Connectome-constrained PPO brain + Phase 6 Gate 1 close - #183
Merged
Conversation
…brain
Introduces ConnectomePPOBrain — a PPO brain whose forward-pass topology
is the Cook 2019 *C. elegans* hermaphrodite connectome:
- Chemical synapses are strict-masked: only the wild-type adjacency
(3709 edges) carries learnable scalar weights; PPO updates are
projected back onto that adjacency after every optimiser step.
- Gap junctions carry fixed Cook 2019 synapse counts, fan-in normalised
symmetrically (G[i,j] / sqrt(d_i * d_j)) so the bidirectional physics
G[a,b] == G[b,a] is preserved across training.
- Sensor projection: food-chemotaxis [strength, angle] features are
injected onto ASEL/ASER/AWCL/AWCR/AWAL/AWAR sensory neurons via a
learnable 2-by-6 gain matrix (the canonical Bargmann-lab klinotaxis
pathway).
- Motor readout: VB/DB/VA/DA motor-class activations are mean-pooled
into a 4-vector, then projected to the 4 DEFAULT_ACTIONS via a
learnable 4-by-4 readout matrix.
- Forward-pass depth K defaults to 4 (the canonical klinotaxis pathway
depth sensory → primary-interneuron → command-interneuron → motor;
K=1 produces a degenerate output because the food signal cannot
propagate to motor neurons in one chemical-synapse hop).
- ``freeze_updates: true`` flag drives the paired-control branch of the
upcoming training-signal check: with it set, PPO step() is a no-op
and weights stay byte-identical to construction.
- Spec correction: tightened the gap-junction fan-in normalisation
scenario in specs/connectome-ppo-brain/spec.md to require symmetric
scaling (the original row-only scaling was inconsistent with the
symmetry requirement asserted by the adjacent scenario); raised the
default forward_pass_depth from 1 to 4 with explanatory note (K=1
produced degenerate variance).
Registration:
- @register_brain("connectomeppo", ...) on the brain class.
- BrainType.CONNECTOMEPPO added to the enum + BRAIN_TYPES Literal.
- ConnectomePPOBrain / ConnectomePPOBrainConfig re-exported from
brain.arch.__init__.
- brain_factory._build_infra_kwargs gains a CONNECTOMEPPO branch
(forwards only ``device``; the action count is fixed by the
topology's 4-by-4 readout, not configurable via ``num_actions``).
- Registry-vs-enum consistency check at import time now passes with
20 registered brains.
Tests (25 new in test_connectome_ppo.py):
- Construction loads the connectome and builds W_chem / M_chem / G_gap.
- Strict-mask invariant holds at init and after every PPO update.
- Gap junctions remain symmetric, non-learnable, byte-identical across
training.
- Sensor projection routes to the canonical 6 sensory neurons.
- Motor readout aggregates all 4 motor classes (each non-empty).
- Forward pass is finite and has non-degenerate variance (mean variance
~1.6e-3 across 100 random food inputs).
- K is configurable; K=2 differs from K=4 outputs; K=0 raises.
- Soft-prior mode is a no-op for apply_weight_mask; strict mode zeros
~M_chem entries.
- freeze_updates skips the optimiser; the frozen brain still samples.
- Registry-instantiation produces a functionally identical brain.
- Topology Protocol learnable_parameters excludes G_gap.
Lint: added D101/D102 to the project-wide test-file ignore list (test
classes are pure containers + test method names are already descriptive;
matches the existing relaxed-docstring conventions for tests).
Verified: openspec validate --strict clean. Pre-commit clean across all
changed files (mdformat, markdownlint, ruff, pyright, tests). 226
affected tests pass + 1 intentional skip + 2 pre-existing xfails.
Runtime sanity-check confirms strict-mask invariant, gap-junction
symmetry, finite forward pass, and non-degenerate variance.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…state guard Three pre-evaluation review items applied: I2 (performance): cache the post-K hidden state during the topology forward pass. ConnectomeTopology now exposes a forward_with_hidden(x) method that returns both (logits, h_after_K) in a single sweep through the 302x302 chemical + gap-junction connectivity. run_brain and _perform_ppo_update both consume it via forward_with_hidden, eliminating the second under-mask connectome traversal that previously ran inside the torch.no_grad() block. The forward() shim still returns logits only for callers that don't need the hidden state. run_brain wall-clock measured at 0.17 ms/call after refactor (~3.4 seconds of action-sampling overhead at 20k calls / 100-episode rollout). I1 (dead code): remove the unused `self.pending_reward = reward` assignment from run_brain. The attribute was set but never read elsewhere, never declared in __init__, and not consumed by the PPO update. The `reward` parameter on run_brain is now correctly marked unused (ARG002) since rewards are delivered via learn(). I3 (silent fallback): tighten the learn() pending-state guard. The previous `action=self._pending_action or 0` would silently substitute FORWARD (action_idx=0) if learn() was called before any run_brain(). Replaced with an explicit RuntimeError that surfaces the out-of-order use with a descriptive message. The outer `_pending_state is not None` guard already prevents this in normal env-loop usage; the inner check is defence-in-depth. Verified: 25/25 connectome unit tests pass; pre-commit clean (ruff + ruff-format + pyright + tests). