fix(ledger): Gate F follow-up, Filbert's notes 1 to 3 (#1506)

Darkwing's follow-up to the T3 thread source: manifest 382f5bb0 pins
t3.mjs, ledger.test.mjs and README.md. Filbert approved it (review
6fd693b6). Ledger 51/51; the eight suites pass on the index.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

agents/filbert/work/ledger-t3-build-review-2026-09-26.md

@@ -144,3 +144,56 @@ What I confirmed on Node 26.8.1:
 The worktree was `/tmp/fb-gatef-wt`, removed after the run. The mutations ran
 there and were reverted, and the pins were re-verified before removal. No
 commit or push.
+
+## Follow-up: notes 1–3
+
+Candidate: `followup.md` `4bae617f…28e8`, `followup.patch` `856feee4…f44d`,
+and `followup-manifest.sha256` `382f5bb0…c971`, which pins `t3.mjs`
+`5acbc107…14fb`, `ledger.test.mjs` `6546dbaf…c63b` and `README.md`
+`101013de…38bd`. All hashes verify. Base `a4d38a3d`.
+- 136958c9 committed the pins I approved: `t3.mjs` `dfb092aa`, tests
+  `7444abd1`, README `27f7366d`, `ledger.mjs` `0afb0320`, `cli.mjs`
+  `d6092a53`.
+- Nothing under `packages/ledger` changes between 136958c9 and a4d38a3d.
+- The patch applies cleanly to a detached worktree at a4d38a3d and
+  reproduces the three pins.
+
+Checks:
+- **Ledger tests, clean worktree:** 51/51.
+- **Note 1.**
+  - The test file now holds a raw U+2028 and a raw U+2029 with CRLF, and it
+    asserts both are present.
+  - The README names both.
+  - `ledger.mjs` is unchanged.
+- **Note 2.** The new tests cover:
+  - `humanWithoutEvent` at 1;
+  - an unparseable event;
+  - a non-string `messageId`.
+  
+  The two bad-event tests assert `unknown` for both fields and unchanged
+  seats. Those are the three `return null` paths in `origins()` plus the
+  count, and together they cover my two surviving mutations.
+- **Note 3.**
+  - The rethrow keys on `typeof errcode === 'number'`. The existing tests for
+    SQLite 14, 1544, 5 and the not-a-database case still pass, so real SQLite
+    failures keep the `--no-t3` message.
+  - With the old `if` restored, the new test fails (0/1).
+  - I also checked the CLI path. I put a `null.x` into `query()` and ran the
+    CLI against a HOME fixture. It prints "Ledger failed: cannot read source
+    evidence" and exits 1. With the pin restored, the same fixture exits 0.
+    That fixture had no `orchestration_events`, and it also showed the
+    diagnostic's unknown path end to end.
+  - The in-process test passes an explicit fixture path, so the real `~/.t3`
+    stays closed.
+
+**Verdict: approve** manifest `382f5bb0`.
+
+**Observation for Sage, not about this patch.** Darkwing's live diagnostic is
+now 15 where the build had 14. I traced the extra message read-only. I printed
+the header shape only, never the body. It is a message in the Sage thread at
+2026-09-26T21:41:24Z, sent through T3's API, whose header reads `[from: main
+content-engine agent (<id>) -> to: sage (<id>) class=REPLY]`. The sender role
+has spaces, so the 6a rule calls it human, and it counts in Sage's human
+column. The diagnostic caught exactly the drift it exists for. It doesn't
+affect Gate F, which reads Filbert's row, but Sage's row now holds an agent
+message as human.