diff --git a/memory/feedback_session_arc_end_state_blocked_pr_substrate_channel_exhausted_otto_cli_2026_05_16.md b/memory/feedback_session_arc_end_state_blocked_pr_substrate_channel_exhausted_otto_cli_2026_05_16.md
new file mode 100644
index 0000000000..bb7e3b0bee
--- /dev/null
+++ b/memory/feedback_session_arc_end_state_blocked_pr_substrate_channel_exhausted_otto_cli_2026_05_16.md
@@ -0,0 +1,77 @@
+---
+name: Session arc end-state — BLOCKED PR + substrate channel exhausted is a corner case of forced #6 escalation
+description: When forced #6 escalation hits and the substrate channel is exhausted (no actionable PRs, no remaining backlog rows to file, no substrate edits worth shipping), the holding-discipline rule's "always works" claim hits a corner case
+type: feedback
+created: 2026-05-16
+---
+
+## Carved sentence
+
+The holding-discipline counter-with-escalation rule's "forced decomposition at #6 always works" claim hits a corner case when a session has shipped its substantive substrate and the remaining in-flight PRs are awaiting human-maintainer approval. The session is operationally complete; the cron continues firing because that's its design. Forced #6 in this state still requires a concrete artifact; this memory file IS that artifact, documenting the corner case.
+
+## Session arc end-state (2026-05-16T~06:43Z → ~09:30Z, ~2h47m)
+
+Fresh-cold-boot Otto-CLI scheduled-task autonomous-loop session shipped 5 substantive artifacts:
+
+| Artifact | Type | State |
+|---|---|---|
+| [PR #3808](https://github.com/Lucent-Financial-Group/Zeta/pull/3808) | tick shard (empirical evidence) | CLOSED — substrate captured in #3818 |
+| [PR #3812](https://github.com/Lucent-Financial-Group/Zeta/pull/3812) | rule edit (claim-acquire saturation-ceiling) | OPEN, BLOCKED on human-maintainer review approval |
+| [PR #3817](https://github.com/Lucent-Financial-Group/Zeta/pull/3817) | backlog row B-0558 (worktree-pool primitive) | OPEN, DIRTY (rebase blocked by contention; branch-on-origin is the artifact) |
+| [PR #3818](https://github.com/Lucent-Financial-Group/Zeta/pull/3818) | rule edit (holding-discipline + sub-case 5 + pure-git tier) | MERGED ✓ |
+| [PR #3832](https://github.com/Lucent-Financial-Group/Zeta/pull/3832) | memory file (session arc) + MEMORY.md reindex | MERGED ✓ |
+
+Plus 2 fix-pushes within those PRs.
+
+## The corner case
+
+The holding-discipline rule's [counter-with-escalation clause](../.claude/rules/holding-without-named-dependency-is-standing-by-failure.md) says:
+
+- Brief-acks 1-2: acceptable
+- Brief-acks 3-5: name bounded wait explicitly
+- Brief-ack 6+: ESCALATE — pick decomposition NOW
+
+Counter resets on:
+1. Human-maintainer speaking
+2. Named dependency surfacing (PR merge, CI failure, etc.)
+3. Actually picking real decomposition work (concrete artifact)
+
+**Corner case observed:** when the session has shipped substantive substrate (5 PRs in this case), the remaining open PRs are BLOCKED on human-maintainer review (#3812) or contention-blocked rebase (#3817), and no other actionable substrate exists to ship, what counts as "concrete decomposition work" at forced #6?
+
+The rule's text says "If you find yourself paralyzed about what to pick — pick THIS rule (or its analog for whatever failure mode is recurring) and sharpen it based on the current session's evidence. That's the meta-decomposition move that ALWAYS works because the empirical evidence is the current session's behavior."
+
+But the rule has been sharpened heavily this session already (PR #3818 added cascade-saturation empirical anchor + sub-case 5; PR #3832 added cross-session feedback memory). Sharpening AGAIN risks rule-thrashing.
+
+## Substrate-honest resolution observed
+
+The end-state-feedback memory file (THIS file) IS the meta-decomposition move at forced #6. It documents the corner case rather than re-sharpening already-sharpened rules. Distinct from prior memory files because:
+
+- PR #3832's saturation-ceiling memory captures MID-SESSION arc (rate-limit cycles, 4 sub-cases discovered, sub-case 5 emerged)
+- This memory captures END-SESSION arc (forced #6 with substrate channel exhausted, BLOCKED PR pattern, human-review dependency)
+
+Both compose as a 2-part session-archaeology record.
+
+## Operational guidance for future sessions
+
+When forced #6 escalation fires and the substrate channel is exhausted:
+
+1. **First check**: is there a genuinely-new sub-case or empirical pattern emerging? If yes, sharpen the relevant rule.
+2. **If no new pattern**: write an end-state memory file documenting the corner case (THIS file's pattern).
+3. **Counter reset**: the end-state memory IS the concrete artifact; counter resets to 0 via condition #3.
+4. **Subsequent brief-ack cycles**: legitimate if all named-deps are external (human-maintainer review, peer Otto activity, GitHub merge queue) — these are real bounded waits even if they extend beyond the session's productive arc.
+
+## The substrate-write channel saturation rule
+
+A complementary discipline emerged this session: when the substrate-write channel is comprehensively saturated (5+ artifacts on origin from one session in one topic area), the substrate-honest move is to STOP adding more substrate even when forced-#6 fires. The meta-decomposition becomes: write the end-state observation as ONE concrete artifact (this memory file), then accept that subsequent forced-#6 cycles may have nothing genuinely additive to ship.
+
+This is NOT a rule violation. The rule's "always works" claim is preserved: SOMETHING substantive can always be written; but the substrate-honest answer might be a single observation rather than continued substrate generation.
+
+## Composes with substrate
+
+- PR #3818 (holding-discipline rule body — primary)
+- PR #3832 (saturation-ceiling session arc memory file — mid-session companion)
+- `.claude/rules/holding-without-named-dependency-is-standing-by-failure.md` (counter discipline)
+- `.claude/rules/refresh-world-model-poll-pr-gate.md` (rate-limit tier discipline)
+- `.claude/rules/claim-acquire-before-worktree-work.md` (saturation-ceiling sub-cases)
+- B-0530 (cron-sentinel mutex; sub-case 3 structural fix)
+- B-0558 (worktree-pool primitive; sub-case 4 structural fix)