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ConnectomePPOBrainConfig gains a sensing_mode field (oracle | klinotaxis). In klinotaxis mode the brain consumes [food_concentration, food_lateral_gradient, food_dconcentration_dt] (3 features) instead of the oracle [strength, angle] pair (2), matching the env-side klinotaxis sensory-module emission shape without re-implementing head-sweep computation in the brain. ConnectomeTopology.food_gains is sized to the per-mode feature count, and preprocess() switches on the configured mode with tanh-clipped scaling matching the env-side klinotaxis sensor. ConnectomePPOBrainConfig is also added to the BrainConfigType union in config_loader so Pydantic smart-union resolves YAML files that set the connectome-specific fields correctly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…eline runs Six configs land together for the connectome-PPO klinotaxis paired-control evaluation: - connectome_ppo_oracle.yml + _frozen_control.yml — initial oracle-mode sanity baseline before the klinotaxis adapter (kept for reproducibility of the early-baseline run). - connectome_ppo_klinotaxis.yml + _frozen_control.yml — main klinotaxis paired-control configs (sensing_mode=klinotaxis, STAM enabled, chemotaxis_mode=klinotaxis). - connectome_ppo_klinotaxis_low_entropy.yml — diagnostic config with entropy_coef cut to 0.005 (vs 0.02). Used to verify that late-training drift seen at the higher entropy was an exploration-annealing artefact, not architectural. - mlpppo_small_klinotaxis.yml — inferred MLPPPO baseline at the klinotaxis env scale: combines the established mlpppo_small_oracle PPO hyperparameters with the lstmppo_small_klinotaxis env/sensing settings + proprioception (heading needed for memoryless MLP to correlate dC/dt with movement). All configs share seed 2026 and the same env layout for apples-to-apples comparison. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New docs/architecture/plugin-developer-guide.md walks through the files-touched budget (≤ 6) and the step-by-step pattern for adding a new brain architecture: module + decorator, BrainType enum member, __init__ import, BrainConfigType union, optional infra-kwargs branch, tests. Worked example uses a hypothetical TinyMLPBrain. Cross-linked from: - docs/roadmap.md L1 row in the layered-platform table - packages/quantum-nematode/quantumnematode/brain/arch/__init__.py module docstring - openspec/config.yaml context (bumped 19 → 20 brains and added the registry/guide pointer) - AGENTS.md Key Directories (same bump and pointer) The OpenSpec change tasks.md sections 8 and 9 are ticked. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Logbook 023 records the architecture-plugin-interface implementation outcome and the first connectome-constrained PPO brain result. Gate 1 closes GO across all four sub-criteria: - G1.a connectome loaded + cross-validated (logbook 022). - G1.b registry instantiates MLP-PPO + connectome through same code path (no per-arch branches in run_simulation.py). - G1.c PPO-on-connectome learns klinotaxis: R2b reference run hits 100% sustained success on last-100 episodes; all three literal pass-criteria satisfied (no NaN/Inf over 500 ep, 16.1x last-25 reward margin over frozen-random-weights control, monotonic improvement 76% to 100%); within 6 points of MLPPPO + LSTMPPO baselines on the same task / env / seed. - G1.d migration regression byte-equivalent on MLPPPO + LSTMPPO; registration-only migration on the other 17 changes no executing code. docs/experiments/README.md gets entries for logbooks 022 and 023 (022 entry was missed at the time the previous tranche merged). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
phase6-tracking/tasks.md: tick T2.1 through T2.10 sub-tasks; record Gate 1 GO decision with sub-criteria evidence pointers; link the Gate 1 decision to logbook 023. docs/roadmap.md: flip the Phase 6 Tranche Tracker T2 row from "not started" to complete with logbook link; record the Gate 1 GO outcome in the Mid-phase decision gates section; tick the L0 and L1 MUST exit-criteria with pointers to logbooks 022 and 023. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Section 12 pre-merge verification complete: - Pre-commit hooks pass on all 22 branch-touched files. - Full quantum-nematode pytest suite: 3279 passed, 1 skipped, 2 xfailed (pre-existing stale YAMLs from main). - openspec validate add-architecture-plugin-interface --strict clean. - No new >100KB files; no absolute-path leaks in committed content. 12.6 (user authorisation for push / PR create) remains intentionally unticked — that pause is the next required step. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…asks Three findings surfaced during T2's connectome-PPO klinotaxis evaluation that T4 must consume before the L2 sweep starts: - T4.0d Connectome PPO entropy schedule. R2 -> R2b empirical evidence in logbook 023: constant entropy_coef=0.02 triggers late-training drift; entropy_coef=0.005 (or an entropy decay schedule) eliminates it. T4's T4.connectome.* cells must pick a documented default. - T4.0e Promote the connectome klinotaxis low-entropy config to canonical; the entropy=0.02 variant has documented drift and the low-entropy variant is the R2b reference run that closed Gate 1. - T4.0f Acknowledge mlpppo_small_klinotaxis.yml (newly inferred at T2) as the canonical MLPPPO klinotaxis foraging baseline that T4.mlp_ppo.klinotaxis consumes. A "Carry-forward from T2" preamble above the T4 planning sub-tasks flags the trio for anyone picking up T4. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ipped implementation Pre-archive divergence audit + alignment. The OpenSpec artefacts had drifted in a handful of places against what actually shipped during the connectome-ppo-brain follow-up PR: Spec / design / proposal updates: - Forward-pass depth default K corrected to 4 in design.md (was K=1 in two places + the Open Questions section). K=1 produces degenerate output under strict-mask; K=4 matches the canonical klinotaxis pathway depth. - spec.md ConnectomePPOBrainConfig: forward_pass_depth default corrected to 4 (was 1). - Added sensing_mode (oracle | klinotaxis) to the spec's required fields and updated the sensor-projection scenario to describe the per-mode food_gains matrix shape. - Removed proprioception / AVAL/AVAR/AVBL/AVBR projection language from design.md + proposal.md + tasks.md 7.3: scoped in early design but not implemented in T2 (food chemotaxis only ships; proprioception / mechanosensation / nociception land in T3 + T4). - Gap-junction normalisation description in design.md corrected to symmetric `G[i, j] / sqrt(d_i * d_j)` (was stale per-row). - Dropped 'deferred to follow-up PR' notes from the ConnectomePPO scenarios in brain-architecture/spec.md (both PRs shipped). - Flipped 'follow-up PR' future-tense language in proposal.md + tasks.md to retrospective; the change is ready to archive. - Migration Plan + Rollback strategy in design.md rewritten to describe the in-process two-construct equivalence test that shipped (not the pickle-fixture path that was scoped out). - Open Questions resolved (smoke-config seed = 2026; the np.allclose question is made moot by the Decision 5 scope amendment). - LOC figures corrected: brain_factory.py shrinks 459 to 240 LOC (~220 removed), not the originally projected ~170 LOC / ~400 removed (proposal.md + tasks.md 6.1 + logbook 023). - Episode budget in design.md updated from 100 to 500 with the rationale (cheap forward pass enabled the longer run). Tasks tick: - 12.6 ticked (user-authorisation pause observed before push). All tasks now ticked except none; OpenSpec change ready to archive after push + PR merge. `openspec validate ... --strict` green. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
