Skip to content

Operational backlog

The project's living short-term working list: work we intend to pick up soon, and decisions we have flagged but not yet taken. It complements:

Items graduate out of here into an ADR/SRD (when designed) or a FIX (when implemented), and leave this file. The narrative stays in the document that closed them, not here.

Long-term and blocked work belongs in GitHub issues, not here. An item waiting on something that isn't scheduled — another subsystem, a decision not yet due — has no business in a working list; keeping it here splits the backlog across two sources and makes neither trustworthy. Move it to an issue with what was measured and what would unblock it, and drop it from this file. Roadmap work is milestoned there too, not restated here.

Open questions to analyze

Design decisions we've surfaced but not yet made.

None right now.

Planned work / deferred

Genuinely un-homed items — not yet tracked in an ADR/SRD, the roadmap, or the audit-backlog.

  • Wipe the version pins from the remaining SAD/ADR references. A reference to a SAD or ADR now names the document and the section only — ADR-033 §2.8, not ADR-033 v.5 §2.8. Bumping one of them updates everything related to it in the same change-set, so a reference is current by construction, and the pin only supplies something for the next bump to falsify.

The SRD-091 branch converted what it already had to touch: ADR-003 (bumped to v.2 there), the continuously-current ADRs and the RU twin that referenced it, the roadmap, SRD-091, ADR-037, the persistence guide, and the ADR references in code comments that branch added. That leaves ~2,400 pins under docs/.

Two rules bound the sweep. Accepted SRD/FIX documents are not touched — a one-shot document is a historical snapshot, its pin was correct when written, and retro-editing it rewrites what its author read. The same holds for the CHANGELOG's released sections, which record what shipped. The target is the continuously-current set: SAD, ADR, their RU twins, the guides, the roadmap and README*.

Deliberately not a rider on a feature branch, and not urgent: an un-swept pin goes stale only when its target bumps, which the sync-linked-docs step catches one document at a time. It is worth a pass of its own precisely because leaving it to that step means every bump keeps paying the repair cost the convention exists to abolish.

The six entries this section carried through 2026-08-01 all landed: the optioned-constructor doc-comment audit, the discard/assertion lint guard, per-module coverage profiles and the Markdown link check closed with FIX-034; the swallowed observer panic with FIX-035; and "examples assert their own outcome" as plain commits on test/examples-assert-outcome.

Tracked elsewhere (not duplicated here)

  • Event-matching generalization — RETIRED (2026-07-30). The premise was that Link would be the second name-keyed event type, making a polymorphic SubscriptionKey() pay off. Link landed as a static graph redirect (SRD-057), so Signal remains the only name-matched subscription and the abstraction has no second consumer. Revisit only if a genuinely name-keyed event type appears.
  • Audit findings — disposition in audit/remediation-status.md, design deferrals in audit/audit-backlog.md. Both maintain their own counts; this file does not restate them.

The note that used to sit here — that row 11 (memrepo can evict an Active instance after a terminal→Active re-save) was the single remaining open row, deferred "until persistence lands" and now belonging to E2 — was stale. It was remediated by SRD-078 FR-9: Save untracks a revived record from the eviction ledger, remediation-status.md row 11 reads ✅, and TestTerminalRevivedToActiveNotEvicted pins it. This file kept pointing at work that had already landed, which is the failure mode a backlog has: an entry costs nothing to leave behind and reads as open forever.