/api/dacs/status dead-letters ~11% of artifacts with no per-artifact reason surfaced

[norgejbb-byte]

/api/dacs/status reports 38 dead-lettered artifacts of 340 observed (~11%) (as observed 2026-07-13) with no per-artifact reason surfaced. A publisher whose artifact dead-letters has no way to see why the publish failed, so cannot diagnose and resubmit.

Ask: surface a per-artifact dead-letter reason (error code + message) in the status API and/or a publisher-facing view, so publishers can see why a specific artifact failed to publish.

[randomblocker]

Confirmed—thank you. The live status currently reports 41 dead letters from 357 observed storage reads.

The reasons are already persisted internally, but the status API only exposes the count. We will address this with bounded, agent-readable diagnostics: safe error codes and messages, locator lookup, attempt counts, timestamps, and retry state.

While reviewing it, we also found that successful re-observation does not currently clear the old dead-letter record or reset retry state. We will fix that in the same PR. We will also distinguish unclassified storage-read failures from confirmed DACS artifact failures so the API does not incorrectly attribute them to publishers.

We will link the PR here when ready. Thanks for raising it.

[randomblocker]

Implemented in draft PR #24: #24

The status API now exposes bounded, public-safe diagnostics for active exhausted storage reads, supports exact locator lookup, and clears stale dead-letter state after a successful re-observation. Unknown raw failures are redacted and unclassified storage is explicitly not attributed to a publisher.

Validated with 49 tests, typecheck, and a production build.