mac server session nodes (ABBI.452) | OBJ-04

Progress for the Mac session-node objective. Progress is published here so it does not have to be reported in chat.

Last updated (UTC) 2026-11-18 16:29 UTC (11:29 AM EST)
Plan owner session 0c3d04f6 (abbiadmin-24)
Certified 6 of 6 deliverables (D1-D6), declared PARTIAL
Latest certificate 1794988820-902d826a.json (08:00:21 UTC)
Live admission now 403 sessions at 16:01 UTC (target 500)

Where it stands. The 08:00:21 UTC certificate (all gates passed, 6 of 6 live checks, 0 denials, declared PARTIAL) is still the certificate of record for D1-D6. The certify booked for 13:36 UTC was released, not run: the D3 and D4 live checks would have failed because their test sessions were ended at 08:01 UTC under the Hive rule. The next certify runs when the open items close; no time is set.

Placement is held at the Architect's direction (12:58 UTC). Menu wiring and the first real menu start wait until the M1-M4 parity matrix and the supporting-process coverage pass on all four Macs. The placement launcher was rolled back on M1, M2 and M3 and stays undeployed. Defect fixes continue (PM ruling 15:0x UTC). Earliest wiring on all four Macs: about 18:15 UTC, if the owners' estimates hold.

500 sessions. New per-Mac caps went live at 15:5x UTC: M1 180, M2 238, M3 32, M4 52 (502 in total). With the memory-reserve items landed (15:57 to 16:01 UTC), live admission is 403: M1 82 (limited by disk), M2 238, M3 32, M4 51. Reaching 500 needs M2's sessions on a terminal server with the raised open-file limit: route A restarts M2's server and is timeboxed to 18:00 UTC; Route A was not ready at 16:16 UTC (the Architect is attached to 3 M2 sessions and 12 are unsaved). If it has not happened and been certified by 18:00 UTC, route B (a second terminal server on M2) is built in scratch only; nothing of it deploys while the parity hold stands.

