🧠 מוניטור

11.10 12:38 onward: reminders kind=sunday week=2026-10-11 skipNearMin=120 maxSend=1 apply=1, driven by local script with 30-60s random gaps; 25 recipients (אביב שרם skipped, 14:25). Dedup across teachers via all items' hourLog (הילי דוד once). Each send logged 'sun 2026-10-11 p{id}' on the teacher's week item. Stops on first failure.

_settings: tpl_sunday_private_one/many (fields שם התלמיד, יום, שעה / מועדים). _week.sundayTextPrivate. reminders kind=sunday: loads waGroupsFast for 1:1 targets; groups gated by sundaySent as before; 1:1 sends logged per key 'sun {week} p{id}' in hourLog so they can go after groups and never twice. Panel preview supports the new templates. Dry run week 11.10: 29 private Sunday messages (אור 3, יעל 16, עדן 7, איתמר 1, זוהר not approved). Notes: shared family group (עטיה x3) gets 3 messages; הילי דוד tagged to יעל+עדן -> duplicate. Decisions also logged: unapproved week -> no reminders (Tohar OK); זאב ח׳ taught by Tohar -> needs a teachers-board row for Tohar (phone + meeting link).

_times.planTeacher(g, roster, teachers): keep current while tagged on >=half; switch to teacher tagged on >half only if active on teachers board (short name label, create_labels_if_missing); else flag. reconcile now handles teacher + hours. Webhook 188757448 change_specific_column_value multiple_person_mkwtafkx -> /api/monday-hook (plus the 3 hour-column hooks); nightly sync-groups. Synthetic cases pass (איתי->יעל, main+substitute keep, minority keep, no tags keep, 2/4 keep, non-teacher flag). Live dry run: only flag זאב ח׳ (section 'לא זוהר', 4/4 tagged טוהר חמדני, label זוהר). Panel overview shows the flag. Simulated hook event returned the flag.

Main board section group_mm1x9qdp renamed 'קבוצת נמר כיתה ז- יעל', 3 active students tagged יעל; groups board 3234060540 color_mm7b10t9 was 'איתי' (inactive). Set label 'יעל'. יעל week 11.10 approved -> preview now includes נמר ז׳: hour 11.10 16:00 + 14.10 15:00 with 3 names. Sunday 08:00 reminder for נמר ז׳ was missed (sundaySent already true for יעל). איתמר checked: only active student נועם חמדני 1:1 Tue/Thu 14:00 (hour reminders 13:00 scheduled, group 120363409895757824); 12 trial-lesson rows tagged איתמר are outside the reminder system.

New api/_times.js reconcile(): per group, member entries (untagged or tagged with group, excl 1:1); a slot moves when support < n/2 and one (day,time) has > n/2; same-day preferred; writes COL.day[i] label + COL.hour[i]; logs source=chance-times. Triggers: Monday webhooks 188750735/37/38 (change_specific_column_value on text_mkwt9ewq/text_mkxdyxet/text_mkzk50mv) -> /api/monday-hook (reconcile all, apply); nightly sync-groups as safety net. 7 synthetic cases pass (1/4 or 2/4 stay, 3/4 move, other-group tag ignored, 2-slot groups, day change, 1:1 ignored). Dry run on live data: only דולפין (4/4). E2E: temp item in test section, edit hour -> webhook -> דולפין 16:50->17:00 within 5s, monitor row; temp item deleted. Week 11.10 reminders verified. Panel guide updated. Test student 3247434700 is now inactive (Tohar).

Reverted groups board 3234061143 hour to 16:50. _week.js: buildWeek keeps g.pool (members + tagged guests with fixed hours); applyOverrides adds names of pool students whose fixed matches the moved slot's new day/time. Verified week 11.10 preview: דולפין sunday 'רביעי 17:00', hour 14.10 16:00 with 4 names; נמר ט׳ moved 16:00->16:10 also has 5 names. Dashboard mismatch text now covers temporary vs permanent. Note: אור changed the 4 students' שעה קבועה rows to 17:00 on 2149921157 (permanent fields) - flagged in panel.

