Connectome degree-preserving rewired-null control (Phase 6a) â degree-statistics, not wiring - #264
Merged
Merged
Conversation
Degree-preserving rewired-null control for the wild-type connectome â tests whether its 029 5th-of-6 standing is its specific C. elegans wiring or merely its degree statistics (Dhiman 2026). Config-option on the existing connectome brain (wiring: wild_type | rewired_degree_preserving + rewire_seed), a hand-rolled seeded degree-preserving double-edge-swap (chemical directed + gap undirected, no networkx dep), matched init/budget vs the 029 connectome cell, paired-seed Wilcoxon + BH-FDR verdict. Byte-identical when wiring: wild_type. Scope trimmed to the gating rewired-null; the learnable-gap-junction control (T7.controls.learnable_gj) is a tracked fast-follow. Authored + reviewed (/nematode-review-spec: matched-init RNG-independence fix applied); validate --strict + pre-commit green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- connectome/rewiring.py: seeded degree-preserving double-edge-swap (directed chemical + undirected gap), no networkx dep; preserves every neuron's in/out degree, rejects self-loops/duplicates, deterministic under seed. 12 tests. - ConnectomePPOBrain: `wiring` (wild_type | rewired_degree_preserving) + `rewire_seed` config fields; the transform runs at the load->topology seam on a DEDICATED RNG so the weight-init stream is unperturbed (matched init vs wild-type for the same seed). wild_type is byte-identical (guarded by a test reconstructing the mask from the loaded Cook adjacency). - TestWiringControl: wild-type == loaded connectome; rewired preserves degree but changes edges; gap stays symmetric; forward pass finite; seed controls the draw. - Rewired-null config: verbatim copy of the combined-C3 connectome cell + the one wiring field. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
scripts/analysis/connectome_structure_controls.py â paired wild-type vs rewired-null comparison on the C3 ranked-success metric, reusing the committed t7_continuous_ranking._plateau_tail metric + the paired-seed Wilcoxon / 80% bootstrap / BH-FDR layer. Pre-registered verdict: SPECIFIC-WIRING (wild-type > rewired at q<0.05) vs DEGREE-STATISTICS (CI spans 0), with the rewired-beats- wild-type edge case reported honestly. 6 tests. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Degree-preserving rewired-null control for the connectome's 029 5th-of-6 standing: is it the specific C. elegans wiring or just the degree statistics? Panel (wild-type vs degree-preserving rewired-null, n=8 paired seeds, 6000ep, matched init/budget): wild-type 52.78% vs rewired-null 56.06% â paired delta d=-3.28, CI[-8.56,+1.61] (spans 0), BH-FDR q=0.770 â indistinguishable (rewired nominally higher). The connectome's ranking is a DEGREE-STATISTICS result, not a wiring result: under PPO weight search the specific wiring confers no advantage over degree-matched alternatives (Beiran & Litwin-Kumar 2025 degeneracy). Does NOT say the wiring is functionless â its value is the evolved fixed weights. A single-seed smoke hinted "specific wiring" (57.2 vs 46.5) but reversed at n=8. Logbook 034 + committed supporting artefacts (controls.json + per-seed CSV + forensics), README row, tracker tick (T7.controls.rewired_null RESOLVED: DEGREE-STATISTICS; learnable_gj deferred as fast-follow), AGENTS.md _rewired_null variant doc, §5-§7 tasks ticked. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change complete + evaluated (DEGREE-STATISTICS, Logbook 034). Archived to openspec/changes/archive/2026-07-05-add-connectome-structure-controls/ with all tasks ticked (§1â§8); the degree-preserving rewired-null wiring requirement is applied to openspec/specs/connectome-ppo-brain/ (heading spacing fixed post-archive). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pre-push `pre-commit run -a` normalisation (magic trailing comma; no behaviour change). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Harness (scripts/analysis/connectome_structure_controls.py): - The reused Wilcoxon is one-sided (H1: wild > rewired), so the `rewired_beats` branch was unreachable â a significant rewired-WIN would be mislabeled "inconclusive". Detect it via the bootstrap CI (ci_hi < 0). + a regression test. - Warn when n < 5 paired seeds (the 1/2^n signed-rank floor makes q<0.05 â `specific_wiring` â unreachable; the n>=8 lesson from bit-memory 030); record `underpowered` in the summary. - Skip manifest lines whose seed token isn't an integer (no crash on a stray line). Rewiring (connectome/rewiring.py): - Guard against a non-simple input graph (parallel edges would silently collapse the weight dicts) with a clear ValueError. + a test. Tests: - Assert the matched-init invariant (food_gains/readout byte-identical between wild_type and rewired at the same seed â the control's load-bearing claim). - A rewired brain survives a PPO update (strict-mask holds on the rewired mask, not just wild-type). - Degree-preservation on a non-regular graph (the regular fixture hid in/out confusion); rewired-beats verdict reachability. None of these change the reported DEGREE-STATISTICS verdict (re-verified). Also: drop the AGENTS.md variant-suffix parenthetical for consistency, and drop the tmp/ path reference from the 034 supporting details. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