Deliverables and tasks

Deliverable / task State Execution type Dependencies Agent / model executing Time elapsed Estimated completion
D1 Decision record CERTIFIED
4 certifies, latest 08:00:21 UTC
Complete None open None (done) not recorded Done; re-checked at the next certify
Today's rulings kept as ruling files: panel rulings on the L8 launcher carry (12:0x), the M3 memory reserve (14:2x) and the route to 500 (14:3x); the PM's ruling that the M2 proof runs on a private terminal socket (11:0x) RECORDED Single tasks None Panel (Grok, OpenAI, Gemini, 2 of 3); for the 12:0x ruling Grok's seat dropped on a credit limit and OpenAI and Gemini agreed. PM for the 11:0x ruling not recorded Done
D2 Live telemetry on M1-M4 CERTIFIED
4 certifies, latest 08:00:21 UTC
Complete None open None (done) not recorded Done
Dashboard selftest window. Window 1 (11:56 UTC) closed with no selftest run: the Pi lane's capture was blocked at sign-in. Window 2 opened 13:05:32 UTC; the selftest ran OK at 13:10:31 UTC; access removed and the dead-man disarmed at 13:11:29 UTC DONE 13:11 UTC Dependent (Hive release, then the dashboard lane's go-ahead, then the filed access change) None now Plan owner with the infrastructure dashboard lane and the Pi lane Window 2: 6 min Done
Session-admission disk check reads the free space macOS reports for important use on M1 and M2 (raw figure on M3 and M4) LIVE M1-M4 since 08:41 UTC Dependent chain (build, audit, deploy M1, then M2-M4) None now not recorded Build done 08:20, all four Macs live 08:41 UTC Done
D3 Session placement working live on the Macs CERTIFIED 05:00:19 and 08:00:21 UTC
WIRING HELD Architect, 12:58 UTC
Certified; follow-up work runs as concurrent lanes, each a dependent chain Architect's hold: parity matrix and supporting processes pass on all four Macs (see below) Build, audit and executor agents; model not recorded Original proof 02:21 to 03:34 UTC (1 h 13 min) Wiring on all four Macs earliest about 18:15 UTC (contingent)
M2 opened for placement (L6): opened 12:49:26 UTC; proof session placed through the real path on M2 (private socket) ended clean at 13:00:54 UTC; independent verify passed 13:12 UTC DONE 13:12 UTC Dependent chain (deploy, M2 tracker check, driver, open, post-check, proof, verify) None now M2 open executor and its verify stage; model not recorded 11:25 to 13:12 UTC (1 h 47 min) Done
Lease driver set to all four Macs (supporting row P1): set 13:54:06 UTC; post-check passed 13:54:51 UTC (34 M1 sessions adopted, none over cap); negative control failed first as required DONE 13:54 UTC Dependent (stage, audit, apply) None now Build and audit agents; model not recorded 13:2x to 13:54 UTC (about 30 min) Done
Balanced placement r16 (a session starts on the Mac with open capacity, as the Architect directed), carrying the L2 and L3 fixes. Audit round 5 passed 14:2x; fix 6 done 15:1x; audit 6 failed narrowly 15:4x; fix 7 running FIX 7 Dependent chain (build, audit, fix); concurrent with the other lanes Audit pass, then landing; deploying it is wiring, so it waits for the Architect's hold Build agent and independent audit agents; model not recorded 11 h 14 min (since 05:15 UTC) not recorded (last stated about 15:00 UTC, passed)
Remote start onto M3 and M4 (supporting row P9). Fix 2 done 16:13 UTC (the router is the only product change); audit 3 passed with conditions; fix 3 running, then a narrow audit 4, then landing the helper and router. Deploy items held until the change gate can classify the placement folder; config and launcher items need real start proofs on M2, M3 and M4. Starts onto M1 refused until the M1 folder-trust answer (P13) FIX 3 Dependent chain (design, build, audit, fix, land) Gate classification of the placement folder; P13 for M1 only Build agent and independent audit agents; model not recorded 3 h 5 min (design written 13:24 UTC) Build about 17:00, audit and land about 17:45 UTC
Session start from the v200c menu goes through placement (L8). Landed and deployed on M1 at 12:25 UTC; rolled back at 12:58:26 UTC for the Architect's hold; M2 and M3 rolled back 13:0x UTC. All four Macs run a launcher without the placement change HELD Dependent Architect's hold lifted; P9 landed; launcher redeployed to M1-M4 from origin; one real menu start not recorded Landed 12:25 UTC; held since 12:58 UTC Earliest about 18:15 UTC (needs the save-continue lane's graceful-quit fix on M3 and M4)
D4 Sessions survive a real restart of M4 CERTIFIED
08:00:21 UTC
Certified; restart follow-up in flight None open for the certificate of record Drain build, audit and install agents; model not recorded Attempt 3: 06:15 to 06:42 UTC (27 min) Done (real restart 06:34:48 UTC, 4 of 4 sessions recovered, independent check 06:47 UTC)
D4 live check at the next certify: its test sessions were ended at 08:01 UTC, so the next certify needs a fresh M4 restart run or re-placement records NEXT Dependent Next certify booked not recorded Not started not recorded