אור (user 98516022) edited students' שעה קבועה on 2149921157 (7.10 17:08, 9.10 12:50) 16:50->17:00 (דולפין) for 4 students; groups board 5104526305 item 3234061143 hour_mm7bz5gh stayed 16:50 (form/reminders source). Set to 17:00. Verified preview week 11.10: sunday text 17:00, hour reminder 14.10 16:00 with 4 names (before: override moved slot but names empty). Panel overview: new problem when >=2 and majority of members' same-day fixed hours differ from group slot. Also confirmed Thursday retry: forms sent 8.10 18:00 to אור/יעל/עדן/איתמר; approved: אור, יעל, עדן, איתמר; זוהר pending.

8.10 week-open 12:00: server stuck 12:16-14:01 (stuck/unstuck rows; Sasha upgraded Lightsail). זוהר DM sent 12:01:53; אור failed (alert); יעל/עדן no rows (run cut at 60s by hanging sends). Fixes: sendWA AbortSignal 25s; week-open two passes (all rows first, then sends); new week-board column link_sent_at records each sent form; re-runs skip already-sent unless force=1 (panel 'send again' passes force); extra Thursday crons 12:00 and 15:00 UTC (15:00/18:00 IDT). זוהר's row backfilled link_sent_at. Pending recipients now: אור, יעל, עדן, איתמר (active+allowed, 1 private). Hour reminders this week: 12 sent incl. יעל בוגלה 1:1 on 7.10 (snapshot fix verified live); בר צברי 6.10 missed (pre-fix).

wa-health: when state open/unknown -> probe = findGroupInfos(test group) 14s, retry after 4s. Explicit error (e.g. Connection Closed) -> 'stuck': WhatsApp from SashaDibka to ALERT_TEST_TO (Sasha own) with restart command + activity_log 'stuck', repeat every 6h, quiet 21-08. Timeout only -> 'slow', no alert (server known loaded; upgrade planned 7.10). Probe ok after stuck -> '✅' message + 'unstuck'. Simona path (QR) unchanged. Tested locally with mocks: stuck alerts, recover closes loop, total <40s (<60s maxDuration).

Settings board 5105268254 had 2 items key=teachers_allow: 3260849533 'אור,זוהר' (07:49) and 3260858393 'אור,יעל,זוהר,עדן' (07:53), from two near-simultaneous panel saves. Deleted the stale one. _settings.load now keeps the latest row per key by 'by' timestamp and save() deletes duplicates. Verified panel reads all four. Also verified prod hour dry run 3.7s OK and wa_groups_snapshot saved 11:52Z.

6.10: group hour reminders sent (5.10 15:00 צפרדע ז׳, 16:00 צפרדע ט׳, 6.10 13:58 ינשוף מתקדמים). Private בר צברי (moved Tue 14:00 -> reminder 13:00) skipped: fetchAllGroups takes 27-35s > 25s cap -> no target -> noGroup, silent. Fix: waGroups timeout 55s; snapshot of group list saved to settings board key wa_groups_snapshot (hourly max) on every successful live listing; hour run resolves privates from waGroupsFast (cache/snapshot) then live; alertSasha on noGroup; wa-health refreshes snapshot when older than 6h. Verified locally: private resolved from snapshot with live lookup disabled.

SSH ubuntu@35.159.20.120 (key id_ed25519, docker needs sudo). Before: load avg ~7 on 2 vCPU, low per-process %CPU (2.3% top), mem avail 932MB, swap 760MB used, disk 36%. sudo docker compose restart (caddy, evolution, greenhook, postgres) at ~13:00 IDT. After: findGroupInfos works, test sendText to test group 201 at 13:15. API still slow (4-9s, some calls >60s), ssh commands hang >10min, sign.builders-tech.com (DocuSeal, same VPS) timed out. Suspected Lightsail CPU burst credits exhausted (high load, low per-process CPU) -> needs Lightsail metrics check / plan upgrade.