* chore(kernel): tighten construction_affordances catalog invariants
- Delete stale make_proposal (deferred from PR #919; no live callers remain)
- Derive _PROPOSAL_FIRST_FAMILIES from _CATALOG keys to eliminate hidden
sync obligation (was a manual restatement of all four catalog entries)
- Introduce _SERVING_AUTHORIZED_FAMILIES frozenset so the serving-allowed
exemption in ConstructionFamily and ConstructionProposal __post_init__
consults one central authority instead of two string-literal comparisons
- Pre-compute _SORTED_FAMILIES at module load so all_diagnostic_families()
returns a constant rather than re-sorting on every call
- Remove _VALID_STATUS ClassVar from ConstructionProposal; the field is
already typed Literal["proposed"] and the indirection adds no safety
No behavior change to serving, assessment, or proposal creation paths.
ADR-0236: small load-bearing PR, single-file, invariants as code.
* refactor(contemplation): unify organ pipelines in pass_manager
- Introduce OrganPipeline descriptor (organ, reader, solver, reason_adapter)
so every routed solve path shares one _solve_and_verify implementation.
Adding a new organ is now a single OrganPipeline(...) declaration.
- Extract _maybe_ask helper that collapses the triply-duplicated
family_by_name / assess_from_family / exercise_ask inline block.
- Add _result() helper that appends the terminal Finding and constructs
ContemplationResult in one call, enforcing terminal-is-last-finding
by construction across every return site.
- Replace assert not isinstance(problem, Refusal) with
InternalContemplationError so the routing invariant is enforceable
under python -O and surfaces a named, debuggable exception.
No behavior change: all Terminal states, Finding sequences, family
resolution, CMB-over-R3 precedence, boundary-first ordering, and
ask/proposal emission logic are preserved exactly.
Extracts the reasoning protocol demonstrated in the contemplation
refactor session into a canonical, reusable guide in AGENTS.md.
- New section: Reasoning and Problem-Solving Discipline (7 steps)
- Covers: read-first, shape-finding, leverage ranking, precise
enumeration, claims-grounded proof, cognitive model connection,
and commit discipline
- Documents the failure modes each step prevents
- CLAUDE.md + GEMINI.md: add pointer to the new protocol section
* fix: final cleanups for CGA substrate migration PR
- Update UI test fixtures to packs/data paths
- Clean CLAUDE.md/GEMINI.md to minimal per AGENTS.md
- Refactor fraction geometric construction into dedicated helper in problem_frame_contracts.py
- No TODOs left in code; notes in PR description
This completes the mechanical readiness for the CGA substrate + registry consolidation work.
* docs: include PR description for review
* chore: enforce forgejo tooling in AGENTS.md
* docs: update bootstrap skill and add git-forgejo config doc per new rules
* docs: remove deprecated git-forgejo wrapper docs
* chore: ignore local .mcp.json configuration file
* fix(ui): remove unsupported assessment.bindings usage to fix TS2339/TS7006 in constructionEvidencePanelModel
Proves topological cost neutrality across identity packs: the Cl(4,1)
versor sandwich product (F <- M * F * ~M) compiled into a single MLX
Metal kernel incurs equal VRAM and latency regardless of whether the
PersonaMotor encodes precision_first vs generosity_first character.
The periodic mx.eval() boundary at every 50th step mirrors the async
token-yielding backpressure of ChatRuntime and confirms that the lazy
MLX computation graph is flushed safely, keeping Active VRAM Delta ~0.
Correctness notes:
- PersonaMotor.apply() is a NumPy function (versor_apply from
algebra.versor). The MLX compiled_field_step wraps the sandwich
arithmetic in pure MLX, structurally mirroring what versor_apply does
so that the Metal kernel fusion path is exercised without re-coupling
this benchmark to the NumPy implementation.
- IdentityManifold has no load_pack() classmethod; identity packs are
constructed directly from ValueAxis instances as the dataclass defines.
- MLX is a soft/optional dependency following the apple_uma_mlx_exact_recall
convention: imported lazily, skipped gracefully when unavailable.
This change removes duplicate provider-specific rules across CLAUDE/GPT55/GROK files and replaces them with thin, provider-neutral shims that point back to the canonical AGENTS.md for all architectural invariants, PR discipline, and trust boundaries.
* feat(bench): add Apple UMA demo package builder
* test(bench): cover Apple UMA demo package builder
* docs(outreach): add Apple UMA demo package runbook
* feat(bench): harden Apple UMA demo package workflow
* feat(workbench-ui): add Apple UMA report card route
* feat(workbench-ui): register Apple UMA route
* feat(workbench-ui): expose Apple UMA route path
* feat(workbench-ui): wire Apple UMA report card into route registry
* test(workbench-ui): fix UI and keyboard tests for Apple UMA route integration
* feat(workbench): add Apple UMA report read model
* test(workbench): cover Apple UMA report projection
* feat(workbench): expose Apple UMA report route
Replace extract_f32_slice list conversion on geometric_product, cga_inner,
versor_condition, versor_apply_with_closure, and versor_apply_with_closure_f64
with PyReadonlyArray1 zero-copy views. Wrong shape, dtype, and non-contiguous
layouts fail loudly. Update Apple UMA benchmark truth table and claim audit
after parity gates pass.
* feat(workbench): route practice evidence through trace endpoint shim
* test(workbench): lock trace practice API route
* fix(workbench): remove stray practice API test token
* feat(workbench): add practice evidence read model
* feat(workbench): add practice evidence endpoint seam
* feat(workbench): persist practice evidence field when present
* test(workbench): cover practice evidence read model seam
Expand the Trace Construction panel from a summary scaffold into a read-only evidence inspection surface.
Adds detail sections for construction proposals, mentions, mention bindings, bound relations, contract assessments, and source-span exactness rows. Preserves the proposal/contract authority boundary: proposals remain diagnostic and contract assessments determine runnable/blocker disposition.
Validation: workbench-ui, smoke, and lane-shas GitHub Actions passed on PR head d20d143b5744d15d73d07a515b466143af6a2d52.