Placement drain hook for Pi-driven Mac restarts (Architect direction 13:5x UTC: the Pis may restart whole Macs). Built 14:2x; audit 1 failed; fix 1 done 15:28; audit 2 passed 15:5x UTC. Install paused at 16:01 UTC for a shellcheck fix and resumed 16:07 UTC. Done: the ten landing filings, INSTALL-HOOK-M6 (CGR-M7-007369, post-check passed, confirmed by the Pi lane), the wording fix and the inert status and dry-run checks. Waiting: the Pi drain keys on M6 (no write to M6's key file 16:30 to 17:00 UTC during the Pi lane's key-narrowing canary, and another session's lock on that file to 16:50 UTC). A further fix round runs in a separate folder INSTALLED keys waiting Dependent chain (design, build, audit, fix, install, keys) Pi lane deploys its restart-lease store on pi-01 (waits for a freeze release there, or 19:30 UTC) Build, audit and install agents; Pi lane for the Pi side; model not recorded About 2 h 34 min (since 13:5x UTC) Pi drain keys earliest 17:00 UTC; drain usable about 30 min after the pi-01 store deploys
D5 500 concurrent sessions CERTIFIED 3 certifies, latest 08:00:21 UTC (measured ceiling)
500 LIVE: OPEN 403 at 16:01 UTC
Concurrent levers, each a dependent chain Route A (M2 terminal server restart), timebox 18:00 UTC; else route B Caps and reserve builders, landing executor, M2 parity lane (lane 28); panel for the route Original ramp 00:11 to 00:27 UTC (16 min) Route A: after the M2 restart (timebox 18:00 UTC) and the M2 cap re-derive. Route B: ready to land about 02:00 to 03:00 UTC on 19 Nov, live after the Architect lifts the parity hold
Real per-session cost measured on M4 (L14): 20 idle sessions; memory 0.138 GiB per added session; idle CPU 0.0041 cores per session. The Architect (via the PM) ruled memory is not the limit and directed the caps be re-derived DONE 12:25:53 UTC Single run None now not recorded 11:56:39 to 12:25:53 UTC (29 min) Done
Per-Mac caps re-derived (memory, terminals, processes, terminal-server files) and deployed, 11 filed items: M1 180, M2 238, M3 32, M4 52 = 502 LIVE 15:5x UTC Dependent chain (build, 4 audit rounds, land, deploy) None now Caps builder, audit agents, landing executor; model not recorded 12:5x to 15:5x UTC (about 3 h) Done
M3 and M4 memory reserve items: all 4 landed and deployed, the last three from 15:57 to 16:01 UTC (CGR-M7-007247, CGR-M7-007249, CGR-M7-007253) once the reserve sheet carried its required disclosure; all post-checks passed, no rollback. Live admission 403 LIVE 16:01 UTC Dependent (disclosure added, then landing) None now Reserve builder, landing agent; model not recorded Rebuilt 15:0x, landed by 16:01 UTC Done
Open-file limit for new terminal servers raised from 256 to 8192: M3 13:53:47, M4 14:06, M2 14:17 UTC. M2's running terminal server keeps 256 until it is restarted DONE 14:17 UTC Concurrent per Mac None now M2 parity lane (lane 28) not recorded Done
Route A: restart M2's terminal server onto the new limit, then re-derive the M2 cap (238 to about 348; projected fleet 502). Waits on all its conditions: every M2 session owner saved with no live work, no attached client, a fresh readiness signal, filed. NOT READY at 16:16 UTC: the Architect is attached to 3 M2 sessions and 12 are unsaved NOT READY open to 18:00 UTC Dependent chain (readiness, restart, re-derive) All restart conditions met; the re-derive refuses any server started before 14:17:12 UTC M2 parity lane runs the restart; OBJ-04 runs the re-derive Not started Timebox 18:00 UTC
Route B (only if route A is not done and certified by 18:00 UTC): a second terminal server on M2 for new sessions, built from 18:00 UTC in scratch only; no deploy while the parity hold stands (its 4 blocking consumers are parity-held files). Design note done 14:5x UTC; about 26.5 h of blocking work PREPARED Dependent 18:00 UTC timebox; the Architect lifting the parity hold for go-live Design-note agent (done); model not recorded Design only Ready to land about 02:00 to 03:00 UTC on 19 Nov if started at 18:00 UTC
M1 disk room: Time Machine local snapshot deleted (13:2x UTC, important-use free 105.8 to 133.6 GiB); its daily backup job retired (13:18:05 UTC); 44 folders archived to M2, each checksum-verified before and after (32.10 GiB freed; M1 about 211 GiB free). M1 admission was still disk-bound at 82 sessions at 15:52 UTC DONE 14:5x UTC Concurrent None now Archive agent; model not recorded not recorded Done
D6 Reversibility and durable record CERTIFIED
4 certifies, latest 08:00:21 UTC
Complete; kept current as work lands None open Plan owner not recorded Done
Reversions used today, each a filed change: placement launcher rolled back on M1 (12:58:26 UTC), M2 and M3 (13:0x UTC); M1 disk floor lowered to 90 GiB (12:53 UTC) and restored to 100 (13:2x UTC). Old slot-claim program kept on each Mac; the M1 archive moves have restore scripts, and the folders cleared for deletion carry a delete-after date of 2026-12-18. The plan owner's running lane record is kept append-only DONE As each change lands None Plan owner not recorded Ongoing

