diff --git a/TODO.md b/TODO.md
index c45f470..a30f539 100644
--- a/TODO.md
+++ b/TODO.md
@@ -243,12 +243,8 @@ Queued 2026-08-29, from building Gitea#11 and #16 (both shipped in v0.7.3, main
     **What is still untested is the part the item is named for: humans, at a table.** Nobody has sat
     down and played a game off the end of its timetable, and the multiplayer vote has never been
     driven through two browsers — what a second player sees while waiting on a first, and whether
-    "waiting on Carol" is legible once Carol has closed her laptop, are still unanswered. v0.7.3 is tested — engine, server,
-    replay, an all-bot regression — and compiled and exercised headlessly, but nobody has sat down,
-    run a game off the end of its timetable and voted. The multiplayer vote in particular has only
-    been driven through `session.intent`, never through two browsers: what a second player sees while
-    waiting on a first, and whether "waiting on Carol" is legible when Carol has closed her laptop,
-    are both unanswered. Worth being the first thing the next play session does.
+    "waiting on Carol" is legible once Carol has closed her laptop, are still unanswered. Worth being
+    the first thing the next play session does.
 
 36. **There is no per-Stage "this train did not move" signal, so "longest an engine sat on a siding"
     cannot be answered.** Gitea#16 asks for it and the comment on that issue said `trainStoodStill`
@@ -263,7 +259,7 @@ Queued 2026-08-29, from building Gitea#11 and #16 (both shipped in v0.7.3, main
     folds a batch of events AFTER `advance` has already mutated past the moment they describe. Worth
     settling with the badge pass (#33) rather than on its own, since that is the only consumer.
 
-37. ~~**Bump the StartOS wrapper to 0.7.3.**~~ — done 2026-08-29 (`74aea24` in
+37. ~~**Bump the StartOS wrapper to 0.7.3, then 0.7.4.**~~ — 0.7.3 done 2026-08-29 (`74aea24` in
     `station-master-startos`). Submodule pinned to `v0.7.3` (`45580d8`), `current.ts` at `0.7.3:0`,
     release notes in all five locales, `README.md` and `instructions.md` updated. No new version file
     and no migration — the outgoing `0.7.2:0`'s `up` was empty, `versions.md`'s common case, so
@@ -275,10 +271,42 @@ Queued 2026-08-29, from building Gitea#11 and #16 (both shipped in v0.7.3, main
     proven rather than assumed by `test/harness.test.ts`, which replays the three files in
     `public/replays` (all recorded under an older ruleset) and asserts every intent still applies.
 
-    **The v0.7.3 tag is made locally and signed, but not pushed** — agents cannot push to
-    `origin`. Until Jesse pushes it, the tag exists only here and in the wrapper's submodule clone;
-    the recorded pin is the commit SHA, which is already on `origin`, so the wrapper is not blocked
-    on it.
+    **Bumped again to `0.7.4:0` the same day** (`085b88b`), pinned to `v0.7.4`, and installed on
+    `phoenix.local` — verified there, not merely packed: the resume log shows `WHISTLE-4086` coming
+    back with its 7 intents and all five new engine code paths present in the served build. Both tags
+    are signed and pushed.
+
+    **Keep doing the whole sequence.** Tag the app, fetch the tag into the wrapper's submodule,
+    bump `current.ts` in place (the outgoing `up` has been empty every time, `versions.md`'s common
+    case), rewrite the notes in all five locales, update `README.md` and `instructions.md`, then
+    `npm run check` / prettier / `make x86` / `make install`. `UPDATING.md` in the wrapper is the
+    authority and has not needed changing.
+
+38. ~~**Gitea#13, #5 and #19 — three rules corrections.**~~ — done 2026-08-29 in v0.7.4
+    (`5e34c73`, `2280276`, `19a6a47`), wrapper `085b88b` as `0.7.4:0`, installed on `phoenix.local`.
+    Each issue carries a comment naming its commit and what was ruled, per #30. Reasoning is in
+    `CHANGELOG.md`; what matters here is what they left behind, below.
+
+39. **NONE OF v0.7.4 HAS BEEN PLAYED BY A HUMAN.** The Yard Office offer, the Red Flag hold and its
+    out-of-phase prompt, and the loaded-Extra make-up rules are all tested end to end, packed, and
+    running on `phoenix.local` — and no person has met any of them at a board. Two are interruptions
+    that stop the Mainline Phase and put a question in front of somebody mid-thought, which is
+    exactly the kind of thing only play reveals. Together with #35 this is now the biggest gap in the
+    project: four features shipped without a table.
+
+40. **A save from before v0.7.4 may not replay, and nobody has been told.** The same shape as #32 but
+    for the main line: the Red Flags intent changed shape, a make-up that was legal may now be
+    refused, and a Yard Office arrival asks a question no older history has an answer for. It fails
+    safe — the server declines the save, names the move and leaves the file untouched — and
+    `WHISTLE-4086` did survive on `phoenix.local`, so it is "may not" rather than "will not". Worth a
+    line wherever the build is announced, and worth knowing when a bug report arrives with a save
+    that will not load.
+
+41. **The bot cannot use the half of Red Flags a human would.** It takes the danger prompt
+    unconditionally and still plays zero flags in 200 games, because the prompt needs a colliding
+    arrival to coincide with holding the card. What it never does is plant a flag ON PURPOSE to buy a
+    Stage for switching, which needs it to know it wants time — a notion it does not have. Reasoning
+    and the measurement are under Bot Performance.
 
 ---