97971bd636
Introduces teaching/ module with three-stage correction pipeline: 1. correction.py — extracts CorrectionCandidate from correction intents, binding correction text to the prior turn it references 2. review.py — validates candidates: rejects identity overrides (17 marker patterns) and empty corrections; produces ReviewedTeachingExample with deterministic SHA-256 review hash 3. store.py — bounded FIFO store for accepted examples; emits PackMutationProposal objects instead of mutating the vocab manifold directly; retrievable by subject Design invariants: - Identity override attempts are rejected at the review gate - Pack mutations are proposal-only (applied=False by default) - All traces are deterministic: same input → same candidate_id and review_hash Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
27 lines
1 KiB
Python
27 lines
1 KiB
Python
"""teaching — correction capture, review, and proposal-only pack mutation.
|
|
|
|
The teaching loop allows CORE to learn from corrections in a controlled,
|
|
auditable way. Corrections flow through three stages:
|
|
|
|
1. Capture — extract a CorrectionCandidate from a correction intent
|
|
2. Review — validate the candidate (identity-safe, bounded, deterministic)
|
|
3. Store — persist reviewed examples; propose pack mutations without applying
|
|
|
|
Identity overrides are rejected at the review stage. Pack mutations are
|
|
emitted as proposals (PackMutationProposal) that require explicit external
|
|
approval before they touch the vocabulary manifold.
|
|
"""
|
|
|
|
from teaching.correction import CorrectionCandidate, extract_correction
|
|
from teaching.review import ReviewedTeachingExample, ReviewOutcome, review_correction
|
|
from teaching.store import TeachingStore, PackMutationProposal
|
|
|
|
__all__ = [
|
|
"CorrectionCandidate",
|
|
"extract_correction",
|
|
"ReviewedTeachingExample",
|
|
"ReviewOutcome",
|
|
"review_correction",
|
|
"TeachingStore",
|
|
"PackMutationProposal",
|
|
]
|