Work in flight

Work State Execution type Dependencies Agent / model executing Time elapsed Estimated completion
Menu wiring on all four Macs (placement through the onboarding menu) HELD Architect, 12:58 UTC Dependent Parity matrix and supporting-process rows below; P9 lands (about 17:45 UTC); launcher redeployed to M1-M4; one real menu start not recorded Held since 12:58 UTC Earliest about 18:15 UTC (contingent)
Next certify, aiming for COMPLETE HELD Dependent Open items closed with real evidence; fresh live evidence for D3 (a live placed keep-alive session) and D4 (a fresh M4 restart run or re-placement records) Completion gate (automatic checks, not a model) Not started not recorded (forecast only once items close)
Gitea canary session on M1 (OBJ-04 runs the canary kit for the integrator lane); it runs one filing from that lane's pre-filled command before its edit SCHEDULED Dependent The integrator lane's order revision 4 and its checksum (by 18:15 UTC); kit re-pinned to it Canary kit; model not recorded Not started Start 18:40 to 18:45 UTC
Plan-owner session restart (planned save-and-continue; the successor resumes from the lane record) PLANNED Single step After the Gitea canary starts Plan owner Not started About 18:50 UTC
This progress page: first publish. Attempt 1 (CGR-M7-007424, accepted 16:22:39 UTC) created the site, but its built-in check read the custom domain once, a second after attaching it, and stopped; the filed rollback removed everything cleanly at 16:24:04 UTC. The check now waits for the domain to become active; re-filing follows an independent audit RE-FILING Dependent chain (fix, audit, check, file, publish, verify) New filing accepted Publish agent and independent auditor; model not recorded Attempt 1: 16:22 to 16:24 UTC not recorded

Open items

Item What it is Owner State What closes it
L2 A corrupt lease file can make the placement program fail and the session start locally, ignoring the cap. Normal writers never produce such a file OBJ-04 OPEN fix built into r16; r16 audit not yet passed r16 passes audit, lands and deploys; re-certify
L3 An old dead lease can briefly make a new one count twice, giving a false refusal one session short of the cap. Errs toward refusing, never overcommitting OBJ-04 OPEN fix built into r16; r16 audit not yet passed Lands with r16
L4 The launcher's placement step needs a standard macOS tool (perl); Ctrl-C during a decision. Ruled 10:5x UTC: Ctrl-C skips placement and starts locally OBJ-04 HELD with L8 (Architect's wiring hold) Fixed inside the L8 launcher change; closes with its deploy and a real menu start
L6 M2 not open for placement OBJ-04 EVIDENCE IN opened 12:49:26 UTC, proof 13:00:54 UTC, verify passed 13:12 UTC Resolution recorded at the next certify
L8 Starting a session from the v200c menu does not yet go through placement OBJ-04 HELD landed 12:25 UTC, rolled back 12:58 UTC for the Architect's hold Hold lifted, launcher redeployed to M1-M4, a real menu start (earliest about 18:15 UTC)
L14 Live admission below the 500 target. Caps total 502; live admission 403 at 16:01 UTC (M1 82, limited by disk; M2 238; M3 32; M4 51). The 500 target is the Architect's and is not waived OBJ-04 OPEN route A timeboxed to 18:00 UTC M2 terminal server on the raised limit and the M2 cap re-derived (route A), or route B; then admission proven live
L1 Sessions placed on M2-M4 cannot yet save and restart there. The 13:0x UTC M2 attempt took a snapshot that was incomplete (session not pinned) and did no restart; restart and read-back on M2 remain unproven M2 parity lane and Hive lane OPEN The owning lane's real save-and-restart of a pinned session on M2-M4
L5 Save-and-restart does not work on M3/M4 yet. The save-continue lane said at 10:4x UTC this is not in its plan and not today; it first needs the local Hive block and session-tracker hook on M3/M4 Save-continue lane and M2 parity lane OPEN not today, per the owning lane The owning lane's sign-off and one save-and-restart per Mac
Checker fix A check script misread a grant whose reason had spaces when M3 was opened. Fix built and audited in a new batch; the old batch is kept intact as the record OBJ-04 UNRECORDED M2 opened 12:49 UTC; the record does not say whether the fixed checker was used Used in a Mac opening
L7, L9, L10 Scope statements, not defects: the ceiling was measured on M2 only (M3/M4 are projections); the dashboard's session-admission action is deliberately off; one isolation check was reworded to what it proves OBJ-04 DISCLOSED Closed only by real new evidence

L11, L12 and L13 recorded that D5, D4 and D3 were not proven on the evening of 17 Nov. All three were overtaken by the certifications below.