ð WalkthroughWalkthroughAdds a degree-preserving double-edge-swap rewiring utility for connectomes, integrates it into ChangesRewired-Null Connectome Control
Estimated code review effort: 3 (Moderate) | ~30 minutes Possibly related PRs
Poem A rabbit rewired the worm's own wires, ðĨ Pre-merge checks | â 5â Passed checks (5 passed)
âĻ Finishing Touchesð Generate docstrings
𧊠Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
ð§đ Nitpick comments (3)
scripts/analysis/connectome_structure_controls.py (2)
32-33: ð Maintainability & Code Quality | ðĩ Trivial | ⥠Quick winImporting a private (
_-prefixed) helper across module boundaries.
_plateau_tailis name-mangled as private int7_continuous_ranking.py, yet it's imported and relied upon here as the load-bearing metric contract. This couples the two scripts to an internal implementation detail rather than a stable public API, so a future rename/refactor of_plateau_tailint7_continuous_ranking.pywill silently break this harness without any import-time signal beyond a linter warning.Consider exporting a public wrapper (e.g.,
plateau_tail) fromt7_continuous_ranking.pyfor cross-module reuse.ðĪ 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 `@scripts/analysis/connectome_structure_controls.py` around lines 32 - 33, The harness is importing the private helper _plateau_tail directly from t7_continuous_ranking, which couples connectome_structure_controls to an internal implementation detail. Update t7_continuous_ranking to expose a stable public wrapper such as plateau_tail, then switch the import in connectome_structure_controls to that public symbol and keep the existing metric usage unchanged. Ensure the cross-module contract references the public API only, while _plateau_tail remains internal if still needed.
52-67: ðïļ Data Integrity & Integration | ðĩ Trivial | ⥠Quick winMalformed manifest lines are silently dropped with no diagnostic.
Lines that don't match
len(parts) != 3 or not parts[1].isdigit()are skipped silently (Line 59-60), unlike the parseable-plateau failure path just below which logs aWARN. A typo'd arm/seed manifest entry (e.g., wrong token count, non-integer seed) would vanish from the analysis with no trace, which is harder to debug than the plateau-parse failures that do get logged. Also, a duplicate<arm> <seed>entry (Line 66) silently overwrites the prior value with no warning.ðĄ Proposed fix: warn on unparseable (non-comment) manifest lines
for line in manifest.read_text().splitlines(): parts = line.split() - if ( - len(parts) != 3 or not parts[1].isdigit() - ): # `<arm> <int seed> <out>`; skip anything else + if not parts or parts[0].startswith("#"): + continue + if len(parts) != 3 or not parts[1].isdigit(): + print(f" WARN unparseable manifest line, skipping: {line!r}") continue arm, seed, out_path = parts[0], int(parts[1]), Path(parts[2]) success = _success(REPO / out_path) if success is None: print(f" WARN {arm} seed {seed}: no parseable plateau in {out_path} - dropped") continue + if seed in arms.get(arm, {}): + print(f" WARN {arm} seed {seed}: duplicate manifest entry, overwriting") arms.setdefault(arm, {})[seed] = successðĪ 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 `@scripts/analysis/connectome_structure_controls.py` around lines 52 - 67, The load(manifest) parser in connectome_structure_controls.py is dropping malformed non-comment manifest lines and duplicate <arm> <seed> entries without any visibility. Update load() to emit a WARN for any line that fails the token-count/int-seed check instead of silently continuing, and add a warning when arms.setdefault(arm, {})[seed] would overwrite an existing seed entry so analysis issues are traceable. Keep the existing _success() warning path intact and use the load() loopâs existing arm/seed parsing to centralize the diagnostics.packages/quantum-nematode/tests/quantumnematode_tests/analysis/test_connectome_structure_controls.py (1)
8-9: ð Maintainability & Code Quality | ðĩ Trivial | ðĪ Low valueHardcoded
parents[5]traversal is brittle to future file moves.The fixed-depth
.resolve().parents[5]walk to locate the repo root (and hencescripts/analysis) will silently resolve to the wrong directory if this test file is ever relocated, without raising an error â it'd just fail to importconnectome_structure_controlsor (worse) find an unrelated directory. Low risk today, but worth a comment or a more robust anchor (e.g., searching upward forpyproject.toml) if this pattern spreads.ðĪ 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/analysis/test_connectome_structure_controls.py` around lines 8 - 9, The path setup in the test module is brittle because _ANALYSIS_DIR relies on Path(__file__).resolve().parents[5], so update the import bootstrap to use a stable repo-root anchor or add an explicit comment documenting the fixed depth; in test_connectome_structure_controls, keep the sys.path insertion tied to a robust locator for scripts/analysis rather than a hardcoded parent index.