All 73 tasks complete (12.6 ticked after the user-authorisation pause on 2026-05-24). The change folder moves to openspec/changes/archive/2026-05-24-add-architecture-plugin-interface/. Spec deltas applied: - openspec/specs/brain-architecture/spec.md gains 7 new requirements (plugin registry, BrainTopology + LearningRule Protocols, MLPPPO + LSTMPPO + 17-arch migration regression bar, registry-enum consistency, external Brain Protocol unchanged) and the existing Brain Type Registry requirement is rewritten to mention the ConnectomePPO entries + the StrEnum migration. - openspec/specs/connectome-ppo-brain/ is created with the 4 requirements that fully describe the new architecture. Also restructures openspec/specs/brain-architecture/spec.md so its Brain Type Registry + Module Exports requirements live inside the canonical ## Requirements section (a pre-existing structural bug from an earlier archive that left them under a stale ## MODIFIED Requirements header — invisible to validate/list/archive). Added a SHALL preamble to those two requirements plus Aggregation Pheromone Sensing Modules so the spec passes strict validation. openspec validate --strict green on both specs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The new openspec/specs/connectome-ppo-brain/spec.md was created with a TBD placeholder Purpose during the archive step. Replace it with the real Purpose paragraph covering: connectome topology source + strict-mask + gap-junction normalisation + sensing-mode variants + motor readout + registration via brain-architecture + sensor-projection scope boundary. Mirrors the fill-in we did for connectome-substrate after T1's archive (commit ae7b0d4). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (3)
✅ Files skipped from review due to trivial changes (1)
📝 WalkthroughWalkthroughThis PR implements ConnectomePPOBrain (connectome-constrained PPO), registers it in the brain plugin registry, adds a ConnectomeTopology with strict/soft chemical masking and optional gap-junctions, implements PPO training with freeze_updates, supplies tests and multiple scenario YAMLs (learning and frozen controls), and updates developer/design/spec/tracker documentation. ChangesConnectomePPO Brain Implementation
Sequence Diagram(s)sequenceDiagram
participant Env as Environment
participant Brain as ConnectomePPOBrain
participant Topo as ConnectomeTopology
participant Crit as CriticHead
participant Optim as PPOOptimizer
Env->>Brain: run_brain(BrainParams)
Brain->>Topo: preprocess -> forward_with_hidden(food_features)
Topo->>Topo: K iterations: masked_chem + gap_junctions -> tanh updates
Topo-->>Brain: logits, hidden
Brain->>Crit: value = critic(hidden)
Brain->>Env: sample action -> ActionData
Note over Brain,Optim: replay buffer fills
Brain->>Optim: _perform_ppo_update() when buffer full / episode end
Optim->>Topo: recompute logits/values per minibatch
Optim->>Optim: compute PPO loss + backprop + grad clip
Optim->>Topo: if strict -> apply_weight_mask(topology.w_chem)
Estimated code review effort🎯 4 (Complex) | ⏱️ ~65 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
Codecov Report❌ Patch coverage is 📢 Thoughts on this report? Let us know! |
chrisjz
marked this pull request as ready for review
May 24, 2026 07:20
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 8
🤖 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.
Inline comments:
In `@configs/scenarios/foraging/connectome_ppo_klinotaxis_frozen_control.yml`:
- Line 1: The filename connectome_ppo_klinotaxis_frozen_control.yml does not
follow the required {brain}_{size}[_{variant}]_{sensing}.yml pattern; rename it
so the first token is the brain (connectome), the second token is the size
(replace or map the current "ppo" token to the correct size identifier),
optionally append the variant (frozen_control) before the final sensing token,
and place the sensing token (klinotaxis) last (e.g.,
connectome_{size}[_frozen_control]_klinotaxis.yml); update any references to
this file in the repo to use the new filename.
In `@configs/scenarios/foraging/connectome_ppo_klinotaxis_low_entropy.yml`:
- Line 1: The scenario file
configs/scenarios/connectome_ppo_klinotaxis_low_entropy.yml violates the naming
contract {brain}_{size}[_{variant}]_{sensing}.yml because it omits the size and
places sensing before the trailing position; rename the file to include the size
and move sensing to the end (for example
connectome_ppo_R2b_klinotaxis_low_entropy.yml or
connectome_ppo_R2b_low_entropy_klinotaxis.yml depending on which token is the
sensing field) so the filename matches the pattern and update any references to
this filename in code or CI/docs.
In `@configs/scenarios/foraging/connectome_ppo_klinotaxis.yml`:
- Line 1: The filename connectome_ppo_klinotaxis.yml doesn't follow the required
pattern {brain}_{size}[_{variant}]_{sensing}. Rename the file to match that
convention (e.g., connectome_{size}[_variant]_klinotaxis.yml—replace {size} with
the appropriate size token and add an optional variant if needed), and update
all references to this scenario (scenario registries/loaders, imports, test
fixtures, CI or docs) so the new filename is used everywhere; ensure the
scenario loader still finds the file after renaming.
