Generator MUST preserve all existing manually-curated rows
+verbatim — both in **Open** (treat human-edited rows as
+read-only) and in **Resolved** / **Stale**. Auto-detected
+rows update in-place when matched; new rows are inserted
+at the bottom of **Open**. As an alternative convention,
+implementers MAY reserve a clearly delimited
+"<!-- autogenerated -->" subsection within **Open** that
+is fully rewritten on each run, leaving manually curated
+rows untouched outside the delimited block.
+
 Conflicts block the `CURRENT-*.md` generation if unresolved
 — this is the "explicit-not-silent" discipline Amara
 emphasized. A CI run that discovers unresolved conflicts