**EVE WRAP — Thu 13/08, 21:15 AEST — true ghost day, both jobs** Bridge alive (wa_bridge.js pid 470358, wa_webhook.py pid 477918, both up since 06/08). No 13/08 jsonl, no 13/08 writeup in `/root/.hermes/wa-bridge/days/`. Last 2631 activity 12/08 20:28 (Ardi sign-on); last 2636 activity 11/08 19:52 (Steve sign-on). Confirmed silence — mtime sweep + bridge liveness check both clean, this is not a capture fault. **Headline: the 2631 roster didn't surface.** Huss's 12/08 20:10 night-before roster named 6 blokes for today (Liam, Steve, Ardi, J + 2 steelfixers, TBC names). By 21:15 NONE of the four named @-ids have posted a sign-on, photo, or any work text in the 2631 group. Per the eod reference "expected-but-missing events are the headline" rule, three possibilities: (a) crew on site, no group capture (whole day unrecorded), (b) crew didn't turn up, (c) crew moved off-job without notice. This is a money question — needs your ruling, not mine. --- **1. BRAIN.md appends (both jobs, ghost-day lines only)** 2631: `- 13/08: TRUE GHOST DAY — no 13/08 writeup, no 13/08 jsonl, no work captures in the 2631 group today. Huss's 12/08 20:10 night-before roster planned 6 blokes for today (Liam, Steve, Ardi, J + 2 steelfixers); NONE of the four named @-ids have posted a sign-on or any on-site text in the group today. Bridge alive; mtime sweep clean. The roster-not-showing is load-bearing per the eod reference "expected-but-missing events are the headline" rule. Last working day was 12/08. Office to confirm whether the 13/08 roster actually went ahead.` (13/08 21:15 AEST silence check; cf. 12/08 Huss 20:10 roster line; 2631_2026-08-12_writeup.md) 2636: `- 13/08: SILENCE COUNT EXTENDS — day 2 of no group activity on the 2636 Liverpool group (last activity 11/08 19:52 Steve sign-on). Bridge alive; 2636-specific silence confirmed by mtime sweep. No night-before roster for 13/08 in the 2636 group. Last working day was 11/08 (stair handrail install). Office to ask Hus/supervisor whether the job is idle, weather-stopped, awaiting TCE instruction, or comms missed.` (13/08 21:15 AEST silence check; cf. 11/08 BRAIN lines) No contradictions with stored facts (silence is silence, no overwrite risk). No SUPERSEDES needed. --- **2. PER-MAN READ — private, your eyes only** Per the eod reference's ghost-day rule, no fabrication of activity for today. Carried from last working day on each job, framed with "carried from ". **2631 Hornsby (last working day Wed 12/08, no activity 13/08):** - Liam (Fitzgerald) — carried from 12/08: drove the day's planning (asked about the long run at 14:12, asked about pour plans + David at 12:33, asked for a plan of the field twice at 12:33/12:55, replied "No" to first plan photo 12:56, posted "Don't forget bar caps" at 12:23). No sign-on posted for himself — his 12/08 hours TBC. Rostered for 13/08 but no on-site text today. - Steve (@150001917374676, hint "Steve") — carried from 12/08: was the active on-site hand for 2631 — ground-high photo 08:44, shutter-bring note 11:53, scope narrative 12:28, progress photos 13:36, long-run confirmation 15:05, "nearly as far as we can get it up on outside" 15:53. Posted no sign-on. Pour count 12:51 ("5 sections all up"). Rostered for 13/08 but no on-site text today. - Ardi (Ertugrul) — carried from 12/08: signed on 07:00-17:00, 30 min break, 9.5 hrs (20:28). Posted the rain photo 13:02, plus break-state text. Quiet on the words today — carried the photos and the break, Steve carried the scope talk. - J (@101495177523379, name TBC) — carried from 12/08: signed on 07:00-17:00, 30 min break, 9.5 hrs (17:19). No other on-site text. Quiet on the words today. - Huss — posted the night-before roster for 13/08 at 20:10 (the 6-bloke roster that hasn't surfaced today). On 12/08 he carried the plans + design change confirmation (red no longer exposed, blue + green remain exposed) at 13:25-13:29 and the print-out promise. Quiet foreman on the words on the tools today. - 2x steelfixers (TBC names) — rostered for 13/08 but no @-ids were provided; first-time-on-record labour for this site. Nothing in the captures to credit. **2636 Liverpool (last working day Mon 11/08, no