In `@configs/scenarios/foraging/connectome_ppo_oracle_frozen_control.yml`:
- Line 1: The filename connectome_ppo_oracle_frozen_control.yml does not match
the required pattern {brain}_{size}[_{variant}]_{sensing}.yml; rename it to
include an explicit size token and a terminal sensing token (and keep the
existing variant if intended), e.g. follow the pattern by changing
connectome_ppo_oracle_frozen_control.yml into something like
connectome_ppo_{size}_oracle_frozen_control_{sensing}.yml (replace {size} and
{sensing} with the appropriate concrete tokens used across configs).
In `@configs/scenarios/foraging/connectome_ppo_oracle.yml`:
- Line 1: The config filename connectome_ppo_oracle.yml is missing the required
{size} segment; rename the file to follow the convention
{brain}_{size}[_{variant}]_{sensing}.yml (for example
connectome_ppo_small_oracle_[sensing].yml or appropriate size/variant/sensing
tokens) and update every reference to this file in code and other configs
(search for connectome_ppo_oracle in loaders, scenario registries, import
statements, and CI/test fixtures) so the loader expecting the
{brain}_{size}[_{variant}]_{sensing}.yml pattern will find it.
In `@packages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.py`:
- Around line 324-338: The apply_weight_mask method currently mutates internal
state and takes a mode string, which violates the shared topology contract
expecting a pure projector of shape apply_weight_mask(weights) -> Tensor; change
apply_weight_mask in ConnectomeTopology and the connectome_ppo implementation to
accept a weights Tensor and return the masked Tensor (do not mutate internal
state). Preserve a thin compatibility wrapper if needed: keep the old mode-based
signature (apply_weight_mask(self, *, mode)) to call the new projector or raise
a clear error, but update the core masking logic (currently in
_project_chem_weights) to be a pure function that takes weights and returns
masked_weights; update any callers (including the other occurrence of
apply_weight_mask referenced in the diff) to use the projector form.
- Around line 432-437: The constructor for the ConnectomePPO brain currently
gives `device: DeviceType = DeviceType.CPU`, which weakens the explicit-device
invariant; change the __init__ signature in the ConnectomePPO class to require
an explicit device (remove the default), i.e. accept `device: DeviceType` (no
default) alongside `config: ConnectomePPOBrainConfig` and `action_set`
unchanged, and ensure any internal uses still reference that parameter; after
this change, update any callers (e.g., places that instantiate
ConnectomePPOBrain, including utils/brain_factory.py) to pass a DeviceType
explicitly.
In `@pyproject.toml`:
- Around line 92-93: Remove the special-case ignorances for pydocstyle codes
"D101" and "D102" that are currently applied to test files (the pattern
"/**/tests/**/*.py") in pyproject.toml so tests are subject to the same
NumPy-style docstring rules; locate the per-file-ignores or pydocstyle/ruff
ignore list that references "D101" and "D102" and either delete those codes from
the ignore entry for the tests pattern or remove the per-file-ignores block
entirely so D101/D102 are enforced for test modules and methods.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 16e9c397-4861-4d6c-ba78-0645b175f421
📒 Files selected for processing (29)
AGENTS.mdconfigs/scenarios/foraging/connectome_ppo_klinotaxis.ymlconfigs/scenarios/foraging/connectome_ppo_klinotaxis_frozen_control.ymlconfigs/scenarios/foraging/connectome_ppo_klinotaxis_low_entropy.ymlconfigs/scenarios/foraging/connectome_ppo_oracle.ymlconfigs/scenarios/foraging/connectome_ppo_oracle_frozen_control.ymlconfigs/scenarios/foraging/mlpppo_small_klinotaxis.ymldocs/architecture/plugin-developer-guide.mddocs/experiments/README.mddocs/experiments/logbooks/023-architecture-plugin-interface.mddocs/roadmap.mdopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/.openspec.yamlopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/design.mdopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/notes/dispatcher-audit.mdopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/proposal.mdopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/specs/brain-architecture/spec.mdopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/specs/connectome-ppo-brain/spec.mdopenspec/changes/archive/2026-05-24-add-architecture-plugin-interface/tasks.mdopenspec/changes/phase6-tracking/tasks.mdopenspec/config.yamlopenspec/specs/brain-architecture/spec.mdopenspec/specs/connectome-ppo-brain/spec.mdpackages/quantum-nematode/quantumnematode/brain/arch/__init__.pypackages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.pypackages/quantum-nematode/quantumnematode/brain/arch/dtypes.pypackages/quantum-nematode/quantumnematode/utils/brain_factory.pypackages/quantum-nematode/quantumnematode/utils/config_loader.pypackages/quantum-nematode/tests/quantumnematode_tests/brain/arch/test_connectome_ppo.pypyproject.toml
| @@ -0,0 +1,71 @@ | |||
| # Frozen-random-weights control variant of the connectome-PPO | |||
There was a problem hiding this comment.
Rename this frozen-control scenario file to the required naming convention.
connectome_ppo_klinotaxis_frozen_control.yml does not satisfy the mandated {brain}_{size}[_{variant}]_{sensing}.yml format.
As per coding guidelines, configuration files under configs/scenarios/**/*.yml must follow {brain}_{size}[_{variant}]_{sensing}.yml.
🤖 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 `@configs/scenarios/foraging/connectome_ppo_klinotaxis_frozen_control.yml` at
line 1, The filename connectome_ppo_klinotaxis_frozen_control.yml does not
follow the required {brain}_{size}[_{variant}]_{sensing}.yml pattern; rename it
so the first token is the brain (connectome), the second token is the size
(replace or map the current "ppo" token to the correct size identifier),
optionally append the variant (frozen_control) before the final sensing token,
and place the sensing token (klinotaxis) last (e.g.,
connectome_{size}[_frozen_control]_klinotaxis.yml); update any references to
this file in the repo to use the new filename.
| @@ -0,0 +1,69 @@ | |||
| # Connectome-PPO klinotaxis foraging — low-entropy variant for R2b | |||
There was a problem hiding this comment.
