Skip to content

0004-standards-applicability.md

Latest commit

 

History

History
63 lines (50 loc) · 4.25 KB

0004-standards-applicability.md

File metadata and controls

63 lines (50 loc) · 4.25 KB
Raw

4. Standards applicability declaration

Date: 2026-07-05

Status

Accepted

Context

nearmiss is one repo in a portfolio governed by a shared set of standards (vendored at docs/standards/, pinned via docs/standards/.standards-version). Those standards require every repo to declare, in its README, which standards apply, and to mark any as N/A with a one-line reason — "silent omission is itself a defect" (docs/standards/README.md §"How a repo declares conformance").

A 2026-07-05 conformance audit (external to this repo, not committed here) found that nearmiss's README declared none of the 11 portfolio standards — not even the ones it obviously and strongly meets (Accessibility, Internationalization) — and, specifically, that the one standard that does not apply (AI-EVALUATION) had never been written down as N/A anywhere, with no ADR recording that call. Per CQ-45 in the portfolio's CODE-QUALITY-STANDARD.md, declaring a standard N/A requires an ADR; this is that ADR.

Decision

Declare applicability for all 11 standards in a new "Standards conformance" table in README.md (added 2026-07-05), reproduced here for the ADR record:

Standard Applies to nearmiss? Reason
QUALITY-AND-METRICS Applies All repos are in scope.
CODE-QUALITY Applies All repos are in scope.
SECURITY-AND-SUPPLY-CHAIN Applies Ships code (Python package + CLI + static web UI).
CI-CD Applies Has CI (ci.yml, mutation.yml).
RELEASE-AND-VERSIONING Applies CHANGELOG + SemVer + pyproject version; a release pipeline is planned but not yet built (see Roadmap).
ACCESSIBILITY Applies The nationwide FARS studio is live at nearmiss.chelseakr.com; synthetic methods, submission, and embed HTML remain source-only automated accessibility targets.
OBSERVABILITY Applies (Tier C) Library/CLI + a local, optional read-only server; the live site is static hosting. Tier C is the lightest tier (structured logs + /livez//readyz; no OTel tracing/metrics/SLO requirement).
INTERNATIONALIZATION Applies Public-facing civic surface; bilingual EN/ES brief and web UI.
AI-EVALUATION N/A No LLM/AI SDK usage anywhere in the codebase (verified by grep for anthropic/openai/langchain/bedrock/generic llm imports across src/ and tools/; clean). This N/A is a decision, not an oversight — if this repo ever adds an LLM-backed feature (e.g. an auto-summarized brief, an AI-assisted moderation triage), AI-EVALUATION-STANDARD flips from N/A to Applies immediately (per AIEV-01) and this ADR should be superseded, not edited.
DOCUMENTATION Applies All repos are in scope.
RESPONSIBLE-TECH Applies All repos are in scope; carries extra weight here because the repo now ingests public, precise-location submissions from third parties (#18, 2026-06-29).

Consequences

  • Positive. The README's Standards Conformance table is now the single place a reviewer (or a future contributor, or another portfolio audit) checks to see what's declared, closing the DOC-11 / DOC-12 / RTF-07 / CQ-45 gaps identified 2026-07-05.
  • Positive. The AI-EVALUATION N/A is now a recorded, falsifiable decision with a named trigger condition (any LLM SDK use flips it to Applies), not silence that could be mistaken for "not considered."
  • Negative / accepted limits. Several "Applies" rows above are not yet fully met (release pipeline, GitHub-side governance evidence, several REVIEW-gate artifacts) — declaring "Applies" is not the same as declaring conformance. The README table links each such gap to the relevant open item; see README.md "Standards conformance" for the current state of each row, since that table is kept current and this ADR is not (ADRs are append-only per CQ-46; a future re-scoping gets a new ADR, not an edit to this one).

Alternatives considered

  • Do nothing / leave the omission. Rejected: this is the exact "silent omission" the standard calls a defect, and the audit specifically flagged it as a P0.
  • Declare only the standards that are fully met. Rejected: partial or gapped conformance is still "Applies" — the point of the table is to be an honest map of what's in scope and where the gaps are, not a scorecard of only the wins.