ðĪ 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_predator_thermal/connectomeppo_small_continuous2d_combined_klinotaxis_rewired_null.yml`:
- Line 1: Rename the scenario config so the variant suffix appears before the
sensing suffix, matching the documented
`{brain}_{size}[_{variant}]_{sensing}.yml` convention. Update the filename to
place `_rewired_null` before `_klinotaxis`, keeping it aligned with the sibling
`connectomeppo_small_continuous2d_combined_klinotaxis.yml` and the AGENTS.md
naming rule.
In `@packages/quantum-nematode/quantumnematode/connectome/rewiring.py`:
- Around line 149-154: The simple-graph validation in rewire_degree_preserving
only checks for duplicate/parallel edges, but it still allows chemical
self-loops through. Update the guard near the chem_edges/gap_edges validation to
also reject any ChemicalSynapse where pre == post, since the Cook loader can
preserve those fixtures unchanged. Keep the fix localized to
rewire_degree_preserving and ensure the ValueError is raised for any self-looped
chemical edge before rewiring proceeds.
---
Nitpick comments:
In
`@packages/quantum-nematode/tests/quantumnematode_tests/analysis/test_connectome_structure_controls.py`:
- Around line 8-9: The path setup in the test module is brittle because
_ANALYSIS_DIR relies on Path(__file__).resolve().parents[5], so update the
import bootstrap to use a stable repo-root anchor or add an explicit comment
documenting the fixed depth; in test_connectome_structure_controls, keep the
sys.path insertion tied to a robust locator for scripts/analysis rather than a
hardcoded parent index.
In `@scripts/analysis/connectome_structure_controls.py`:
- Around line 32-33: The harness is importing the private helper _plateau_tail
directly from t7_continuous_ranking, which couples connectome_structure_controls
to an internal implementation detail. Update t7_continuous_ranking to expose a
stable public wrapper such as plateau_tail, then switch the import in
connectome_structure_controls to that public symbol and keep the existing metric
usage unchanged. Ensure the cross-module contract references the public API
only, while _plateau_tail remains internal if still needed.
- Around line 52-67: The load(manifest) parser in
connectome_structure_controls.py is dropping malformed non-comment manifest
lines and duplicate <arm> <seed> entries without any visibility. Update load()
to emit a WARN for any line that fails the token-count/int-seed check instead of
silently continuing, and add a warning when arms.setdefault(arm, {})[seed] would
overwrite an existing seed entry so analysis issues are traceable. Keep the
existing _success() warning path intact and use the load() loopâs existing
arm/seed parsing to centralize the diagnostics.