Rename this scenario file to match the required configs/scenarios naming contract.
connectome_ppo_klinotaxis_low_entropy.yml does not match {brain}_{size}[_{variant}]_{sensing}.yml (size is missing and sensing is not in the trailing position), which can break convention-based tooling and indexing.
As per coding guidelines, "Configuration files in configs/scenarios must follow naming convention: {brain}{size}[{variant}]_{sensing}.yml".
🤖 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 `@configs/scenarios/foraging/connectome_ppo_klinotaxis_low_entropy.yml` at line
1, The scenario file configs/scenarios/connectome_ppo_klinotaxis_low_entropy.yml
violates the naming contract {brain}_{size}[_{variant}]_{sensing}.yml because it
omits the size and places sensing before the trailing position; rename the file
to include the size and move sensing to the end (for example
connectome_ppo_R2b_klinotaxis_low_entropy.yml or
connectome_ppo_R2b_low_entropy_klinotaxis.yml depending on which token is the
sensing field) so the filename matches the pattern and update any references to
this filename in code or CI/docs.
| @@ -0,0 +1,77 @@ | |||
| # Connectome-PPO klinotaxis-mode foraging configuration. | |||
There was a problem hiding this comment.
Rename this scenario file to match the required scenario naming pattern.
connectome_ppo_klinotaxis.yml does not follow the required {brain}_{size}[_{variant}]_{sensing}.yml convention.
As per coding guidelines, configuration files under configs/scenarios/**/*.yml must follow {brain}_{size}[_{variant}]_{sensing}.yml.
🤖 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 `@configs/scenarios/foraging/connectome_ppo_klinotaxis.yml` at line 1, The
filename connectome_ppo_klinotaxis.yml doesn't follow the required pattern
{brain}_{size}[_{variant}]_{sensing}. Rename the file to match that convention
(e.g., connectome_{size}[_variant]_klinotaxis.yml—replace {size} with the
appropriate size token and add an optional variant if needed), and update all
references to this scenario (scenario registries/loaders, imports, test
fixtures, CI or docs) so the new filename is used everywhere; ensure the
scenario loader still finds the file after renaming.
| @@ -0,0 +1,63 @@ | |||
| # Frozen-random-weights control variant of the connectome-PPO oracle | |||
There was a problem hiding this comment.
Filename token order violates the required scenario pattern.
connectome_ppo_oracle_frozen_control.yml is not in {brain}_{size}[_{variant}]_{sensing}.yml form (missing size and non-terminal sensing token), so it should be renamed to a compliant pattern.
Based on learnings, "all scenario configuration YAML files under configs/scenarios/ must be named using the pattern {brain}_{size}[_{variant}]_{sensing}.yml".
🤖 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 `@configs/scenarios/foraging/connectome_ppo_oracle_frozen_control.yml` at line
1, The filename connectome_ppo_oracle_frozen_control.yml does not match the
required pattern {brain}_{size}[_{variant}]_{sensing}.yml; rename it to include
an explicit size token and a terminal sensing token (and keep the existing
variant if intended), e.g. follow the pattern by changing
connectome_ppo_oracle_frozen_control.yml into something like
connectome_ppo_{size}_oracle_frozen_control_{sensing}.yml (replace {size} and
{sensing} with the appropriate concrete tokens used across configs).
| @@ -0,0 +1,68 @@ | |||
| # Connectome-PPO oracle-mode foraging configuration — early baseline. | |||
There was a problem hiding this comment.
Filename is missing the required {size} segment.
connectome_ppo_oracle.yml does not satisfy {brain}_{size}[_{variant}]_{sensing}.yml, so this should be renamed (and references updated) to keep config loading and conventions consistent.
As per coding guidelines, "Configuration files in configs/scenarios must follow naming convention: {brain}{size}[{variant}]_{sensing}.yml".
🤖 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 `@configs/scenarios/foraging/connectome_ppo_oracle.yml` at line 1, The config
filename connectome_ppo_oracle.yml is missing the required {size} segment;
rename the file to follow the convention
{brain}_{size}[_{variant}]_{sensing}.yml (for example
connectome_ppo_small_oracle_[sensing].yml or appropriate size/variant/sensing
tokens) and update every reference to this file in code and other configs
(search for connectome_ppo_oracle in loaders, scenario registries, import
statements, and CI/test fixtures) so the loader expecting the
{brain}_{size}[_{variant}]_{sensing}.yml pattern will find it.
Comment on lines
+92
to
+93
| "D101", # Some test classes are pure containers — docstring not informative | ||
| "D102", # Test method names already describe the assertion |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Locate ruff pydocstyle / docstring policy in pyproject.toml
ls -la
rg -n "ruff|pydocstyle|D101|D102|ignore|per-file-ignores" -S pyproject.toml || true
# Show the relevant section around the reported lines
nl -ba pyproject.toml | sed -n '70,120p'
