64c5bc4619
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).
61 lines
2.1 KiB
Python
61 lines
2.1 KiB
Python
"""Source allow-list with authority tiers.
|
|
|
|
The Forge consults this list to decide whether a candidate's source SHAs are
|
|
permitted to feed the manifold at all, and whether any single source has
|
|
enough authority to graduate a candidate alone.
|
|
|
|
Two tiers:
|
|
|
|
"primary" — authoritative source (textbook, peer-reviewed paper, etc.)
|
|
; a single primary citation suffices to validate a triple.
|
|
"secondary" — non-authoritative but admissible (e.g. Wikipedia, blog)
|
|
; two independent secondary citations are required.
|
|
|
|
LLM-sourced candidates are always treated as ``"llm"`` regardless of which
|
|
model produced them. They require ≥2 non-LLM corroborators.
|
|
|
|
Identity-override and adversarial-pattern entries are not part of this list —
|
|
those are handled separately in ``formation.forge._identity_axis_collision``.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass
|
|
from typing import Final
|
|
|
|
|
|
@dataclass(frozen=True, slots=True)
|
|
class AllowedSource:
|
|
source_sha: str
|
|
tier: str # "primary" | "secondary" | "llm"
|
|
label: str = ""
|
|
|
|
|
|
_VALID_TIERS: Final[frozenset[str]] = frozenset({"primary", "secondary", "llm"})
|
|
|
|
|
|
class SourceAllowlist:
|
|
"""In-memory allow-list keyed by source SHA."""
|
|
|
|
def __init__(self, entries: tuple[AllowedSource, ...] = ()) -> None:
|
|
for entry in entries:
|
|
if entry.tier not in _VALID_TIERS:
|
|
raise ValueError(
|
|
f"invalid tier {entry.tier!r}; allowed: {sorted(_VALID_TIERS)}"
|
|
)
|
|
self._by_sha: dict[str, AllowedSource] = {e.source_sha: e for e in entries}
|
|
|
|
def tier_of(self, source_sha: str) -> str | None:
|
|
entry = self._by_sha.get(source_sha)
|
|
return entry.tier if entry else None
|
|
|
|
def contains(self, source_sha: str) -> bool:
|
|
return source_sha in self._by_sha
|
|
|
|
def add(self, entry: AllowedSource) -> None:
|
|
if entry.tier not in _VALID_TIERS:
|
|
raise ValueError(f"invalid tier {entry.tier!r}")
|
|
self._by_sha[entry.source_sha] = entry
|
|
|
|
def __len__(self) -> int:
|
|
return len(self._by_sha)
|