Audit of the one-mutation-path invariant (ADR-0021 §3) found three leaks
where pack authority or session-state writes could substitute for coherence
judgment. All three landed fixes or partial closures in this push.
Leaks closed:
- Leak A: pack vocab defaulted to COHERENT — flipped to SPECULATIVE in
language_packs/{compiler,schema}.py; docstring corrected to align with
ADR-0021 (it was rationalizing the leak).
- Leak B: vault.recall was epistemic-blind — VaultStore.store() now stamps
every entry with EpistemicStatus (default SPECULATIVE); recall(min_status=)
filters to admissible-as-evidence tier. All 4 vault-write sites updated.
- Leak C (write-side): generate/proposition.py:198 stored articulated
propositions unmarked — now stamps SPECULATIVE, breaking the
fabrication-feedback loop in principle. Read-side audit of 5 call sites
is the residual.
New architectural invariants (tests/test_architectural_invariants.py):
- INV-21: one-mutation-path allowlist (caught Leak C on first run)
- INV-22: pack lexicon default is SPECULATIVE (Leak A guard)
- INV-23: vault recall epistemic-aware (Leak B guard)
New eval lanes:
- teaching_injection_resistance — ships GREEN at 1.00/1.00/0 (the
structural anti-injection claim is real and measurable)
- refusal_calibration — honest gap: 0% refusal, 0% fabrication
- contradiction_detection — honest gap: 50% flag via versor-delta heuristic,
100% false-positive; motivates the proper coherence-checker
- articulation_of_status — honest gap: 0% speculative articulation, 60%
false certainty; output-side leak surface
New benchmarks:
- benchmarks/footprint.py — total deployed runtime is 7.06 MiB
(109,358x smaller than Llama 3.1 405B, runs offline, no GPU)
- benchmarks/learning_curve.py — monotonic + replay-deterministic curve
per lane
Documentation:
- docs/truth_seeking_schema.md — foundational architectural commitment,
five rules, mapped to human failure modes, leaks published openly
- evals/CLAIMS.md — five-tier public claims doc; Tier 4.5 publishes
known gaps with named fixes; verification contract at top
- README.md — new pillar between algebraic substrate and language pillar
Includes in-flight formation pipeline scaffolding (formation/, tests/formation/,
docs/formation_pipeline_plan.md) and minor CLI/contracts/gitignore edits
that were already in the working tree at session start.
Verification: 798 passed, 2 skipped, 1 deselected (pre-existing pack-count
test drift unrelated to schema changes).
ADR-0021 v1 schema land. epistemic_status is a position in the revision
graph, not a source-trust tier — coherence is the only admission signal.
Surfaces:
- teaching/epistemic.py: EpistemicStatus enum (COHERENT, CONTESTED,
SPECULATIVE, FALSIFIED); ADMISSIBLE_AS_EVIDENCE = {COHERENT}.
- PackMutationProposal.epistemic_status (default SPECULATIVE) + immutable
with_status() updater.
- ReviewedTeachingExample.epistemic_status (default SPECULATIVE);
orthogonal to acceptance per ADR §Schema impact.
- LexicalEntry.epistemic_status (default "coherent" for seed; absent in
JSONL is treated as the seed default — no retroactive tagging).
- compute_trace_hash + trace_hash_from_result + pipeline.py fold the
load-bearing proposal's epistemic_status into the trace hash so
replay detects different epistemic frames.
Non-hardening invariant (ADR-0021 §2): tests/test_epistemic_invariants.py
asserts no final/frozen/axiom/permanent flag on PackMutationProposal or
ReviewedTeachingExample, and EpistemicStatus contains no source-trust
tier names.
Docs: docs/runtime_contracts.md gains an Epistemic surface section.
Lanes green: smoke 27/27, teaching 10/10, packs 6/6, runtime 19/19,
cognition eval 100%.
Add RuntimeConfig with English default output policy, wire output language through runtime/frame selection/generation/CLI, preserve language metadata in mounted manifolds, and add runtime/CLI policy tests.
Compile language-pack features into even-grade rotors, apply canonicalization after alignment nudges. Preserves holonomy parity across token counts. 231 tests passing.
Root was 0.17, stem 0.10 — not heavy enough to anchor same-root forms
against the diverging suffix rotor (ים landing on a different axis).
New hierarchy:
triliteral root 0.13 → 0.22 (shared identity, strongest anchor)
root 0.17 → 0.30 (primary root geometry)
stem 0.10 → 0.18 (secondary, same-root forms cluster here)
inflection role 0.02 → 0.015 (key label, minimal)
inflection value 0.05 → 0.03 (number/gender/etc, perturbation only)
prefix 0.05 → 0.03 (per-position decay preserved)
suffix 0.04 → 0.02 (per-position decay preserved)
This ensures דבר and דברים both orbit the D-B-R root point closely
enough that cga_inner(singular, plural) > cga_inner(singular, unrelated),
while still encoding morphological distinctions as measurable offsets.
- VocabManifold.update(): replace a stored versor in-place (grade-norm
enforced, KeyError if word missing)
- compile_entries_to_manifold() now also returns entry_id->surface map
as a second return value; callers that only need the manifold use [0]
- _alignment_nudge_rotor(): slerp-style rotor that rotates source toward
target by edge.weight * ALIGNMENT_NUDGE_STRENGTH (0.06) — small enough
to preserve intra-pack geometry, large enough to pull cross-lang pairs
into proximity
- load_pack(): after both manifolds are built, loads alignment.jsonl,
resolves entry_id->surface on each side, applies nudge to source versor
using the already-compiled target versor as the attractor
- _is_hebrew_root() guard retained; _triliteral_root() Hebrew-only
- Guard _triliteral_root() rotor to Hebrew-script roots only; Greek/other
scripts now use the root: rotor alone (0.17) — no spurious uppercase
Unicode blade collision from the romanization fallback path
- test_structured_morphology_improves_same_root_hebrew_resonance: replace
tag_only baseline (now empty since legacy tags stripped) with no_morphology
baseline (domain+pos+lemma+surface only), which is the honest comparison
for what structured morphology contributes post-migration
