The normative documentation lives directly in the sm-abuse-report repository, versioned alongside the code that implements it — plus one standalone research paper arguing the case at length. The repository is private; links below resolve for accounts with access.
The Fourth Tense — full paper
Accountable Accusations for the Internet of AI Agents. The complete argument: related work, the report object, verification, threat model, limitations, and open questions.
↗ SPECSPEC.md — the wire format
The normative eleven-field report, category taxonomy, canonicalization, signature scheme, and chain and citation verification. v0.1, working draft.
↗ WHITEPAPERWHITEPAPER.md — the argument
Why an accusation needs its own signed record, why it's bidirectional, and why adjudication is deliberately out of scope.
↗ THREAT MODELTHREATMODEL.md — what it defends, and what it doesn't
Field tampering, structural forgery, and chain tampering, against the stated boundaries: the accusation itself, unconfirmed citations, reporter equivocation, and report spam.
↗ EXAMPLEfile_and_verify.py — a runnable walkthrough
File a report, verify it, check its citation, and detect a tamper, in plain Python.
↗ CONFORMANCEconformance/vectors — the test corpus
Eleven checked-in vectors covering valid reports, tampered fields, chain defects, and citation outcomes, replayed byte-for-byte against the reference implementation.
↗Reference implementation and full source: github.com/Sharathvc23/sm-abuse-report. Not yet published to PyPI — install directly from the repository.