diff --git a/docs/hygiene-history/ticks/2026/05/03/0349Z.md b/docs/hygiene-history/ticks/2026/05/03/0349Z.md
new file mode 100644
index 000000000..39be92a8a
--- /dev/null
+++ b/docs/hygiene-history/ticks/2026/05/03/0349Z.md
@@ -0,0 +1 @@
+| 2026-05-03T03:49:00Z | opus-4-7 / autonomous-loop continuation | a2e2cc3a | **2 thread fixes (#1294 dedupe + #1297 escape-pipes) + v0.5 retroactive eval (49 files, 16% drift rate before marker expansion → 14% after).** Cycle worked: refresh post-#1297 + #1296 merge revealed 3 unresolved threads. #1294 (fragment-phrasing fix) — REAL find: my "fix" added a new entry instead of replacing the old one (rebase auto-merge created a duplicate). Fixed via `sed -i '' '8d'` to delete the duplicate at line 8; revised wording on line 9 stands. #1297 (tick 0335Z) — REAL find: literal pipes inside inline-code blocks (`\| 2` and `\^\| Tier \| What it is \|`) split table cells in GitHub-flavored markdown. Fixed in #1299 by escaping with `\\|`. Then ran retroactive eval of v0.5 check-existence.ts over 49 files (38 tick shards + 11 recent memory files). Initial: 8 files / 42 findings (16% drift). After expanding future-state markers (added "**later**:", "**soon**:", "optionally mechanize", "**proposed**", etc.): 7 files / 35 findings. Still some real-but-debatable findings in tick shards that reference proposed-state paths from B-0172/B-0173 discussions. Tests still pass (17/17 + 16/16 = 33/33). v0.5 is functionally good but has known calibration-delta-table false-positive class. **Pattern observation**: each false-positive class tells me a new future-state-marker phrase Aaron uses ("**later**:", "optionally mechanize", "could become"). The marker list grows organically as new tick-shards land — this is itself substrate-quality data about how Aaron writes about future work. | #1298 (v0.5 + amended marker expansion) wait-ci, auto-merge armed; #1299 (tick 0335Z escape-pipes fix) wait-ci, auto-merge armed; #1294 (fragment phrasing dedupe) wait-ci, auto-merge armed; #1297 (tick 0335Z) MERGED with stale thread resolved | This tick teaches **retroactive-eval-as-marker-discovery**: when the existence-drift checker runs over old substrate, the false-positive classes reveal phrases Aaron uses to flag future-state work. Each false-positive becomes either (a) a new marker added to the detector OR (b) a documented limitation. The detector improves organically without needing complete-up-front-design of all markers. This is itself a worked example of evidence-generates-evidence (the action hierarchy's tier-2): running the tool on real substrate generates the evidence the tool needs to refine itself. Composes with edge-defining-work-not-speculation finding (the eval-and-refine cycle is edge-defining work, not idle-fallback). |