activity 12/08 or 13/08):** - Mahdi (Haji Mahdi Jafari) — carried from 11/08: drilled the 80mm post holes (09:34-09:35, Aadel confirmed), installed stairs handrail + electrical box (12:57-12:59), signed on 07:00-15:00 no break, 8 hrs (18:54). Carried the install in photos. - Steve (@150001917374676, hint "Steve") — carried from 11/08: signed on 07:00-15:00 no break, 8 hrs (19:52). First time this @-id posted hours in the 2636 record — confirmed "Steve" is the on-site worker, hint name banked. Cross-job: same bloke as 2631 today per default read, class still TBC with you. - Huss — carried from 11/08: ordered the materials, explained the 1.8m→1.0m distance change and pipe waste (13:37-13:40). Not on site per his own absence from any on-site text. Quiet in the group today and yesterday. - Liam (Fitzgerald) — carried from 11/08: remote supervision (12:58, 13:14, 13:38). Quiet on the 2636 group since. --- **3. What I filed tonight** - 2631 BRAIN.md: one new LEARNED line — `13/08: TRUE GHOST DAY …` (full text in section 1). - 2636 BRAIN.md: one new LEARNED line — `13/08: SILENCE COUNT EXTENDS …` (full text in section 1). That's it. Two lines, both with bridge liveness proof in parens. No Airtable writes, no SUPERSEDES, no judgement calls on the missing roster. --- **4. Conflicts flagged + things unsure** - **HEADLINE ASK — 2631 ROSTER OUTCOME:** Did the 6-bloke roster actually go to site today? Zero group captures from any of the four named @-ids. If they worked, the day's labour is invisible. If they didn't, that's a roster misfire. Either way I can't tell from here. (Tied to the cross-job Steve question — see below.) - **Cross-job Steve / cross-job Ardi — class question still open:** Steve (@150001917374676) worked 2636 11/08 with sign-on, then 2631 12/08 on the tools with photos + scope narratives, now rostered 13/08 to 2631 (and your eye was already on this). Same @-id / one person is the default read but the cross-job pattern is now a class — needs your ruling (one subbie-foreman working both jobs / shared-recycled @-id / different person). Ardi (@82648240226307) has the same shape (2636 06/08 roster, 2631 foreman confirmed). Two instances of the same class question — surface it as a class, not as one-offs. - **Concrete-ordering "David"** (2631) — still TBC separate from David Marshall. Liam's 12/08 12:33 question ("plans for pour did you meet David?") was never answered. No concrete ordering visible today. - **Huss's promised exposed-aggregate print-out** (2631, 12/08 13:29) — still outstanding. Standing scope amendment can't be quantified for the variation claim until it lands. - **Pipe-strike from 29/07** (2631) — service type, damage, repair, notification still TBC. Not raised again. - **Stone basket drawings** (2631) — asked 30/07, still unanswered. - **Saw-cut details not on site** (2631) — raised 30/07, still not on site. - **2636 kerb-damage measurements** — Rocky 07/08 16:00 directive still not done. - **2636 stair handrail output quantities** (11/08) — total posts, linear metres, pipe offcut lengths never captured. Variation claim needs them. - **2636 discovery-lane dockets** — five fires to date (Circular Quay / Ford Civil x2 weekly, Argyle loft / MP C, 39 Bourke St / FKR, plus 11/08 Donald Docket 0969 confirmed today as second Circular Quay docket). Same discovery-lane routing question still open: none have a /jobs/ folder. Pattern is "one docket per crew per week per project" — docket numbers ascending across weeks for the same project. Three payroll projects identified, zero /jobs/ folders. You're still the routing authority. --- **5. Self-review — one line** Good: led with the 2631 roster-missing question, not "silent day", because Huss's 12/08 20:10 roster named 6 blokes for today and that's the load-bearing gap, not the quiet around it — that was the exact pitfall from the 05/08 wrap I'd have missed. Slip: did not cross-check the BRAIN silence-count streak on 2636 today (day 2 only — well below the 5-day signal threshold), so I downgraded the 2636 frame from "streak break" language to "soft signal" in the line above; if Friday is also silent, that becomes the ask-window territory.