Skip to content

v165.6.1 — PaperGenerationProvenanceReceipt

Choose a tag to compare

@EmergentMonk EmergentMonk released this 19 May 00:47
· 451 commits to main since this release
d5eb108

Release Notes — v165.6.1
v165.6.1 — PaperGenerationProvenanceReceipt
Added

Introduced deterministic provenance receipts for generated research documents, papers, preprints, and publication-bound research artifacts.

New module:

src/qec/analysis/paper_generation_provenance_receipts.py

New artifacts:

GeneratedDocumentIdentity
GenerationSessionReference
CitationBoundaryReference
ReviewBoundaryReference
DocumentClaimInheritance
PublicationIntentDeclaration
PaperGenerationProvenanceReceipt

New aggregate identity:

paper_generation_provenance_receipt_hash
Deterministic Provenance Guarantees

v165.6.1 establishes deterministic contracts for:

document provenance lineage
publication-boundary governance
review-boundary inheritance
citation-boundary inheritance
claim inheritance
replay-safe paper provenance
publication-intent governance

The receipt layer now:

recomputes all hashes deterministically
validates all child artifacts before aggregate validation
validates upstream manifest lineage
enforces review/citation boundaries
rejects publication escalation
rejects hidden scientific-authority semantics
rejects autonomous-publication semantics
Publication Boundary Law

v165.6.1 formalizes:

A generated paper is not evidence.

Generated documents:

are not automatically correct
are not automatically peer reviewed
are not automatically publication-ready
are not automatically reproducible
are not automatically authoritative

The receipt records:

provenance lineage
review lineage
citation lineage
claim inheritance
publication boundaries

The receipt is the boundary.

Validation Hardening

Added deterministic enforcement for:

INVALID_HASH_FORMAT
HASH_MISMATCH
publication-boundary enforcement
claim inheritance enforcement
review-boundary enforcement
citation-boundary enforcement

Canonical validation now rejects:

publication laundering
review-state escalation
symbolic-only claim escalation
undeclared publication authority
autonomous-publication semantics
hidden scientific-authority semantics
Provenance Chain Semantics

provenance_chain_complete is now:

recomputed deterministically
not trusted from input
gated by review boundaries
gated by citation boundaries
gated by claim inheritance
gated by publication intent
gated by upstream manifest lineage

This prevents:

incomplete provenance escalation

through forged publication receipts.

Claim Inheritance Law

Claim scope inheritance MUST remain bounded.

If upstream provenance is:

SYMBOLIC_ONLY

the downstream document may NOT silently escalate to:

publication authority
benchmark authority
peer-review authority

without:

completed review boundaries
explicit provenance continuity

This blocks:

epistemic escalation drift

inside generated research chains.

Tests

Added:

tests/test_paper_generation_provenance_receipts.py

Coverage includes:

canonical hash stability
deterministic provenance ordering
publication_allowed recomputation
provenance_chain_complete recomputation
review-boundary enforcement
citation-boundary enforcement
claim inheritance enforcement
malformed hash rejection
forbidden runtime semantics rejection
immutable payload validation
PYTHONHASHSEED replay stability
decoder-boundary enforcement
autonomous-publication rejection

Validation completed successfully:

PYTHONPATH=src pytest -q tests/test_paper_generation_provenance_receipts.py

PYTHONPATH=src pytest -q -ra

→ 17973 passed
Scope Boundary

v165.6.1 does NOT:

generate papers
verify theorem correctness
perform autonomous publication
execute inference
browse the internet
replace peer review
modify src/qec/decoder/
modify dependency/package metadata

This release establishes deterministic publication provenance receipts only.

The decoder remains untouched.

Files Added
src/qec/analysis/paper_generation_provenance_receipts.py
tests/test_paper_generation_provenance_receipts.py
Gremlin Notes (Non-Canonical)

publication-laundering goblin attempted to convert symbolic-only outputs directly into publication authority and was denied escalation privileges.
peer-review goblin attempted marking NO_REVIEW artifacts as publication-ready and was destroyed by provenance-chain recomputation law.
citation-boundary goblin attempted silently inheriting inaccessible references into authoritative claims and was trapped behind citation-boundary enforcement.
truth-paper goblin attempted inserting “AI proved theorem” into canonical provenance payloads and was exiled by forbidden semantic validation.
autonomous-journal goblin attempted granting papers self-validating publication authority and was denied all constitutional standing.
coffee-break goblin attempted validating only aggregate provenance receipts while ignoring upstream manifest lineage and was forced to recursively validate the entire epistemic publication graph.

Extra Gremlin Notes:

"Why the Gremlin notes?" See: Where the Goblins Came From - https://openai.com/index/where-the-goblins-came-from/
"It's also funny that's why!"