ðŠ Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
âđïļ Review info
âïļ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 13e3b839-e330-4809-b128-aee53de44f3d
â Files ignored due to path filters (1)
docs/experiments/logbooks/supporting/034-connectome-structure-controls/success-per-seed.csvis excluded by!**/*.csv
ð Files selected for processing (19)
AGENTS.mdconfigs/scenarios/foraging_predator_thermal/connectomeppo_small_continuous2d_combined_klinotaxis_rewired_null.ymldocs/experiments/README.mddocs/experiments/logbooks/034-connectome-structure-controls.mddocs/experiments/logbooks/supporting/034-connectome-structure-controls/controls.jsondocs/experiments/logbooks/supporting/034-connectome-structure-controls/details.mdopenspec/changes/archive/2026-07-05-add-connectome-structure-controls/.openspec.yamlopenspec/changes/archive/2026-07-05-add-connectome-structure-controls/design.mdopenspec/changes/archive/2026-07-05-add-connectome-structure-controls/proposal.mdopenspec/changes/archive/2026-07-05-add-connectome-structure-controls/specs/connectome-ppo-brain/spec.mdopenspec/changes/archive/2026-07-05-add-connectome-structure-controls/tasks.mdopenspec/changes/phase6-tracking/tasks.mdopenspec/specs/connectome-ppo-brain/spec.mdpackages/quantum-nematode/quantumnematode/brain/arch/connectome_ppo.pypackages/quantum-nematode/quantumnematode/connectome/rewiring.pypackages/quantum-nematode/tests/quantumnematode_tests/analysis/test_connectome_structure_controls.pypackages/quantum-nematode/tests/quantumnematode_tests/brain/arch/test_connectome_ppo.pypackages/quantum-nematode/tests/quantumnematode_tests/connectome/test_rewiring.pyscripts/analysis/connectome_structure_controls.py
| @@ -0,0 +1,142 @@ | |||
| # Degree-preserving rewired-null control of the ConnectomePPO combined-C3 cell. | |||
There was a problem hiding this comment.
ð Maintainability & Code Quality | ð Major | ⥠Quick win
Filename places the new variant suffix after the sensing suffix.
Per AGENTS.md's own convention ({brain}_{size}[_{variant}]_{sensing}.yml, sensing suffixes listed last), _rewired_null is documented as a variant suffix but is appended here after the _klinotaxis sensing suffix, unlike the sibling connectomeppo_small_continuous2d_combined_klinotaxis.yml file this is meant to pair with. Consider connectomeppo_small_continuous2d_combined_rewired_null_klinotaxis.yml to match the documented ordering.
As per coding guidelines: configs/scenarios/**/*.yml: "Scenario configuration files must follow the naming convention {brain}_{size}[_{variant}]_{sensing}.yml."
ðĪ 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_predator_thermal/connectomeppo_small_continuous2d_combined_klinotaxis_rewired_null.yml`
at line 1, Rename the scenario config so the variant suffix appears before the
sensing suffix, matching the documented
`{brain}_{size}[_{variant}]_{sensing}.yml` convention. Update the filename to
place `_rewired_null` before `_klinotaxis`, keeping it aligned with the sibling
`connectomeppo_small_continuous2d_combined_klinotaxis.yml` and the AGENTS.md
naming rule.
Source: Coding guidelines
Comment on lines
+149
to
+154
| # The swap assumes a SIMPLE input graph (no parallel edges) - the weight dicts below key on the | ||
| # edge tuple and would silently collapse duplicates. The Cook loader dedups both edge types, so | ||
| # this is a guard against a hand-built/alternate-loader fixture, not a live path. | ||
| if len(set(chem_edges)) != len(chem_edges) or len(set(gap_edges)) != len(gap_edges): | ||
| msg = "rewire_degree_preserving requires a simple connectome (no duplicate/parallel edges)" | ||
| raise ValueError(msg) |
There was a problem hiding this comment.
ðïļ Data Integrity & Integration | ðĄ Minor | ⥠Quick win
ð§Đ Analysis chain
ð Script executed:
#!/bin/bash
# Confirm whether ChemicalSynapse forbids self-loops (pre == post) at construction.
rg -n -A 15 'class ChemicalSynapse' packages/quantum-nematode/quantumnematode/connectome/model.pyRepository: SyntheticBrains/nematode
Length of output: 706
ð Script executed:
#!/bin/bash
set -euo pipefail
echo "== model.py around ChemicalSynapse =="
sed -n '55,130p' packages/quantum-nematode/quantumnematode/connectome/model.py
echo
echo "== rewiring.py around the degree-preserving swap and output checks =="
sed -n '1,240p' packages/quantum-nematode/quantumnematode/connectome/rewiring.pyRepository: SyntheticBrains/nematode
Length of output: 10650
ð Script executed:
#!/bin/bash
set -euo pipefail
echo "== search for self-loop handling in connectome code =="
rg -n -A 3 -B 3 'self-loop|pre == post|a < b|duplicate|parallel' packages/quantum-nematode/quantumnematode/connectome
echo
echo "== locate parser/loader entry points =="
rg -n -A 5 -B 5 'ChemicalSynapse|GapJunction|Connectome' packages/quantum-nematode/quantumnematode/connectomeRepository: SyntheticBrains/nematode
Length of output: 45748
ð Script executed:
#!/bin/bash
set -euo pipefail
echo "== chemical edge parsing and any self-loop filtering in loader.py =="
sed -n '120,240p' packages/quantum-nematode/quantumnematode/connectome/loader.py
echo
echo "== direct search for chemical self-loop checks =="
rg -n 'pre == post|a == b|self-loop' packages/quantum-nematode/quantumnematode/connectome/loader.pyRepository: SyntheticBrains/nematode
Length of output: 4686
Reject chemical self-loops here. ChemicalSynapse accepts pre == post, and the Cook loader doesnât filter them, so a fixture with a pre-existing self-loop can still be returned unchanged and violate the simple-graph contract.
