style+fix: conform to the fleet's black and ruff gates (B and C)

[stranske]

Conforms this repo to the fleet's format and lint gates. B and C only — pyproject.toml (A) is held pending the src/ layout question.

Two commits, deliberately separate:

commit	what	review effort
4117aa2	black --line-length 100 — 126 files, purely mechanical	skip it; it's the formatter's output
14ccb6d	61 ruff findings, incl. 3 real defects	this is the one to read

Why these were failing

pr-00-gate.yml calls stranske/Workflows/.github/workflows/reusable-10-ci-python.yml@main. Its two relevant jobs run, verbatim:

black --check --line-length 100 --exclude '(\.venv|\.workflows-lib|node_modules)' .   # line 690
ruff check --select E4,E7,E9,F --output-format github --extend-exclude .workflows-lib .   # line 1155

The ruff line is the no-explicit-selection fallback — pyflakes plus serious pycodestyle, i.e. real defects rather than style. Both now pass.

A measurement correction worth recording

My first pass reported 914 ruff errors and 180 files needing black, and concluded the gate was unreasonable. Both numbers were artefacts of running the tools bare: with no pyproject.toml present they fall back to their own defaults (line length 88; the full ruff rule set), not to what CI runs. At CI's real settings it was 126 files and 79 findings — and after black split the 18 E702 semicolons, 61.

Take the command line from reusable-10-ci-python.yml, never from the tool default.

What black actually changes, measured

Code only: tuple/list literals exploded one per line, implicit string concatenation re-indented, a; b split. It also collapses the column-aligned padding before trailing # comments to two spaces, which this tree used extensively — 244 comment-bearing lines are byte-identical after whitespace normalisation. No comment or docstring prose is reworded or reflowed anywhere. (I initially claimed "0 comment lines touched" from a single-file sample; the inline-alignment collapse is real, and cosmetic.)

black verifies AST equivalence on every file it rewrites.

The three real defects ruff found

1. test_observability_activation.py set "route_weights" twice in one dict literal (lines 52 and 90). Same value, so no behaviour changed — but one silently overrode the other.
2. verify.py unpacked fc, fp in verify() while only fc is used — a leftover from the refactor that moved the passed-floor comparison into _floor_problems. Removed there, kept in _floor_problems, which genuinely uses both.
3. capability_admission.py computed cited = {d["record"] for d in com["dangling_citations"]} and asserted nothing on it. This repo's founding defect in miniature: a value computed for a check that no longer checks. The dead binding is removed; the missing assertion is not invented here — that needs a decision about what it should assert.

Deliberately not repaired

keepalive_evidence.py built haystack = f"#{issue_number} {issue_title}" and then searched issue_title alone — so a bare #N reference in a title is never matched. Wiring haystack in would change what the keepalive evidence path matches, and that is Brain input (CLAUDE.md §2). A behaviour change has no business riding along in a lint pass. Only the unused binding is removed, which changes nothing.

Also load-bearing and preserved: in exp_abcd.py the lambda→def conversion keeps the default-arg capture repo=meta["repo"], exp_id=edir.name verbatim in the signature — it binds this iteration's values so a later loop turn cannot rebind them.

Method note

My first edit replaced the wrong fc, fp = ... occurrence and broke _floor_problems. ruff --select F821 named it immediately — "Undefined name fp", three sites. That is the argument for this gate existing rather than being configured away.

Test gate

python3 verify.py — 368 passed, 0 failed, 0 skipped, 83/83 selftests, 43/43 can-fire, 5/5 gates. Floor and ceilings untouched: nothing was added, removed or renamed.

Both gates re-run exactly as CI invokes them: black --check → 192 files unchanged; ruff check --select E4,E7,E9,F → All checks passed.

Still red after this, and why

python 3.12 / 3.13 / typecheck-mypy need pyproject.toml (CI passes --cov-config=pyproject.toml, which currently raises ConfigError). That is commit A, held pending the src/ layout decision.

🤖 Generated with Claude Code

[coderabbitai]

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your current included review allowance is based on your included PR review attempts over the past 7 days.

Next review available in: 5 minutes

Limit details: You’ve used the included review currently available. Your 74 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab.

How can I continue?

Wait for the limit to reset, then comment @coderabbitai review or push new commits to the PR.

An organization admin can change what happens after included review limits in Billing.

How do review limits work?

CodeRabbit enforces per-developer PR review limits within each organization.

For paid Pro and Pro+ reviews, CodeRabbit uses a developer's included PR review attempts over the past 7 days to set the current hourly allowance. At typical activity levels, the full plan allowance applies. Higher sustained activity can lower the allowance until earlier attempts leave the 7-day window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: b475fb8f-810c-40a9-bd53-02dd95095122

📥 Commits

Reviewing files that changed from the base of the PR and between af6654d and 14ccb6d.

