docs(checkpoint): e2e flake killed at a6a4e6f (2026-09-24 evening)

CHECKPOINT-encryption.md

@@ -1,15 +1,63 @@
-# Checkpoint — NIP-46 session restore on startup (2026-09-24)
+# Checkpoint — e2e flake killed (2026-09-24 evening)
 
 ## Where things are
 - Project: `/home/avi/Projects/Keynctr`
-- Branch: `master` @ **`0982dad`** ("feat(nip46): restore saved signer
+- Branch: `master` @ **`a6a4e6f`** ("test(nip46): kill the e2e flake —
-  sessions on startup/unlock — no fresh scan"). Previous: `73bf17c`
+  local relay for strict kind-0, poison-tolerant vault lock"). Previous:
-  (checkpoint), `f53bc56` (sign timeout leash).
+  `0982dad` (session restore), `73bf17c` (prior checkpoint).
 - Working tree: clean for tracked files. Untracked intentionally NOT
   committed: `COSMIC_THEME.md`, `KeynectrAppIconPossibility02.jpeg`,
   `deferred/` (stays deferred).
-- Release binary: **rebuilt at `0982dad`**. Frontend untouched — no
+- Release binary: **rebuilt at `a6a4e6f`** (test-only commit; same code
-  renderer rebuild needed. Restart Electron before retesting.
+  as `0982dad` for the app itself). Restart Electron before retesting.
+
+## What was completed this session
+1. **Root-caused and killed the intermittent e2e failure** that showed
+   up as 3–5 simultaneous test failures roughly 1 run in 4:
+   - REAL flake: `nip46_bunker_connect_params_match_spec_against_strict_amber`
+     published its kind-0 through the DEFAULT relay set — the real
+     internet (damus + the known-hanging nostr.band) — so "at least one
+     relay must accept the signed kind-0" was a network lottery against
+     the 6s send timeout. The test now pins settings to the in-process
+     relay like the QR test always did. Suite has zero network
+     dependency now.
+   - CASCADE amplifier: a panic while holding the test-only
+     `VAULT_ENV_LOCK` poisoned the mutex, so every later test died on
+     `PoisonError`. All four lock sites are now poison-tolerant
+     (`unwrap_or_else(into_inner)`) — a real failure reports as ONE.
+- Verification (all green at `a6a4e6f`): **10/10 consecutive green
+  e2e runs** (was ~1 in 4 failing), runtime now uniform ~21.5s (was
+  bimodal — long tail was network waiting). `cargo test` 216 unit + 5
+  e2e passed, `cargo clippy --all-targets` 0 warnings, `cargo fmt
+  --check` clean, `cargo build --release` green. Frontend untouched.
+
+## Commits added (newest first)
+- `a6a4e6f` test(nip46): kill the e2e flake — local relay for strict kind-0, poison-tolerant vault lock
+
+## How to resume / reproduce
+- Restart Electron (new release binary). With the vault already
+  unlocked/unencrypted, the backend logs `[NIP46] restoring session:
+  peer=... as npub1...` then `identity check on restored session:
+  PASS` and the profile goes Connected without showing Amber a new
+  scan. CLI trace: `~/Tools/keynctr-debug/` logs.
+- Flake regression check: `for i in 1..10; do cargo test --test
+  nip46_e2e; done` — all runs must pass in ~21–22s.
+- Then do the still-pending live proof: Publish name / publish a note
+  and approve in Amber within 2 minutes (120s leash from `f53bc56`).
+
+## Outstanding / next steps
+1. **Live sign/publish through real Amber** (covers the 120s leash and
+   the restored session in one shot).
+2. Prune the 11 stale profileless `nip46_connections` rows (cosmetic;
+   restore already skips them).
+
+---
+
+# Checkpoint — NIP-46 session restore on startup (2026-09-24)
+
+## Where things are
+- Branch `master` @ `0982dad`; see git for hashes. Details below are
+  as written when `0982dad` was HEAD.
 
 ## What was completed this session
 1. **Session restore without a fresh scan (`0982dad`)**: the WIP from