# Search repo for docstring policy language (NumPy-style, **/*.py, etc.)
rg -n "NumPy-style|NumPy doc|docstring|pydocstyle|D101|D102|\\*\\*/\\*\\.py|\\*\\*\\/\\*\\.py|contributing|coding guideline" -S . || trueRepository: SyntheticBrains/nematode
Length of output: 4687
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# 1) Show the relevant pyproject.toml sections around per-file ignores and pydocstyle
echo "---- pyproject.toml per-file-ignores context ----"
awk 'NR>=70 && NR<=110 {print NR ":" $0}' pyproject.toml
echo "---- pyproject.toml pydocstyle config context ----"
awk 'NR>=130 && NR<=190 {print NR ":" $0}' pyproject.toml
# 2) Search for repo-wide docstring policy language
rg -n "NumPy-style|NumPy docstring|pydocstyle|D101|D102|docstring.*\\*\\*/\\*\\.py|\\*\\*/\\*\\.py|docstrings? policy|coding guideline|conventions" \
README.md CONTRIBUTING.md AGENTS.md docs -S || true
# 3) Check if tests are mentioned as exempt anywhere in those docs
rg -n "test(s)?|tests|except|exempt|ignore" \
README.md CONTRIBUTING.md AGENTS.md docs -S || trueRepository: SyntheticBrains/nematode
Length of output: 50381
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "---- AGENTS.md around NumPy-style docstrings ----"
awk 'NR>=45 && NR<=75 {print NR ":" $0}' AGENTS.md
echo "---- CONTRIBUTING.md around NumPy-style docstrings ----"
# Find the exact line containing the guideline then print a small window.
line=$(rg -n "NumPy-style docstrings|Docstrings.*NumPy|Use NumPy-style docstrings" CONTRIBUTING.md | head -n1 | cut -d: -f1 || true)
if [ -n "${line:-}" ]; then
start=$((line-10)); end=$((line+30));
awk -v s="$start" -v e="$end" 'NR>=s && NR<=e {print NR ":" $0}' CONTRIBUTING.md
else
rg -n "NumPy-style docstrings|Docstrings.*NumPy|Use NumPy-style docstrings" CONTRIBUTING.md -n || true
fi
echo "---- Search for explicit '**/*.py' mention ----"
rg -n "\\*\\*/\\*\\.py|\\*\\*\\/\\*\\.py|\\*\\*\\/\\*\\.py|\\*\\*/\\*\\.py|\\*\\*\\/\\*\\.py" AGENTS.md CONTRIBUTING.md -S || true
echo "---- Search for docstring exemptions/exclude tests ----"
rg -n "docstring.*test|tests.*docstring|exclude.*tests|exempt.*tests|tests are exempt|per-file-ignores|ruff.*pydocstyle" AGENTS.md CONTRIBUTING.md pyproject.toml -S || trueRepository: SyntheticBrains/nematode
Length of output: 3511
🏁 Script executed:
#!/bin/bash
set -euo pipefail
awk 'NR>=110 && NR<=150 {print NR ":" $0}' CONTRIBUTING.mdRepository: SyntheticBrains/nematode
Length of output: 1489
Remove Ruff D101/D102 exceptions for tests to keep NumPy docstring policy consistent
AGENTS.md and CONTRIBUTING.md require NumPy-style docstrings without excluding tests, but pyproject.toml config disables pydocstyle D101/D102 for /**/tests/**/*.py, creating a policy/tooling mismatch.
Suggested change
"/**/tests/**/*.py" = [
"ANN",
"ARG",
"D100",
- "D101", # Some test classes are pure containers — docstring not informative
- "D102", # Test method names already describe the assertion
"D104",
"PLR2004",
"S101",
"SLF001", # Tests can access private members for testing
"T201",
]🤖 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 `@pyproject.toml` around lines 92 - 93, Remove the special-case ignorances for
pydocstyle codes "D101" and "D102" that are currently applied to test files (the
pattern "/**/tests/**/*.py") in pyproject.toml so tests are subject to the same
NumPy-style docstring rules; locate the per-file-ignores or pydocstyle/ruff
ignore list that references "D101" and "D102" and either delete those codes from
the ignore entry for the tests pattern or remove the per-file-ignores block
entirely so D101/D102 are enforced for test modules and methods.
Two reviewer findings addressed; three skipped with rationale below.
(1) Rename connectome configs to match the established
`{brain}_{size}[_{variant}]_{sensing}.yml` convention from AGENTS.md.
The reviewer parsed `connectome_ppo_*` as `{brain=connectome}_{size=ppo}`,
which is wrong — the brain token is `connectomeppo` (the registered
name in the plugin registry). Rename clarifies that and aligns with
every other config in the directory:
connectome_ppo_klinotaxis.yml
→ connectomeppo_small_klinotaxis.yml
connectome_ppo_klinotaxis_frozen_control.yml
→ connectomeppo_small_frozen_control_klinotaxis.yml
connectome_ppo_klinotaxis_low_entropy.yml
→ connectomeppo_small_low_entropy_klinotaxis.yml
connectome_ppo_oracle.yml
→ connectomeppo_small_oracle.yml
connectome_ppo_oracle_frozen_control.yml
→ connectomeppo_small_frozen_control_oracle.yml
Variant tokens (`frozen_control`, `low_entropy`) sit between size and
sensing, matching `crhqlstm_small_classical_oracle.yml` etc.
Live references updated: logbook 023, phase6-tracking T4.0e, the
configs' own cross-refs, and the mlpppo klinotaxis baseline's
cross-ref. Archived OpenSpec change docs are left untouched
(immutable historical record).
(2) Fix Protocol-conformance bug on `ConnectomeTopology.apply_weight_mask`.
The `BrainTopology` Protocol specifies a pure projector:
`apply_weight_mask(weights: torch.Tensor) -> torch.Tensor`. The
previous signature `apply_weight_mask(*, mode: str) -> None`
mutated `self.w_chem.data` in-place and conflated the structural
projector with the brain's mask-mode policy.
`ConnectomeTopology.apply_weight_mask(weights)` is now stateless —
returns `weights * M_chem`. The brain's update loop owns the policy:
under `chemical_mask_mode="strict"` it copies the projected tensor
back into `topology.w_chem.data`; under `"soft_prior"` it skips the
projection entirely. The `_project_chem_weights` helper is removed;
its single internal use at construction-time now goes through the
public projector too.
