Doc: record the 2026-07-01 improve-in-place pass (ARCHITECTURE §10)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

docs/ARCHITECTURE.md

@@ -2,6 +2,10 @@
 
 Documentation generated from a full read of the codebase (2026-07-01).
 
+> **Note:** §7 (known issues) and §9 (improve-in-place plan) describe the code as
+> it was *before* the 2026-07-01 improvement pass. See **§10** for what has since
+> been fixed and what remains.
+
 ## 1. What this is
 
 PQS is a **player queueing and mission-management system for a pod-based
@@ -268,4 +272,48 @@ layout (field order, delimiters) can only be inferred from the PHP. One part of
 the contract is already known and must be kept: `getFSplayers.php` **always
 returns exactly 16 lines** in fixed pod order, inactive pods zeroed rather than
 omitted, because the game console addresses all 16 positions positionally.
-```
+
+## 10. Improvement work completed (2026-07-01)
+
+An "improve in place" pass was executed against the plan in §9. All changes were
+tested on a local WSL2 Ubuntu stack (PHP 8.3 / MariaDB; see `dev/`).
+
+**New infrastructure**
+- `includes/db.php` — one shared mysqli connection + parameterized query helpers
+  (`pqs_db`, `pqs_exec`, `pqs_rows`, `pqs_row`, `pqs_val`); also the single
+  source of truth for `PQS_TIMEZONE`. Targets **PHP 7.4+** (mysqlnd).
+- `includes/queue.php` — all seat-taking domain logic, serialized by a
+  `GET_LOCK` advisory lock and wrapped in transactions:
+  `pqs_enqueue_player`, `pqs_add_player_to_mission`, `pqs_add_player_to_missions`,
+  `pqs_enqueue_group`, `pqs_claim_group_slot`, `pqs_remove_player`.
+- `dev/` — local test-stack setup + `migrations/` (001: queue+mission → InnoDB;
+  002: currentmission+pastmissions → InnoDB).
+
+**§9 plan status**
+1. **Mission-fill races — DONE.** Every writer (kiosk single/group, staff
+   new/add-multiple, delete, commit) goes through the shared lock + a
+   transaction; capacity is judged by real `COUNT(*)` and `numplayers` is
+   rewritten from it, so it can't drift or overfill. Verified with 120–140
+   concurrent/mixed writers: zero overfill, zero drift.
+2. **Parameterize + strip DEBUG — DONE for all user-input paths.** api.php
+   (also un-broke the info-panel refresh, whose JSON the DEBUG text was
+   corrupting), console.php, registration.php, history.php, search.php.
+3. **Centralize DB / kill `mysql_*` — DONE.** db.php; 11 `mysql_error()`
+   landmines fixed; config.php unfetched-`pqs_pods` bug fixed.
+4. **Timezone + pod loop — DONE.** Timezone centralized in db.php; the 16 pod
+   branches in console.php collapsed to loops (fixed 16 kept as the hardware
+   contract).
+
+**Deliberately not changed**
+- `getFSgame.php` / `getFSplayers.php` — injection-safe already; their output is
+  the pod/AHK contract (see §4, `ahk/README.md`). Not refactored until the AHK
+  scripts are recovered and the exact format is confirmed.
+- Internal display includes (`getqueue`/`getconsole`/`getcurrent`/`getreg`/
+  `getgroup`) — read-only with internally-derived mission IDs; left on their
+  legacy connection (only `getgroup`'s `$_POST[group]` was int-cast). Converting
+  them to helpers is consistency-only.
+
+**Deployment prerequisites**
+- Run migrations 001 and 002 against the `pqs` database.
+- The new code targets **PHP 7.4+**. If the venue box is still PHP 5.4, upgrading
+  its PHP is the real prerequisite before deploying these changes.