memory: the owed items and the SPYRJA fold; the mutation tool's collection-error hole

persistent-memory.md

@@ -31,7 +31,7 @@ _As of 2026-09-27:_
   inputs; carry() measures a sent-then-changed form against what it SENT (an
   answer set back mid-flight survives); the two double-matching r2b anchors
   are re-anchored AND `scripts/mutation_check.py` now refuses any anchor that
-  matches more than once; the r2_flow C3 prose points at as_antislop S5b.
+  matches more than once; the r2_flow C3 prose points at as_antislop S5b. A hulda hunt on that (SPYRJA) found 8 more, all folded (`213071b`): report input now guards the lone native submit too, report controls are baselined at mount (a bare select no longer holds every reload), contenteditable counts.
   **S5b MERGED AND PUSHED** (`0233ca6`, operator "merge and push",
   2026-09-28; booth-dev gate 1068 / 428/428 + seam pass clean): focus
   survives a swap, a floating status line, and leaving with an unsent answer
@@ -107,6 +107,7 @@ _As of 2026-09-27:_
 
 ## Recent decisions
 
+- `[2026-09-28]` ⚠ **`mutation_check` counted ANY non-zero exit as proof, a collection error included** — found by the SPYRJA hunt (hulda). Only pytest exit 1 proves now; timeouts, byte-exact restore, a run lock and overlap-aware anchor counting landed with it. The first full run under the hardened tool: 433/433, so no row had been proving by a broken import. READ BEFORE TRUSTING ANY INSTRUMENT'S "red": ask what else turns it red.
 - `[2026-09-28]` ✅ **A posted doc could run script; raw HTML is now escaped and doc hrefs guarded** — Prime ruled ESCAPE; READ BEFORE RENDERING ANY AUTHOR TEXT `|safe` or touching `links.is_safe_href`, which now guards docs too → `persistent-memory.d/2026-09-28-a-posted-doc-could-run-script.md`
 - `[2026-09-27]` ✅ **One submit saves every ask on the page; the heid panel found the async window** — READ BEFORE TOUCHING THE SUBMIT PATH OF embed.js OR base.html: a batch reads forms at the press while the page stays live, and only a test that acts inside the flight can see it → `persistent-memory.d/2026-09-27-one-submit-saves-every-ask.md`
 - `[2026-09-24]` ⏸ **The upload route's three lifecycle gaps: DEFERRED** — the pickup-id `mkdir` sits outside the try (a FileExistsError race), `rmtree(ignore_errors=True)` hides its own failure, and `except Exception` misses CancelledError. All three are rare; the operator was told and merged without them. Tracked in `225ba32`'s commit message.