The Architect's wiring hold: parity and supporting processes

Set 12:58 UTC: no menu wiring and no first real menu start until the M1-M4 parity matrix and the supporting-process coverage pass. The Architect also asked for a Mac column in the onboarding menu. No partial (M1/M2-only) wiring. Estimates below are each owner's, as recorded.

Row Gap Owner State / estimate (UTC)
P1Lease driver served M2 and M3 onlyOBJ-04CLOSED 13:54:51
P2M2 session puller copied active rows only (ghost M1 sessions on M2)STW 48DONE on M2 (14:3x report)
P3 / matrix 7bRestart scripts on M2-M4 differ from originSave-continue laneAbout 16:00; M3/M4 pass only with that lane's graceful-quit fix
P4Reapers, janitor, idle and auto-pause jobs, PM reconcilers run on M1 onlyLane 28About 17:00 (janitor part held for a ruling)
P5 / matrix 8bM3/M4 session rows did not reach the M1 rosterSTW 48 (plus lane 28 install)P5 VERIFIED 15:39:08, using an OBJ-04 canary session on M3 (15:29 to 15:38); lane 28 install part 16:00 to 17:00
P6, P7Architect-message, teamwork and conflict-check see M1 (and M2) onlyLane 28 (PM slot)About 15:00; closure not yet in the record
P8Save-continue snapshot and restart pieces on M1 onlySave-continue laneAbout 16:30
P9 / matrix 11bRemote start reaches M2 onlyOBJ-04FIX 3 audit 3 passed with conditions; last recorded estimate: land about 17:45
P10aConsole sessions table shows M1 onlyConsole lane (abbiui 95)Release 3, 16:30 to 18:00
P10bConsole PM and Monitor tabs and the session dashboard show M1 onlyInfrastructure dashboard laneAfter today's certificate
P11No tracker push job on M2; own-host filter on M1STW 48Filter about 15:20, M2 push 16:00
P12Session-close poller per hostSTW 48DEFERRED panel 3-0
P13On M1 the current Claude build does not keep folder trust for the home folder, so manual starts stop at a trust promptLane 28; the setting is the Architect'sAWAITING the Architect's yes or no; M1 remote starts and M1 wiring depend on it
P14Blank original-session field on M1 root rowsSTW 48CLOSED 14:01, by design, no change
P15M2, M3 and M4 have no session-launch helperLane 28no estimate recorded
Matrix 7cOn M2, M3 and M4 the save-continue snapshot row goes to the local Hive onlyHive lane (or save-continue lane)About 16:00; worst case 17:15
Mac columnMac column in the onboarding menu (the Architect's request) and a cross-host menu listingM2 follow-up lane; lane 28 carries it to M2-M4Owned; the lane reports landing
Matrix 4d/5b5 resumed M2 sessions had no boot prompt and no MCP toolsLane 28 (owners restart)About 15:30; closure not yet in the record
Matrix 4c, 9, 12b, F4a/F5Local-Hive forwarder, terminal versions (per-host exception), file-resolver 403, M3 shell setup and authLane 28; PM ruling for 9CLOSED (F4b closes with P9)

Certification history

When (UTC) Result Deliverables certified Declared
2026-11-17 21:51CertifiedD1, D2, D6not recorded on this page
2026-11-18 02:00:20PASS, all gates, 0 denialsD1, D2, D5, D6PARTIAL
2026-11-18 05:00:19PASS, all gates, 0 denialsD1, D2, D3, D5, D6PARTIAL
2026-11-18 08:00:21PASS, all six gates, 6 of 6 live checks, 0 denialsD1, D2, D3, D4, D5, D6PARTIAL
2026-11-18 13:36 (booked)Released, not run: the D3 and D4 live checks would fail (test sessions ended 08:01 UTC) and the result would still be PARTIALNoneNot run

Times are UTC unless marked EST (UTC minus 5 hours); "13:0x" means a time recorded only to the ten-minute mark. Sources: the plan owner's running lane record (as read at 16:29 UTC) and the supporting-process owner and estimate table (to 15:5x UTC). Fields shown as "not recorded" were not in those sources and are left unknown rather than guessed. Where the record does not name a deliverable for a piece of work, this page groups it under the deliverable it serves. This page shows state; it does not rule on completion. The final close needs a COMPLETE certificate and the Architect's own typed confirmation.