ðĪ 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/connectome/rewiring.py` around
lines 149 - 154, The simple-graph validation in rewire_degree_preserving only
checks for duplicate/parallel edges, but it still allows chemical self-loops
through. Update the guard near the chem_edges/gap_edges validation to also
reject any ChemicalSynapse where pre == post, since the Cook loader can preserve
those fixtures unchanged. Keep the fix localized to rewire_degree_preserving and
ensure the ValueError is raised for any self-looped chemical edge before
rewiring proceeds.
Codecov Reportâ Patch coverage is ðĒ Thoughts on this report? Let us know! |
This was referenced Jul 5, 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
The credibility control for the connectome half of the Phase-6a synthesis. The continuous-substrate ranking (Logbook 029) placed the wild-type C. elegans connectome 5th of 6 under PPO weight search. This change answers the obvious question: is that its specific wiring, or just its degree / sparsity statistics?
A degree-preserving rewired-null (Dhiman 2026) â rewire the connectome so every neuron keeps its exact in/out degree but which neurons connect is scrambled â as a config option on the existing connectome brain (
wiring: wild_type | rewired_degree_preserving+rewire_seed), byte-identical when wild-type, matched initialisation (dedicated RNG) + budget. Full write-up: Logbook 034.Result â DEGREE-STATISTICS
Panel: wild-type vs rewired-null, n=8 paired seeds, 6000ep, same recipe/budget as the 029 connectome cell.
Paired delta (wild â rewired): d = â3.28, CI[â8.56, +1.61] (spans 0), BH-FDR q = 0.770 â indistinguishable (rewired nominally higher).
The connectome's ranking is a degree-statistics result, not a wiring result. Under PPO weight search â which re-optimises the chemical weights freely on whatever topology it is given â the specific C. elegans wiring confers no advantage over degree-matched alternatives (the Beiran & Litwin-Kumar 2025 degeneracy prediction). It does not say the wiring is functionless â its value is presumably the evolved fixed weights + specific I/O routing, not a better substrate for gradient descent.
A single-seed smoke hinted "specific wiring" (57.2 vs 46.5) but reversed at n=8 â a clean caution that connectome results must not be read off one seed.
Mechanism
connectome/rewiring.pyâ a seeded degree-preserving double-edge-swap (directed configuration-model for chemical synapses, undirected for gap junctions; no networkx dependency). On the real connectome: 91% of chemical edges moved, degree exactly preserved, mixes cleanly.ConnectomePPOBraingains thewiring/rewire_seedfields; the transform runs at the loadâtopology seam on a dedicated RNG so the weight-init stream is unperturbed (matched init vs wild-type for the same seed). Byte-identical whenwiring: wild_type(guarded by a test).scripts/analysis/connectome_structure_controls.pyâ paired wild-vs-rewired comparison reusing the committed 029 ranked metric + the paired-seed Wilcoxon / bootstrap / BH-FDR layer.Self-review
Ran a multi-angle review of the branch (edge-swap correctness, seam+harness, tests/conventions). The algorithm and the matched-init seam verified correct; applied fixes for a one-sided-Wilcoxon verdict-branch reachability bug, two load-bearing test gaps (matched-init assertion, rewired-brain-trains), and cheap robustness guards. None changed the verdict.
Scope
Ships the gating rewired-null; the learnable-gap-junction control (
T7.controls.learnable_gj) is a tracked fast-follow.Gates
openspec validate --strictvalid; change archived (spec requirement applied); fullpytest -m "not nightly"â 4090 passed; fullpre-commit run -aclean.ðĪ Generated with Claude Code
Summary by CodeRabbit
New Features
Documentation