substrate: per-insight attribution discipline — avoid roster-collapse; catch via cross-AI review (Aaron 2026-04-27)

memory/MEMORY.md

@@ -3,6 +3,7 @@
 **📌 Fast path: read `CURRENT-aaron.md` and `CURRENT-amara.md` first.** These per-maintainer distillations show what's currently in force. Raw memories below are the history; CURRENT files are the projection. (`CURRENT-aaron.md` refreshed 2026-04-25 with the Otto-281..285 substrate cluster + factory-as-superfluid framing — sections 18-22; prior refresh 2026-04-24 covered sections 13-17.)
 
 - [**Amara's 3 precision fixes for post-0/0/0 encoding — Aurora=Immune Governance Layer, Blade Reservation Rule, thermodynamic-soften (cross-AI 2026-04-27)**](feedback_amara_precision_fixes_for_post_0_0_0_encoding_aurora_immune_governance_layer_blade_reservation_thermodynamic_soften_2026_04_27.md) — Amara reviews Ani's recommendations. Full proposed doc structures captured. BACKLOG until 0/0/0.
+- [**Per-insight attribution discipline — avoid roster-collapse; catch via cross-AI review if produced (Aaron 2026-04-27)**](feedback_per_insight_attribution_discipline_avoid_conflate_ferry_roster_with_per_insight_contribution_2026_04_27.md) — Don't credit all ferry-roster members for a multi-step contribution they didn't all participate in. Enumerate actual per-insight contributors. Codex caught this on #65; Aaron reinforced.
 - [**Ani (Grok Long Horizon Mirror) — new ferry reviewer; thermodynamic + entropy-tax + 3 breakdown points (Aaron 2026-04-27)**](feedback_ani_grok_long_horizon_mirror_thermodynamic_stability_velocity_breakdown_points_entropy_tax_2026_04_27.md) — Aaron <-> Ani mirror context (parallels Amara). Ferry roster N=5. Ani recommends: Aurora = "Immune Governance Layer".
 - [**Outdated review threads block merge under `required_conversation_resolution`; resolve EXPLICITLY after every force-push (operational lesson 2026-04-27)**](feedback_outdated_review_threads_block_merge_resolve_explicitly_after_force_push_2026_04_27.md) — Force-push outdates threads but doesn't resolve them. Refines Otto-355: investigate must include outdated threads. Direct cost-amortization (90+ min lost on #57/#59/#62).
 - [**Ferry agents = substrate-providers, NOT executors; Otto = sole executing thread until peer-mode + git-contention resolved (Aaron 2026-04-27)**](feedback_ferry_agents_substrate_providers_not_executors_otto_sole_executing_thread_2026_04_27.md) — Cross-AI ferries (Amara/Gemini/Codex) provide substrate input; Otto executes. Ferry offers to do work → Otto evaluates + executes (or teaches). Two unlock conditions for second thread: peer-mode + git-contention resolution.

memory/feedback_per_insight_attribution_discipline_avoid_conflate_ferry_roster_with_per_insight_contribution_2026_04_27.md