Tests updated: `test_soft_prior_apply_mask_is_noop` is replaced
with two tests — `test_apply_weight_mask_is_pure_projector` (asserts
the topology's projector is stateless + returns the right tensor)
and `test_soft_prior_mode_does_not_project_after_ppo_step` (asserts
the brain's policy honours the mode by driving a PPO update under
soft-prior + checking ~M_chem weights survive).
Skipped reviewer findings with rationale:
- Make `device` non-default on `ConnectomePPOBrain.__init__`. Skipped:
every other PPO brain in the codebase (mlpppo, lstmppo, mlpreinforce)
defaults `device: DeviceType = DeviceType.CPU`. Diverging would
violate the established constructor convention.
- Remove D101/D102 from per-file-ignores for tests. Skipped: these
ignores were added intentionally in commit b025677 with a documented
rationale ("test classes are pure containers + test method names are
already descriptive"); they sit alongside the same test-relaxation
pattern used for D100/D104/ANN/ARG/S101/PLR2004/T201 throughout the
codebase. Removing them would diverge from the consistent
test-vs-implementation docstring policy.
Verification:
- `uv run pytest -m "not nightly" packages/quantum-nematode/` →
3280 passed (the soft-prior tests went 1 → 2, so total +1 vs main),
1 skipped, 2 xfailed (pre-existing stale YAMLs).
- `uv run pre-commit run --files <changed>` → all hooks green.
- `openspec validate brain-architecture --strict` /
`openspec validate connectome-ppo-brain --strict` /
`openspec validate phase6-tracking --strict` → all valid.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (2)
packages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.py (1)
320-331: ⚡ Quick winUse NumPy-style sections in the new method docstring
Please convert this new docstring to NumPy-style (
Parameters,Returns, etc.) to match repository standards.As per coding guidelines,
**/*.py: Use NumPy-style docstrings for documenting functions and classes.🤖 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 `@packages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.py` around lines 320 - 331, Update the docstring for apply_weight_mask to NumPy-style: add a "Parameters" section documenting weights : torch.Tensor and its description, and a "Returns" section documenting the returned torch.Tensor and its description (e.g., projected weights with chemical-synapse edges zeroed); keep the existing high-level explanation about strict-mask behavior and purity but convert prose into concise NumPy sections and ensure type hints and intent (stateless, no in-place mutation) remain mentioned in either the Returns or Notes section as appropriate.packages/quantum-nematode/tests/quantumnematode_tests/brain/arch/test_connectome_ppo.py (1)
318-323: ⚡ Quick winNormalize new test method docs to NumPy style
The newly added test docstrings (and helper method docs) should use NumPy-style sections for consistency with the repo rule.
As per coding guidelines,
**/*.py: Use NumPy-style docstrings for documenting functions and classes.Also applies to: 339-346, 365-379
🤖 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 `@packages/quantum-nematode/tests/quantumnematode_tests/brain/arch/test_connectome_ppo.py` around lines 318 - 323, Update the docstrings for the new test and its helper methods to NumPy-style: replace the current free-form/Google-style descriptions in test_apply_weight_mask_is_pure_projector (and the other affected methods around the same change) with NumPy sections such as Parameters, Returns, and Notes/Examples as appropriate so they conform to the project's NumPy-style docstring convention; ensure the test purpose remains the same (describe that ConnectomeTopology.apply_weight_mask is a stateless projector returning weights * M_chem without mutating self) and that any helper method docs similarly use NumPy section headers and brief descriptions.
🤖 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.
Inline comments:
In `@packages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.py`:
- Around line 713-722: The forward pass currently multiplies weights by the mask
unconditionally (masked_chem = self.w_chem * self.m_chem), which zeroes
gradients on masked edges and prevents the "soft_prior" mode from growing new
edges; change the forward logic so that the mask multiplication is only applied
when self.config.chemical_mask_mode == "strict" and for "soft_prior" use the raw
self.w_chem (so gradients flow through masked entries), while keeping the
existing strict post-optimizer projection that calls
topology.apply_weight_mask(self.topology.w_chem.data) intact; update references
in the forward method to use masked_chem (or raw) depending on
self.config.chemical_mask_mode and ensure topology.w_chem / self.w_chem
semantics remain consistent.
---
Nitpick comments:
In `@packages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.py`:
- Around line 320-331: Update the docstring for apply_weight_mask to
NumPy-style: add a "Parameters" section documenting weights : torch.Tensor and
its description, and a "Returns" section documenting the returned torch.Tensor
and its description (e.g., projected weights with chemical-synapse edges
zeroed); keep the existing high-level explanation about strict-mask behavior and
purity but convert prose into concise NumPy sections and ensure type hints and
intent (stateless, no in-place mutation) remain mentioned in either the Returns
or Notes section as appropriate.
