docs(phase5): revise M6/M6.5/M7 plan per May 2026 literature scan

[chrisjz]

Summary

Plan-only PR (no code, no tests). Revises the remaining Phase 5 milestones (M6 Transgenerational Memory, M6.5 NEAT Architecture Evolution, M7 Phase 5 Synthesis Logbook) in the OpenSpec tracker + corresponding roadmap rows, based on a May 2026 literature scan run ahead of starting M6 implementation in the next PR.

2 files, 40 insertions / 24 deletions, 1 commit.

What changed and why

M6 — Transgenerational Memory (substantive)

• Status flipped pending → ready to start (all gates cleared: M3 GO ✅, M4 closed STOP ✅, M5 closed STOP ✅).
• Decision gate corrected. The original (F1 ≥ 40%, F3 ≤ 10%) would have marked a successful biological replication as STOP — Kaletsky et al. (Murphy lab, eLife May 2025) report F2 choice indices in the 0.5-0.6 range, and the Vidal-Gadea independent validation (eLife Nov 2025) confirms F1+F2 retention well above the 10% floor. Revised gate: GO if F1 ≥40%, F2 ≥25%, F3 ≥15%, with monotone non-increasing decay F0 ≥ F1 ≥ F2 ≥ F3.
• Literature controversy note added — Posner 2023 is actively contested but not retracted (Hunter lab critique + Murphy rebuttal + Vidal-Gadea validation all cited). Current consensus: "real but fragile." This motivates rigorous methodology rather than novelty.
• New sub-task M6.5b (TEI-on vs TEI-off lineage comparison) added per the bioRxiv Apr 2026 cross-species finding that TEI is itself a species-conditional trait. Old M6.5 sub-task renamed M6.5a to disambiguate from milestone M6.5.

M6.5 — NEAT Architecture Evolution (reframed)

• Reframed from generic NEAT-vs-PPO ablation to a direct test of M5's architecture-asymmetry hypothesis.
• Independent corroboration cited: Resendez Prado (arXiv 2604.03565, Apr 2026) identifies a "transparent regime" (same-architecture self-play) that suppresses Baldwin/Red Queen signal — same hypothesis as our M5 STOP diagnosis from a different research group.
• TensorNEAT now mature (ACM TELO 2025 final) — ~500× speedup vs neat-python; compute concern materially weakened.
• New verdict gate: reuse M5's lag-matrix instrument (logbook 017) — GO if matched-capacity arm produces own-vs-cross lag delta ≤ −0.05 while asymmetric arm reproduces M5's +0.017.
• OPTIONAL marker retained with clarifying parenthetical pointing at the Phase 6 quantum re-evaluation deferral path (M7.4).

M7 — Phase 5 Synthesis Logbook (citation augmentation)

Added synthesis-scope paragraph + three independent 2025/2026 corroborating results to cite when writing the Phase 5 synthesis:

• Resendez Prado 2026 (chess Baldwin transparent regime)
• Mougi 2026 (Sci Reports — dual-trait predator-prey decoupling)
• Chen 2025 (arXiv 2512.15732 — Red Queen's Trap in HFT)

Review findings (caught + fixed pre-commit)

Spawned a review agent before committing. Caught 3 blocking issues, all genuine, all verified independently against primary sources:

1. "Sharma et al." was wrong — actual author of arXiv 2604.03565 is Diego Armando Resendez Prado (sole author). Verified via WebFetch of arXiv abstract. Renamed at 5 sites.
2. "Sun et al." was wrong — actual first author of eLife 105673 is Rachel Kaletsky (Murphy lab). Verified via WebFetch + Web search. Renamed at 4 sites.
3. Quantitative claim "F1-F4 = 0.2-0.4 of F0" not directly supported by the eLife abstract. Replaced with the verified Kaletsky F2 ~0.5-0.6 PA14 anchor + a note that thresholds are intentionally permissive given documented protocol-sensitivity.

Also addressed should-fix items: dropped unverifiable "GECCO 2025 Best Paper" framing; renamed M6.5 sub-task → M6.5a to resolve namespace collision with milestone M6.5; clarified OPTIONAL vs M5-closure tension.

Test plan

• uv run openspec validate phase5-tracking --strict passes
• uv run pre-commit run -a all hooks green (mdformat, markdownlint, ruff check + format, pyright, tests)
• All cited URLs verified to resolve (sample of 4-5 most load-bearing checked via WebFetch + Web search; remaining checked via Web search title-match)
• All author attributions verified against primary sources

What's deferred

M6 implementation (agent/transgenerational_memory.py + configs + F0-F3 experiment + logbook 018) ships in a follow-up PR. This PR is plan-only to keep the scope tight.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Documentation
  • Updated Phase 5 milestone M6 (Transgenerational Memory) status to "ready to start" with revised decision gates and refined success criteria for F1/F2/F3 retention
  • Reframed M6.5 to directly test the M5 architecture-asymmetry hypothesis using matched vs asymmetric capacity evaluation
  • Enhanced M7 synthesis scope with Phase 5 cross-linking and detailed methodology contributions checklist

[coderabbitai]

📝 Walkthrough

Walkthrough

Phase 5 milestone planning is updated to advance M6 (Transgenerational Memory) from "not started" to "ready to start" with revised F1/F2/F3 TEI retention decision gates. M6.5 is reframed as a hypothesis closure test of M5's architecture-asymmetry claim. M7 synthesis scope is enhanced to integrate Phase 5 STOP findings.

