a-debt-nothing-lists-owed-fidelity-reviews

Summary

The design record for itd-2609150819445595, from the four decisions on the intent (2026-09-20).

Scope

1. One reader of the review marker across shipped/ in internal/core/intent: for every shipped intent, the first marker line (existingMarker is the authority) yields one of OWED, INGESTED, DEAD_LETTER, or none; the reader returns the set with each intent's receipt id where one exists and the dead-letter reason where one was recorded. It reads; it never rewrites a request (decision 2).
2. The owed set (decision 3): OWED plus none; DEAD_LETTER is its own heading with the reason and is not counted; INGESTED is absent. A no-marker entry says a receipt is minted on re-emit.
3. Three surfaces (decision 4): bare abcd intent audit renders the listing in text and --json (one entry per shipped intent: id, state, receipt, the re-emit command; no local-tier path); bare abcd intent adds the owed count beside its bucket counts from the same reader; the record dispatcher's next move for a shipped intent reads the marker and says "fidelity review owed, receipt rcp-…, re-emit with abcd intent audit <itd-N>" in place of the current "none".
4. No gate (decision 1): nothing in lint, record-lint or the release cut reads the owed set.

Out of scope

Running the auditor (host-delegated); a legacy marker; any stamp on a no-marker intent (re-emit mints the receipt as it does today).

Approach

Test-first: fixtures with one shipped intent per state, plus a shipped intent with a duplicated marker to pin that the first marker wins. One lane. The reader is the one canonical primitive; the count on the lifecycle board and the dispatcher's next move call it rather than re-scanning.

How the criteria are satisfied

1 to 4 by pieces 1 and 2; 5 and 8 by piece 3's listing; 6 by piece 3's board count; 7 by piece 3's dispatcher move.