Status: Accepted
Date: 2026-07-06

Context

Six GOVERNANCE-themed known-issues had accumulated about the integrity of the
.context/ and ADR tooling: the adr-index gate checked existence not freshness;
check-undocumented-decisions.sh silently skipped any file mentioning
index.yaml; generated remediation plans omitted the now-required value field;
the frontmatter schema’s additionalProperties: false was not enforced by a real
validator; effort used single-letter values against the UPPER_CASE-word
convention (ADR-050); and known-issues themselves had no enforcement gate. A
verification pass confirmed all six were still live against current source. A
3-reviewer plan-review then examined a single consolidated remediation plan.

Decision

1. Fix by value, not in one batch. The six issues are split by their own
   value grades rather than bundled. The three quick, genuinely-valuable gaps
   (adr-index freshness, undocumented-decisions false-negative, remediation-plan
   value — all individually value: MEDIUM/S-effort) go in
   governance-tooling-hardening-2026-07-06.md. The two value: LOW,
   high-blast-radius gaps (effort full-word enum migration; a real JSON-schema
   validator) go in a separate deferred plan,
   governance-enum-and-schema-hardening-2026-07-06.md.

2. Do not gate valuable work behind low-value work. The consolidated plan
   originally sequenced the LOW-value effort-enum migration as a hard prerequisite
   to the valuable remediation-plan value fix. Two independent reviewers
   (Strategic, Risk) flagged that a blended value: MEDIUM hid a LOW majority-
   effort tail and that the coupling blocked a live fix. The split removes the
   coupling and keeps each plan sorting honestly under the read protocol.

3. Known-issues enforcement stays rejected (reaffirms the deferral). The
   known-issues-lack-enforcement gap is a settled do_not_proceed_for_now
   design-debate verdict — self-assigned severity, creation-date cannot detect
   neglect, a wall-clock expiry gate is non-deterministic, and any such gate is
   trivially silenceable. It is not actioned by either plan and stays ACTIVE as
   a tracked deferral. This complements ADR-046 (which established the known-issue
   context type) by recording that the type deliberately carries no enforcement.

4. Generated remediation plans get a static value: MEDIUM, not a derived
   grade. Deriving value from the score gap would invent a second heuristic
   for no proven benefit; a static default is the minimal fix that makes generated
   plans satisfy the validator. Revisit only if the grades prove systematically
   wrong.

5. The deferred plan sequences the enum migration before the schema validator,
   and does not enforce additionalProperties: false against generated
   remediation plans until they have a dedicated schema or a path-glob exemption.
   The effort enum must be final before a strict validator locks it in, and a
   strict validator would otherwise reject generated remediation plans on their
   extra keys.

Consequences

• Easier: the read protocol stays honest — the MEDIUM quick-wins plan is not
  weighed down by a LOW deferred tail, and the valuable remediation-plan fix ships
  without waiting on an optional migration.
• Prevents re-litigation: a future agent will not re-attempt known-issues
  enforcement (Decision 3), re-bundle the enum/schema work into the quick-wins
  plan (Decisions 1-2), or add a gapValue() heuristic without cause (Decision 4).
• Ongoing: the deferred plan carries genuinely open design questions (effort
  vocabulary, schema-validator runner, remediation-plan schema scope) to be
  resolved in its own plan-review when it reaches the top of the queue.
• Observed irony worth noting: the quick-wins plan is currently exempted from
  the check-undocumented-decisions scan by the very index.yaml-substring
  false-negative (G3) it exists to fix — this ADR gives it real coverage
  regardless, and G3’s fix will bring it under the scan legitimately.