Changes

Phase 5 Milestone Refinement

Layer / File(s)	Summary
M6 status and decision gate definition docs/roadmap.md, openspec/changes/phase5-tracking/tasks.md	M6 status changed to "ready to start" with explicit GO/STOP decision gate: F1/F2/F3 retention thresholds and monotone non-increasing decay criterion, replacing prior F1/F3-only framing.
M6 narrative context and literature grounding openspec/changes/phase5-tracking/tasks.md	Added May 2026 literature controversy note; expanded protocol-sensitivity explanation and substrate readiness framing with species-conditional variability sources and stronger validation evidence statements.
M6 experimental checklist and task breakdown openspec/changes/phase5-tracking/tasks.md	Clarified M6.5a replication steps with revised gate framing and added M6.5b TEI-on vs TEI-off lineage comparison to isolate inheritance-substrate effects from residual policy learning.
M6.5 reframing as M5 architecture-asymmetry hypothesis closure docs/roadmap.md, openspec/changes/phase5-tracking/tasks.md	Reframed M6.5 from generic NEAT planning into a hypothesis-driven test of M5's architecture-asymmetry claim, defining matched-capacity vs asymmetric NEAT head-to-head evaluation using M5 lag-matrix instrumentation.
M7 synthesis scope and negative findings integration openspec/changes/phase5-tracking/tasks.md	Enhanced M7 with "Synthesis scope" narrative cross-linking Phase 5 STOP findings and methodology contributions; refined checklist for synthesizing negative findings and Phase 6 trigger recommendation.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~4 minutes

Possibly related PRs

• SyntheticBrains/nematode#131: Introduced Phase 5 milestone tracker scaffolding that is now revised and extended by this PR's M6/M6.5/M7 updates.
• SyntheticBrains/nematode#143: Updated M5 instrumentation framing (Baldwin secondary readouts and lag-matrix) that is now leveraged downstream in this PR's M6.5 hypothesis-closure design.
• SyntheticBrains/nematode#20: Created initial docs/roadmap.md content; this PR refines and expands the M6/M6.5 milestone entries.

Poem

🐰 A roadmap rewoven with gates drawn anew,
M6 springs forth when M5's tale rings true.
TEI retention and NEAT's matched dance,
Phase 5 findings in synthesis advance! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and specifically summarizes the main change: revising three Phase 5 milestones (M6/M6.5/M7) based on a May 2026 literature scan, which directly matches the documented changes to both the roadmap and task tracking files.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/m6-plan-revision

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

docs/roadmap.md (1)

511-511: 💤 Low value

Verify readability of extended table cell prose.

Line 511 contains substantial narrative (~300 words including literature controversy note, decision gate correction, cross-species variability, and substrate readiness) embedded within a Markdown table cell. While the content is substantive and well-justified, extremely long table cells can reduce readability and make diff review difficult.

Consider extracting the detailed literature context to a separate section below the milestone tracker table, leaving only a brief status summary in the table cell itself (e.g., "ready to start — May 2026 literature scan cleared controversy; see detailed context below").

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/roadmap.md` at line 511, The M6 table cell contains an overly long
narrative; extract the extended literature/context prose out of the table and
move it to a new subsection below the milestone tracker (e.g., "M6 — Detailed
Literature & Decision Gate Rationale"), leaving the table row for M6 (the "M6 |
Transgenerational memory (gated on M3 ✅ + M4/M5 ✅) | HIGH | …") with a short
summary like "ready to start — May 2026 literature scan cleared controversy; see
detailed context below"; update any internal links or anchors so references to
M6 point to the new subsection and keep the original decision-gate numbers
(F1/F2/F3 thresholds and substrate notes) in the moved detailed text.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@docs/roadmap.md`:
- Line 511: The M6 table cell contains an overly long narrative; extract the
extended literature/context prose out of the table and move it to a new
subsection below the milestone tracker (e.g., "M6 — Detailed Literature &
Decision Gate Rationale"), leaving the table row for M6 (the "M6 |
Transgenerational memory (gated on M3 ✅ + M4/M5 ✅) | HIGH | …") with a short
summary like "ready to start — May 2026 literature scan cleared controversy; see
detailed context below"; update any internal links or anchors so references to
M6 point to the new subsection and keep the original decision-gate numbers
(F1/F2/F3 thresholds and substrate notes) in the moved detailed text.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 097cf996-b3a9-4ad1-b15c-e6a3b3e179b0

📥 Commits

Reviewing files that changed from the base of the PR and between b10e5d1 and 33670a9.

📒 Files selected for processing (2)

• docs/roadmap.md
• openspec/changes/phase5-tracking/tasks.md