Protocol module
if.story lint rules
Guardrails for publishing narrative logs: fail fast on missing traps, missing evidence links, leaked secrets, bad URLs, and unknown product IDs.
Black/white: these rules define checks; they do not claim a deployed runtime or enforcement exists.
Black / white
Verified (about the checks)
- The lints below are concrete and implementable as publish-time checks.
- They prevent the most common trust failures: overclaiming, leakage, and link drift.
Not verified / out of scope
- No proof of truth, correctness, safety, or intent.
- No deployed runtime or performance guarantees.
Publish-blocking lints
Fail the publish if any of these are violated.
Missing trap (required)
Fail if the narrative or JSON metadata lacks a defensive trap section/object with: risk, do_not, safe_path, evidence_links.
Missing evidence_links (required)
Fail if there are no stable public links a third party can fetch to review context (one URL per line in Markdown).
Secrets leakage
Fail if artifacts contain credentials, private keys, session cookies, or other sensitive content.
URL hygiene
Fail if a single line contains multiple https links, if canonical links include querystrings, or if links point to internal-only infrastructure.
Unknown product_ids
Fail if product IDs in metadata do not exist in the published registry (`/llm/if.registry.json`).
Forbidden overclaim wording
Fail if the text implies guarantees we don’t verify (truth, safety, compliance). Use QUANTUM READY only when a PQ receipt exists.