hygiene: first backlog-refactor fire (knowledge-absorb on AutoDream row)

docs/BACKLOG.md

@@ -5425,6 +5425,26 @@ systems. This track claims the space.
   Cadence: 24h + 5 sessions. Consolidation *deletes*
   contradicted facts. See `memory/reference_autodream_feature.md`.
 
+  **Knowledge-absorb update (2026-04-23, first backlog-refactor
+  fire per FACTORY-HYGIENE row #54):** the factory's
+  near-term position on AutoDream has since evolved — rather
+  than building a parallel retraction-native consolidator as
+  a standalone research project, the factory first lands a
+  thin **extension-overlay policy** on top of Anthropic's
+  feature so upstream upgrades are inherited automatically.
+  See `docs/research/autodream-extension-and-cadence-2026-04-23.md` (lands via PR #155)
+  (lands via PR #155) and the corresponding
+  `docs/FACTORY-HYGIENE.md` row #53 (lands via PR #155) for the cadenced
+  enforcement surface. The better-dream-mode research project
+  below is now framed as the **more-ambitious second step**:
+  if the Anthropic feature proves insufficient for the
+  factory's memory discipline even after the four
+  factory-overlay steps (A/B/C/D in the research doc), *then*
+  a retraction-native replacement earns its research weight.
+  Until that signal fires, this row stays queued at its
+  current priority but is no longer the primary AutoDream
+  response.
+
   Zeta already has memory-architecture primitives AutoDream
   does not:
 

docs/hygiene-history/backlog-refactor-history.md

@@ -0,0 +1,15 @@
+# Backlog-refactor audit — fire history
+
+Per-fire ledger for FACTORY-HYGIENE row #54 (backlog-refactor
+cadenced audit). Schema per row #47 (date / agent /
+pre-count / post-count / passes fired / findings /
+next-expected).
+
+Authoritative policy:
+`docs/FACTORY-HYGIENE.md` row #54 (lands via PR #166) +
+per-user feedback memory
+`feedback_backlog_hygiene_cadenced_refactor_look_for_overlap_not_just_dump_2026_04_23.md`.
+
+| Date | Agent | Pre-count | Post-count | Passes fired | Findings | Next expected |
+|---|---|---|---|---|---|---|
+| 2026-04-23 | Claude (first row-#54 fire, bounded pilot) | ~60 rows (approx on main) | ~60 (no retire this fire) | Knowledge-absorb only (pass d) | Updated the "Retraction-native memory-consolidation (better dream mode)" research row with a cross-ref to the newer AutoDream extension-overlay policy (PR #155) + hygiene row #53. The better-dream-mode project is reframed as the **more-ambitious second step** conditional on the overlay policy proving insufficient; not the primary AutoDream response anymore. | Next full-cadence fire at 5-10 rounds from now. Bounded-pilot ticks can fire opportunistically on-touch when a new research/policy landing changes older rows' framing. |