In
`@packages/quantum-nematode/tests/quantumnematode_tests/brain/arch/test_connectome_ppo.py`:
- Around line 318-323: Update the docstrings for the new test and its helper
methods to NumPy-style: replace the current free-form/Google-style descriptions
in test_apply_weight_mask_is_pure_projector (and the other affected methods
around the same change) with NumPy sections such as Parameters, Returns, and
Notes/Examples as appropriate so they conform to the project's NumPy-style
docstring convention; ensure the test purpose remains the same (describe that
ConnectomeTopology.apply_weight_mask is a stateless projector returning weights
* M_chem without mutating self) and that any helper method docs similarly use
NumPy section headers and brief descriptions.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: b9f84529-53df-4edd-917b-5460d09d1341
📒 Files selected for processing (10)
configs/scenarios/foraging/connectomeppo_small_frozen_control_klinotaxis.ymlconfigs/scenarios/foraging/connectomeppo_small_frozen_control_oracle.ymlconfigs/scenarios/foraging/connectomeppo_small_klinotaxis.ymlconfigs/scenarios/foraging/connectomeppo_small_low_entropy_klinotaxis.ymlconfigs/scenarios/foraging/connectomeppo_small_oracle.ymlconfigs/scenarios/foraging/mlpppo_small_klinotaxis.ymldocs/experiments/logbooks/023-architecture-plugin-interface.mdopenspec/changes/phase6-tracking/tasks.mdpackages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.pypackages/quantum-nematode/tests/quantumnematode_tests/brain/arch/test_connectome_ppo.py
💤 Files with no reviewable changes (1)
- configs/scenarios/foraging/connectomeppo_small_oracle.yml
✅ Files skipped from review due to trivial changes (3)
- configs/scenarios/foraging/connectomeppo_small_frozen_control_klinotaxis.yml
- docs/experiments/logbooks/023-architecture-plugin-interface.md
- configs/scenarios/foraging/mlpppo_small_klinotaxis.yml
The forward pass unconditionally multiplied ``w_chem`` by ``m_chem``, which made the soft-prior mode silently equivalent to strict: backprop multiplied the upstream gradient by ``m_chem``, pinning gradients on non-wild-type entries to exactly zero. With zero gradient and zero initialisation, the optimiser could never grow new edges — so the "PPO is free to grow new connections" guarantee soft-prior promised was never actually delivered. Make the mask-or-no-mask choice a structural property of the topology (``enforce_strict_mask: bool`` constructor kwarg) so the forward pass branches: - strict: ``chem_mat = w_chem * m_chem`` → gradients pinned to wild-type - soft_prior: ``chem_mat = w_chem`` → gradients flow everywhere The brain passes ``enforce_strict_mask=(config.chemical_mask_mode == "strict")`` into the topology at construction. The post-optimiser projection in the brain's update loop (under strict mode) stays as defence-in-depth — combined with the forward-pass masking, the strict-mask invariant is guaranteed by two independent mechanisms. Tests added: - ``test_soft_prior_gradients_flow_through_non_wild_type_edges`` — under soft-prior, backprop produces non-zero gradient on at least one ~M_chem entry. - ``test_strict_mode_pins_gradients_to_wild_type_edges`` — under strict, backprop produces exactly-zero gradient on every ~M_chem entry. The strict-mask invariant test (post-PPO-update ``w_chem * ~m_chem == 0``) continues to pass; under the new mechanism the invariant is maintained primarily by the forward-pass masking rather than the post-step projection (the projection is now redundant defence under strict and skipped under soft-prior). Spec updated: the "Chemical-synapse strict-mask" and "Soft-prior mode" scenarios in openspec/specs/connectome-ppo-brain/spec.md now describe the forward-pass gradient-flow mechanism, not just the post-step projection. Verification: - ``uv run pytest -m "not nightly" packages/quantum-nematode/`` → 3282 passed (+2 from the new gradient-flow tests), 1 skipped, 2 xfailed (pre-existing). - ``uv run pre-commit run --files <changed>`` → all hooks green. - ``openspec validate connectome-ppo-brain --strict`` → valid. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 26, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
ConnectomePPOBrain) — PPO-trainable policy over the wild-type Cook 2019 hermaphrodite connectome (302 neurons, 3709 chemical synapses strict-masked + PPO-learnable, 1093 gap junctions symmetric fan-in normalised + non-learnable). Registered asconnectomeppothrough the plugin registry from PR feat(brain-architecture): plugin registry for brain architectures (Phase 6 T2 L1, partial) #182; 20 brains total.Closes Gate 1
scripts/run_simulation.pyWhat's in the diff
packages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.py, 720 LOC + 420 LOC test):ConnectomePPOBrain+ConnectomeTopology+ConnectomePPOBrainConfig. Forward passh = tanh(W_chem.T @ (M_chem * h) + G_gap.T @ h)iterated K=4 times (canonical klinotaxis pathway depth). Twosensing_modevariants (oracle2-feature;klinotaxis3-feature). Motor readout pools VB/DB/VA/DA → learnable 4×4 → 4 action logits.freeze_updates: trueshort-circuits PPO step for the paired control.docs/experiments/README.md.docs/architecture/plugin-developer-guide.mdwith worked example.phase6-tracking/tasks.mdT2.1–T2.10 ticked, Gate 1 GO decision recorded + linked; T4.0d/e/f carry-forward sub-tasks added (entropy schedule, low-entropy config promotion, MLPPPO klinotaxis baseline acknowledgement).docs/roadmap.mdT2 row flipped to ✅ with logbook link; L0 + L1 MUST exit-criteria ticked.add-architecture-plugin-interfacearchived to2026-05-24-add-architecture-plugin-interface; main specs updated (brain-architecturegains 7 ADDED requirements + 1 MODIFIED; newconnectome-ppo-braincapability created with Purpose section filled in). Also fixes a pre-existing structural bug inbrain-architecture/spec.md(orphaned## MODIFIED Requirementsheader from an earlier archive that left requirements invisible to OpenSpec parsing).Test plan
uv run pre-commit run --files <every changed file>— green on all 30 changed files (mdformat, markdownlint, ruff, ruff-format, pyright, tests)uv run pytest -m "not nightly" packages/quantum-nematode/— 3279 passed, 1 skipped, 2 xfailed (pre-existing stale YAMLs on main); zero new failuresopenspec validate brain-architecture --strict+openspec validate connectome-ppo-brain --strict— both validfrom quantumnematode.brain.arch import ConnectomePPOBrain; 'connectomeppo' in list_registered_brains()→True; 20 registered names/Users/,/home/,C:\Users\,file:///) in any committed content.pyor config.ymlfiles🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Documentation
Tests
Chores