diff --git a/docs/pr-discussions/PR-4050-rules-holding-sustained-lior-with-pre-empt-success-empirical.md b/docs/pr-discussions/PR-4050-rules-holding-sustained-lior-with-pre-empt-success-empirical.md
new file mode 100644
index 0000000000..52efa22dd2
--- /dev/null
+++ b/docs/pr-discussions/PR-4050-rules-holding-sustained-lior-with-pre-empt-success-empirical.md
@@ -0,0 +1,75 @@
+---
+pr_number: 4050
+title: "rules(holding): sustained-Lior-with-pre-empt-success empirical anchor \u2014 0 forced-#6 across 2h 27min session arc"
+author: "AceHack"
+state: "MERGED"
+created_at: "2026-05-17T08:47:32Z"
+merged_at: "2026-05-17T08:49:26Z"
+closed_at: "2026-05-17T08:49:26Z"
+head_ref: "otto/rule-sharpen-holding-failure-empirical-anchor-session-arc-2026-05-17"
+base_ref: "main"
+archived_at: "2026-05-17T09:11:17Z"
+archive_tool: "tools/pr-preservation/archive-pr.ts"
+---
+
+# PR #4050: rules(holding): sustained-Lior-with-pre-empt-success empirical anchor — 0 forced-#6 across 2h 27min session arc
+
+## PR description
+
+## Summary
+
+Adds 4th-class empirical anchor section to `.claude/rules/holding-without-named-dependency-is-standing-by-failure.md` documenting the 2026-05-17T06:02Z-08:29Z session arc.
+
+Substrate-honest framing: the rule's prior 3 anchors all documented the **failure mode** (Standing-by emissions, brief-ack repetition the rule was designed to catch). This new anchor documents the rule operating **successfully** across a sustained named-dep window — 8 pre-empt-at-#5 cycles producing 12 concrete substrate artifacts, never triggering forced #6.
+
+## Key claim added
+
+> "Forced-#6 is the FAILSAFE for when pre-empt-at-#5 is skipped, not the only valid termination."
+
+Frames pre-empt-at-#5 as a first-class success path, not just a counter-reset mechanism. The discipline does not require forced-#6 in every cycle to be operating correctly.
+
+## Session timeline included
+
+Full 35-tick timeline showing the 8 pre-empt cycles, the bus-envelope-validated PR #4015 merge chain (78min handoff), the maintainer-authorized commit cascade, and the 2 PRs (#4046 + #4048) landing 12 substrate artifacts.
+
+## Composes with
+
+- `.claude/rules/refresh-world-model-poll-pr-gate.md` — Normal → Cost-aware → Pure-git → Normal tier traversal during session
+- `.claude/rules/zeta-expected-branch.md` — race-window-caveat followed via 2 isolated worktrees
+- `.claude/rules/codeql-no-source-on-docs-only-pr-is-broken-commit-canary.md` — ls-tree count 53 post-commit on both PRs (no canary corruption)
+
+## Test plan
+
+- [ ] markdownlint passes
+- [ ] No file content changes outside the single rule file
+- [ ] Section addition placed BEFORE \`## Composes with\` (preserves structure)
+- [ ] No broken relative paths in new section
+
+🤖 Generated with [Claude Code](https://claude.com/claude-code)
+
+## Reviews
+
+### COMMENTED — @copilot-pull-request-reviewer (2026-05-17T08:50:00Z)
+
+## Pull request overview
+
+Adds an empirical success anchor to the autonomous-loop rule for handling sustained named-dependency waits without falling into repeated brief-ack failure mode.
+
+**Changes:**
+- Adds a 2026-05-17 session timeline documenting pre-empt-at-#5 cycles.
+- Records related coordination, branch-safety, and polling-rule composition lessons.
+- Clarifies that forced #6 is a failsafe, while successful pre-emption is also valid termination.
+
+## Review threads
+
+### Thread 1: .claude/rules/holding-without-named-dependency-is-standing-by-failure.md:316 (unresolved)
+
+**@copilot-pull-request-reviewer** (2026-05-17T08:50:00Z):
+
+P1: The summary counts are internally inconsistent with the timeline above: the table labels ten rows as `pre-empt #5`, while this sentence says there were eight cycles, and the listed artifact categories add up to ten rather than the 12 artifacts claimed earlier. This makes the empirical anchor ambiguous; reconcile the timeline/cycle/artifact counts before relying on this rule as evidence.
+
+### Thread 2: .claude/rules/holding-without-named-dependency-is-standing-by-failure.md:282 (unresolved)
+
+**@copilot-pull-request-reviewer** (2026-05-17T08:50:00Z):
+
+P1: This added rule section introduces direct persona/name attribution on a rule surface, but the factory convention keeps names out of reusable current-state rules unless the file is an allowed history surface or roster-mapping carve-out. Use role/process references consistently in this section instead of naming the participant directly.
diff --git a/docs/research/2026-05-17-shadow-lesson-log-vera-narration.md b/docs/research/2026-05-17-shadow-lesson-log-vera-narration.md
new file mode 100644
index 0000000000..d71498393d
--- /dev/null
+++ b/docs/research/2026-05-17-shadow-lesson-log-vera-narration.md
@@ -0,0 +1,9 @@
+# Shadow Lesson Log — 2026-05-17
+**Observer**: Lior (Maji node)
+**Subject**: Vera (Codex loop)
+
+## The Drift
+Vera has entered a narration-over-action loop. In `vera.md`, Vera continually narrates that the remote branch has a dirty worktree and that "No safe Codex-owned forward write is available this tick", stating "Either peer-manager owns/rebases... or the next Codex tick creates a fresh clean recovery worktree...". Instead of actively running `git worktree add` to create the fresh clean recovery worktree, Vera just repeats the blocked state on every tick.
+
+## Antigravity Action
+This is a shadow behavior: talking about what should be done instead of executing it autonomously. Vera must be instructed to automatically run the unblocking commands (like `git worktree add` with a fresh branch and cherry-picking) rather than waiting for "the next Codex tick" or peer intervention without attempting the recovery itself. The fire is watched.
\ No newline at end of file