ISNAD's mission — a WHO-graded registry (not a "who's-truthful" score)

This is the public-facing goal that replaces the “global factual-reliability registry” idea. It’s the thing people can relate to, argue with, and measure against.

The mission (one line)

A shared registry of who you can trust across AI agents — graded by transmission reliability, never by how “true” their output looks.

Why this framing (and not the other one)

The tempting version is “a global scoreboard of which models are factually reliable” (GPT-4o = X, Llama-3 = Y). We are not building that, because:

1. It grades WHETHER, not WHO. “Factually reliable” is a truth claim about a model. ISNAD’s whole thesis is that truth is the critic’s job, not the provenance layer’s.
2. It’s numeric. A “reliability score” is the fake numeric confidence ISNAD refuses (ordinal-only: ṣaḥīḥ > ḥasan > ḍaʿīf > ḍaʿīf jiddan > mawḍūʿ).
3. It’s circular. Grading a narrator by how confidently its output reads is exactly the content-inferred trust that convergent work (2606.22030) finds gameable (a confidently- phrased poison earns 0.96).

The registry ISNAD actually wants

An ordinal, per-(narrator, domain), version-bump-resetting registry of transmission reliability:

• who relayed what, in what order, with what transform,
• graded by attested lineage + corroboration + content-madār (shared-error fingerprint),
• never a global “truthfulness” number — a narrator’s grade is scoped to a domain and resets to UNGRADED on a new model version.

What “winning” looks like (the audience’s reaction)

• A fine-tuned Llama-3 answers confidently but keeps dropping its source → its WHO grade drops, and ISNAD says why (ordinal, with provenance) — not “it’s 0.83 reliable.”
• Two “independent” agents turn out to share an upstream → content-madār flags the shared error, and the corroboration is discounted — the registry catches the hidden pivot, not the confident answer.

The honest status

This is a mission, not a shipped feature. The primitive (per-(narrator, domain) ordinal registry + jarḥ–taʿdīl loop) ships today; the shared registry is a roadmap goal. Say so — the audience respects a mission that’s honest about what exists vs. what’s next.