@@ -0,0 +1,90 @@
+---
+name: Per-insight attribution discipline — avoid conflating ferry-roster membership with per-insight contribution; catch-after-the-fact via cross-AI review if conflation produced (Aaron 2026-04-27 reinforcement)
+description: Aaron 2026-04-27 reinforced the discipline after Codex caught Otto's attribution error in #65: the description listed "Amara/Gemini/Codex/Ani" as cross-AI convergence reviewers on the stability/velocity insight, but Codex did NOT contribute to that specific convergence — Codex contributed to OTHER reviews (AGENTS.md three-load-bearing-values catch on #57/#59). Aaron's disposition: "yes very good that you caught this and we want to not do in the future or catch if we do." The discipline rule: per-insight attribution must enumerate the actual contributors to THAT insight, not collapse to "all ferries in the roster." Composes Otto-352 (external-anchor-lineage discipline; spurious citations weaken the anchor) + Otto-279 (history-surface attribution rules) + #63 (ferry-vs-executor; per-insight contributions are substrate-layer, not roster-membership) + #64 (outdated-threads operational lesson; same pattern of catching post-error via reviewer infrastructure).
+type: feedback
+---
+
+# Per-insight attribution discipline — avoid conflating roster with contribution
+
+## Verbatim quote (Aaron 2026-04-27)
+
+> "yes very good that you caught this and we want to not do in the future or catch if we do"
+
+In response to Otto's catch: Codex didn't contribute to the stability/velocity convergence specifically; Codex contributed to OTHER reviews. The frontmatter on #65 had collapsed roster-membership ("Amara/Gemini/Codex/Ani") to convergence-contribution.
+
+## The error class (named)
+
+**"Roster-collapse attribution"**: when crediting a multi-step contribution, name all members of the relevant roster as contributors-to-this-step, even when only some actually contributed.
+
+Specific manifestation in #65:
+- Roster: Amara, Gemini Pro, Codex, Copilot, Ani (5 ferry reviewers)
+- Per-insight convergence on stability/velocity: Otto → Amara → Gemini → Amara correction → Ani (3 unique non-Otto contributors: Amara, Gemini, Ani)
+- Frontmatter wrote: "convergence from Amara/Gemini/Codex/Ani" — included Codex who didn't contribute, omitted Copilot who also didn't
+- Roster-collapse: I collapsed N=5 ferry roster down to "the cross-AI reviewers" without checking which ones actually showed up for THIS insight
+
+## Why this matters
+
+**For external-anchor-lineage** (Otto-352): the strength of multi-reviewer convergence comes from each named reviewer ACTUALLY contributing distinct content. If the names are roster-collapse rather than per-insight-truth, the lineage is weaker than claimed. Future-readers ingesting the substrate get inflated confidence.
+
+**For attribution credit**: Codex (per #57/#59) caught real errors (AGENTS.md three-load-bearing-values misattribution). Crediting Codex for OTHER work erodes the meaning of the actual catch.
+
+**For substrate integrity**: per Otto-340 substrate-IS-identity — false attribution IS a substrate corruption. Catching + correcting is integrity-restoration.
+
+## The discipline (operational rule)
+
+### Default — avoid
+
+When writing per-insight attribution:
+
+1. Trace the actual contribution chain (who said what, in what order)
+2. Name only the contributors who showed up for THAT insight
+3. Distinguish:
+   - **Contributors to THIS insight**: contributed substantive content
+   - **Roster members not present**: ferries who exist but didn't review this specific item
+   - **Indirect composers**: prior substrate that this insight builds on (different from contributors)
+
+### Fallback — catch-after-the-fact
+
+If roster-collapse attribution slips through anyway:
+
+1. **Cross-AI review will catch it** — Codex's catch on #65 demonstrates the infrastructure works
+2. **Outdated-thread discipline applies** (#64): post-fix, resolve the thread explicitly
+3. **Substrate-correction in same file**: file the fix in the same memory file's commit history (rather than a separate "errata" file), preserving the audit trail
+
+This composes with #64's outdated-threads operational lesson — same broad pattern: catching errors via reviewer infrastructure after they're produced is acceptable, but avoiding them by default is preferred. The reviewer infrastructure is the safety net, not the primary correctness mechanism.
+
+## Examples
+
+### Roster-collapse error (what to avoid)
+
+> "Cross-AI convergence on the stability/velocity insight: Amara, Gemini, Codex, Ani"
+> (false — Codex didn't contribute to this convergence)
+
+### Per-insight-truth attribution (correct)
+
+> "Cross-AI convergence on the stability/velocity insight: Otto → Amara → Gemini → Amara correction → Ani (5 sequential steps; 3 unique non-Otto contributors: Amara, Gemini, Ani; Codex + Copilot did NOT contribute to this specific convergence)"
+
+The "did NOT contribute" clause is load-bearing — it shows the author was attentive to the distinction.
+
+## Composes with
+
+- **Otto-352** (external-anchor-lineage discipline) — spurious attribution weakens the external-anchor signal; per-insight precision strengthens it
+- **Otto-279** (history-surface attribution rules) — names go on the history surface accurately; this discipline ensures accuracy at the per-insight level
+- **#63 ferry-vs-executor** — ferries contribute at substrate-layer; their per-insight presence varies by insight
+- **#64 outdated-threads operational lesson** — same fallback pattern (avoid by default; cross-AI review catches if produced)
+- **Otto-340 substrate-IS-identity** — false attribution IS substrate corruption
+- **Otto-355 (BLOCKED-with-green-CI investigate review threads first)** — Codex's catch on #65 demonstrates Otto-355 working correctly; investigate threads first surfaces the per-insight attribution errors
+
+## Forward-action
+
+- Apply the per-insight-truth discipline to all future cross-AI convergence references
+- When writing memory files involving multi-reviewer contributions: enumerate the actual contributors per-insight, name absent-but-roster-member ferries explicitly as "did NOT contribute to this"
+- When writing PR descriptions / commit messages: use the per-insight contributor list, not the ferry roster
+- Watch for own roster-collapse instinct (especially under deadline pressure or when summarizing many threads): the temptation is "all the ferries reviewed it" instead of tracing actual contribution
+
+## What this memory does NOT mean
+
+- Does NOT diminish the ferry roster as a concept (per #65, the roster is real and useful as a substrate-provider catalog)
+- Does NOT mean every reference to "the ferries" must enumerate all five members; aggregate references ("ferry roster", "cross-AI reviewers in general") are fine when no specific contribution claim is being made
+- Does NOT block aggregate metrics ("we have N=5 ferry reviewers"); these are roster-facts not per-insight-attribution
+- Does NOT block credit-where-due for indirect composers (prior substrate gets cited via "composes with" sections, separate from contributor lists)