Plan: mark the library built and the shadow run gathering

Four and five are done and six is deliberately not. Everything six
needs exists; what is missing is weeks of real batches, and building
it now would throw away the only reason the shadow run exists.

Records what to watch for in the log, and the one decision the tests
forced: a copy that arrives with partners is trusted at once, because
holding it back would mean only ever proposing advertising detection
had already found twice. A cue gets no shortcut, which is the
independence intro and outro are waiting for.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018AwnALZL6bD4VBtATDVv3B

1 files changed, 33 insertions, 6 deletions

diff --git a/docs/superpowers/plans/2026-10-20-corrections-and-signature-library.md b/docs/superpowers/plans/2026-10-20-corrections-and-signature-library.md
index fb4ad2c..b93497f 100644
--- a/docs/superpowers/plans/2026-10-20-corrections-and-signature-library.md
+++ b/docs/superpowers/plans/2026-10-20-corrections-and-signature-library.md
@@ -282,7 +282,7 @@ thing.
 and had no tests; its output was an answer, not code to keep. Everything needed to rebuild it
 properly is above.
 
-### 4. Store and age signatures — NOT STARTED
+### 4. Store and age signatures — DONE 2026-10-24
 
 Two kinds, which behave nothing alike:
 
@@ -295,14 +295,40 @@ Confidence is a count of sightings across **separate fetch occasions**, which is
 independence test the withheld intro/outro offsets have been waiting for. Never promote on one
 sighting.
 
-### 5. Shadow run — NOT STARTED
+### 5. Shadow run — DONE 2026-10-24, now gathering
 
 Match against each batch, log what it *would* have cut, publish nothing. A week or two answers
 whether it finds advertising or accumulates junk, before one extra second is skipped.
 
-### 6. Publish matches, and unlock intro/outro — NOT STARTED
+**Running now.** Every detection pass learns from what it found and reports what the library
+would add, marking as *novel* only the proposals detection did not already publish. Those are
+logged individually with their votes, occasions and partner counts.
 
-Only once the shadow run has earned it.
+**It matches before it harvests.** Harvesting first would let a batch's own signatures match the
+batch they came from, and every run would report itself a triumph.
+
+**One decision the tests forced, worth knowing.** Requiring two separate occasions before a
+signature is matched against would mean the library could only ever propose advertising that
+detection had already found twice -- precisely when it has nothing to add. So a *copy* that
+arrives with partners is trusted at once: a break is only published because the audio recurred
+across episodes and correlation proved it, which is not a single unsupported sighting. A *cue*
+gets no such shortcut, which is the independence test intro and outro have been waiting for.
+
+**What to look for in the log.** `shadow -- learned N, reinforced N, library N (N worth
+matching); would cut N, N of them new`. The novel count is the number that matters. If it stays
+at zero for weeks the library is adding nothing; if it climbs with proposals that do not sound
+like advertising, the thresholds are wrong and nothing has been skipped in the meantime.
+
+### 6. Publish matches, and unlock intro/outro — DELIBERATELY NOT STARTED
+
+Only once the shadow run has earned it, and it has had one pass. Everything it needs is now
+built; what is missing is evidence, and evidence takes weeks of real batches. Building this now
+would throw away the only reason the shadow run exists.
+
+**What would make it ready:** novel proposals that hold up when listened to, and cue signatures
+reaching two or more separate fetch occasions. The second is what unlocks the withheld intro and
+outro offsets -- an opening theme confirmed across *weeks* is the show's, where one confirmed
+across a single batch of eight was an advertisement.
 
 ---
 
@@ -366,5 +392,6 @@ everything measured, though not directly caught in a probe.
   refresh rather than at launch.
 - **Library and Settings redesign (step 4 of the UI plan) not started.** Smallest wins; nothing
   depends on it. Mockups in https://claude.ai/artifact/AnywWE7j7e96QUpXvF9yaW
-- **Intro/outro offsets still withheld.** They need sightings spanning more than one fetch
-  occasion — which step 4 of the library delivers.
+- **Intro/outro offsets still withheld.** The machinery to unlock them exists now: a cue is
+  trusted only after two separate fetch occasions, which is exactly the independence they need.
+  What is missing is weeks of real sightings. Nothing to build; something to wait for.