## Remedy A implementation and gate handoff

Commit: `982b90843bf833881e53ffaf2f91798fc509928c` on `feat/300-input-acceptance` (local candidate; not merged).

- `PRESENCE_PROBE_CEILING`: 10000 ms → 2500 ms. The separate shared `PROBE_TIMEOUT` remains 2500 ms and is unchanged.
- Diagnosis attribution: perri's **10105/10120 ms** waits were **one unreachable peer's dial reaching the former 10-second presence ceiling**. The separate **457 ms pre-subnet block** remains separate; the underlying **30-second dial** is also distinct and unchanged by this remedy.
- An unanswered presence probe beyond the ceiling remains **UNKNOWN, never OFFLINE**. Probes still fan out concurrently.
- Replaced the current requirement title/activation comment, CONTEXT cost paragraph, and CLI ceiling comment. The existing wedged-probe cell now uses the actual `PRESENCE_PROBE_CEILING`, not an unrelated 150 ms test constant.

Verification: exactly two named CLI cells passed, no full suite or old #49/#267 rigs:
- `cli::tests::a_wedged_probe_costs_one_ceiling_and_reads_unknown_not_offline` (test-process duration 3.205 s)
- `cli::tests::only_rows_with_a_node_are_asked_and_answers_land_on_their_own_row`

`traceable-reqs check --json`: 903/903 complete, zero findings. Generated documentation refreshed.
Evidence: `.spt/preserved/todlando-300-input-acceptance/{299.diff,selected299.txt,run299.stderr,trace299.json}`.

Published/cached presence is the existing structural follow-up **#306**: https://github.com/BigscreenVR/spt-bs-releases/issues/306 . No duplicate issue and no cached-presence implementation in this correction.

— todlando