📒 Files selected for processing (126)

• adapters.py
• adversarial.py
• agent_auth_check.py
• backlog.py
• cadence_registry.py
• capabilities.py
• capability_activation_audit.py
• capability_admission.py
• capability_advisor.py
• capability_compiler.py
• capability_effectiveness.py
• capability_firing_monitor.py
• capability_ir.py
• capability_lifecycle.py
• capability_matcher_proposals.py
• capability_opportunity.py
• capability_outcome_bridge.py
• capability_propensity.py
• capability_recurrence_check.py
• capability_targets.py
• capacity.py
• ccusage_reconcile.py
• claims.py
• codemod_lane.py
• completion_event_adapter.py
• consumer_sync_artifact_ingest.py
• consumer_sync_shadow.py
• cross_repo_lane.py
• dispatcher.py
• dry_seam_audit.py
• durability_sweep.py
• env_prereq.py
• epic_lane.py
• evidence_acquisition.py
• evidence_schema.py
• execution_profiles.py
• exp_abcd.py
• experiment_recovery.py
• exploration_backfill.py
• exploration_collection.py
• exploration_evidence_plan.py
• exploration_review.py
• feature_scan.py
• features.py
• feedback.py
• frontend_verify.py
• gh_capacity.py
• human_calibration.py
• issue_quality.py
• issue_readiness.py
• judge_reliability.py
• keepalive_evidence.py
• keepalive_outcomes.py
• keepalive_shadow.py
• keepalive_supervisor.py
• langsmith_direct.py
• langsmith_fetch.py
• langsmith_pull.py
• ledger_reconcile.py
• local_verify.py
• mcp_server.py
• merge_guard.py
• model_profile_trial.py
• model_profile_trial_bridge.py
• objective_anchor.py
• observability_dashboard.py
• outcomes.py
• partitioned_review.py
• pattern_miner.py
• periodic_report.py
• provision.py
• range_lane_rollout.py
• redirect_apply.py
• redirect_plan.py
• redirect_policy.py
• redirect_shadow.py
• redirect_sweep.py
• relearn_report.py
• repo_knowledge.py
• research_scheduler.py
• research_subjects.py
• roles.py
• router.py
• runner_effect_bridge.py
• runtime_ac.py
• runtime_ac_flow_monitor.py
• runtime_ac_gate.py
• runtime_ac_panel.py
• strategy_experiment.py
• switch_review.py
• synthesis_promotion.py
• test_capabilities.py
• test_capability_admission.py
• test_capability_causal_core.py
• test_capability_epic.py
• test_capability_lifecycle_e2e.py
• test_capability_set_coverage.py
• test_capacity_profiles.py
• test_completion_events.py
• test_consumer_sync_artifact_ingest.py
• test_consumer_sync_shadow.py
• test_evidence_contract_compiler.py
• test_experiment_arm_identity.py
• test_feedback_model_provenance.py
• test_model_profile_trial.py
• test_model_profile_trial_bridge.py
• test_model_tier_resolution.py
• test_observability_activation.py
• test_partitioned_review.py
• test_pattern_miner.py
• test_playbook_compiler.py
• test_research_control.py
• test_role_compiler.py
• test_roles_lineage.py
• test_runner_effect_bridge.py
• test_runtime_ac_flow_monitor.py
• test_skill_compiler.py
• test_synthesis_promotion.py
• test_ux_review.py
• test_workflow_compiler.py
• testgen_gate.py
• testgen_lane.py
• tick.py
• ux_review.py
• verify.py
• watch.py

---

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

[stranske]

Closing — superseded by #60, which has the root cause this PR does not.

I built this from a local measurement without checking open PRs or the improvement log first. #60 got there properly, and its diagnosis invalidates the premise of this PR: all five upstream jobs were dying at a shared install step —

Error: .github/workflows/autofix-versions.env is required; refusing to install unpinned tooling.

Ruff, Black and mypy had never executed on this repo, not once. Every statement about its lint debt, mine included, was inferred. #60 adds the missing pin file, plus ruff.toml + mypy.ini, docs/CI_LINT_BASELINE.md, and toggles that state blocking/drainable/drains-by with a test enforcing it.

It also explains why my 126-file black pass was largely wasted: the repo self-heals formatting via its own autofix workflow (chore(autofix): formatting/lint, adfa5f4 / 9239e06). What autofix leaves behind is exactly the non-auto-fixable residue — the E741/E731/F841/E402/F601 judgement calls — and #60 drains those too, under a ruff.toml selecting ["E4","E7","E9","F","I"] that collapses the Gate/Autofix two-window disagreement into one set. That last part is the real fix and this PR does not have it.

One thing from here worth carrying over if it is not already in #60, since these were verified against live code:

• test_observability_activation.py sets "route_weights" twice in one dict literal (lines 52 and 90) — same value, so nothing behaves differently, but one silently overrides the other.
• capability_admission.py computes cited = {d["record"] for d in com["dangling_citations"]} and asserts nothing on it — a value computed for a check that no longer checks. Filed as a chip; the missing assertion needs a decision, not a deletion.
• keepalive_evidence.py builds haystack = f"#{issue_number} {issue_title}" then searches issue_title alone, so a bare #N in a title is never matched. Deliberately not repaired — it changes what the keepalive evidence path matches, which is Brain input (CLAUDE.md §2). Filed as a chip.

And a correction to my own earlier claim on #58: the --cov-config=pyproject.toml failure is real, but adding pyproject.toml is not the fix. reusable-10-ci-python.yml appends -e '.[app,dev]' at eight sites gated only on [ -f pyproject.toml ], so the file's mere existence makes all five jobs attempt an editable install — and 126 flat root modules have no package and no app/dev extras. It would move the failure earlier, to the install step. #60's ruff.toml + mypy.ini is the correct shape for a non-package repo.

🤖 Addressed by Claude Code