diff --git a/docs/hygiene-history/ticks/2026/05/03/0257Z.md b/docs/hygiene-history/ticks/2026/05/03/0257Z.md
new file mode 100644
index 000000000..3c939d122
--- /dev/null
+++ b/docs/hygiene-history/ticks/2026/05/03/0257Z.md
@@ -0,0 +1 @@
+| 2026-05-03T02:57:00Z | opus-4-7 / autonomous-loop continuation | a2e2cc3a | **Second calibration data point landed (65% on B-0172, up from 48% on B-0173) — KEY NEW FINDING: principle-strong + specific-weak pattern is context-dependent.** Cycle worked: filed guess #002 on B-0172 plugin-packaging (PR #1282 — pre-recovery self-prediction included as meta-calibration). Then executed GROUND-TRUTH-RECOVERY: 26/40 = 65% across 4 layers (vs guess #001's 19/40 = 48%). **Pattern progression**: guess #001 (B-0173, no prior specific-context) scored 3/10 on specific-implementation layer; guess #002 (B-0172, recent PR #1262 path-correction context) scored 7/10 on the same layer. **Hypothesis**: specific-context-density predicts specific-layer accuracy; the principle-strong + specific-weak gap narrows when recent context is present. **Pre-recovery self-prediction**: 2/3 correct (architectural PARTIAL-MATCH ✓ + substrate-content MIXED ✓ + specific MOSTLY-OFF predicted but actual MOSTLY-MATCH ✗ — over-predicted weakness when context present). Architectural-layer gap (Aaron's verbatim *"so we can take advantage of hooks in harnesses"* + promotion-trigger maturity-gate) replicated guess #001's principle-strong + frame-specific-weak pattern. Cross-row composition layer scored well (7/10) — got right rows; mis-categorized B-0173 as composes_with (actual depends_on) because architecturally hooks must precede plugin packaging (without hooks, packaging is bare-skill-grouping per Aaron's exact phrase). | #1283 (B-0172 ground-truth recovery + delta) wait-ci, auto-merge armed; #1282 (guess #002 B-0172) wait-ci, auto-merge armed; #1281 (tick-0251Z) wait-ci, auto-merge armed; #1280 (B-0173 ground-truth recovery + delta) wait-ci, auto-merge armed; #1278 (guess-then-verify protocol memo) MERGED | This tick teaches **context-dependent calibration as a refinement to the principle-strong + specific-weak pattern**: Otto's specific-implementation accuracy is not a fixed weakness; it varies as a function of recent specific-context density. When PR fixes / doc reads / commit context exist for a specific architectural layer, accuracy approaches principle-layer accuracy. When absent, specific-layer accuracy degrades to baseline ~30%. Future-Otto: don't auto-predict weakness on specific-implementation; instead, predict-by-context-density. The hypothesis is testable on subsequent guesses (guess #003+). |