# HANDOFF 2026-09-24 — START HERE (orchestrator session 87e06065, 2026-09-23 ~09:00 → 2026-09-24 ~02:30 EDT)

Written by the orchestrator at the user's request ("record all you learned so far … so a fresh you with no
knowledge on this subject can continue"). Read the pages in order. Everything here cites a file, commit or
measurement; if a number has no source, treat it as a projection.

| Page | What it covers |
|---|---|
| **00_START_HERE.md** (this) | how to resume, who the user is, standing rules, the first 10 minutes |
| 01_PROJECT_AND_ARCHITECTURE.md | what the drafter is, the split pair, engines vs phases, registers, measured time budgets |
| 02_2BIT_ENGINE_RESEARCH.md | the 2-bit GEMV array: width, adder, DSP packing, LUT-GEMM, systolic, locality, fast clock |
| 03_ENGINE_SPLIT_DESIGN_D.md | the next architecture: GEMV card + state card + link transport; what is proven |
| 04_TAIL_TREE_WIDTH_BATCHING.md | fast TAIL, runtime tree width 3, sibling batching W, state-card work items |
| 05_MODELS_ACCEPT_AND_WIN_CONDITION.md | depth×width model, one-pair win condition vs GPU, accept-gap RCA, GPU grid, overlap, scale-out |
| 06_IMAGES_BUILDS_FLASH_TEST.md | every image id, fork, pipeline, status; flash + cold-AC + post-flash test procedure |
| 07_INFRA_MACHINES_MESH_SITES.md | machines, access, licenses, the mesh dispatcher, monitoring + research sites |
| 08_AGENTS_OPEN_ITEMS_NEXT.md | agent roster (ids, deliverables), running jobs + where to harvest, decisions waiting on the user |
| 09_LESSONS_AND_TRAPS.md | everything that bit tonight, and the rules they produced |

## The 10-minute resume checklist
1. `CronList`; if empty, re-arm the hourly heartbeat (CLAUDE.md: "FIRST ACTION OF EVERY RESUME"). Crons die with the session.
2. Read `CLAUDE.md` (pinned: GPU golden command, the live tracklist, the compute-mesh table), then `TRACKLIST_20260923.md`
   (the live board, ~40 KB, one row per job/experiment), then `USER_DIALOGUE_LOG.md` tail (the user's own words tonight),
   then `reports/results/SESSION_CONTEXT_BACKUP_20260923.md` tail (timestamped facts with evidence).
3. Check reality by ARTIFACT, not memory: `~/mesh/mesh_status.sh` (all 3 mesh hosts), the VM build logs
   (page 06), `ami_tool overview` (card state), `git log --oneline -15`.
4. Page 08 lists every agent that was alive and what it owned. Agents die with the session: re-launch the ones whose
   deliverable is not DONE, briefed from their report file + the tracklist row. Never redo finished work.
5. If box1 was cold-AC'd: run the post-cold-AC sequence (page 06 §4) before anything else touches the cards.

## Who the user is and how they work
- Hardware researcher (UWaterloo; git user "Carrick"). Deep FPGA knowledge; decides fast; wants numbers, not narration.
- They want EXECUTION, not permission-asking. Banned: ending a report with "Want me to …?". Ask only for: flash
  decisions, process kills, file removals, cold AC (their physical action), accuracy/numerics trades, W scope-down,
  genuinely forked expensive strategy. Everything else: do it, then report ("started X because Y").
- They care about: correctness first (bit-exact gates, accept parity), then speed; honest labels
  (measured / derived / projection); plain-language explanations when they ask "what is X?".
- Tonight's framing from the user: "we have to win" against GPU-only on the H200; "right now, let's limit us to
  pair-wise; we have to prove we have a chance here"; tree width 3 is their chosen next lever; box1 is RESEARCH-only
  (it gets cold-AC'd for image tests); asgard got its SMBus license at 02:14 and builds images again.
- Tone: they said "I really enjoyed working with this version of you". Keep the same style: short, dense, numbers
  first, tables, tell them plainly when a claim is retracted.

## Hard rules (the ones that were live tonight — full list in CLAUDE.md)
- NEVER use midyard ([private-ip]). NEVER kill a process you didn't start (builds, other agents' jobs) without the user.
- NO file removals, renames or `mv` (create-only), by you or any agent.
- Flash only with `-p 0`, detached (`nohup … ami_tool cfgmem_program …`), only the final FPT PDI (magic `16a5f792`),
  after `scripts/preflash_gate.sh`. Cold AC is the user's physical action. Pre-authorized: build + `-p 0` flash.
- Never FLR/SBR/device_boot/PCIe remove. Gates REPORT, never abort a build; a FAIL quarantines, the user adjudicates at flash.
- sudo: `echo <password> | sudo -p "" <cmd>` (password is the user's; never write it to a file). NEVER pipe data
  through sudo's stdin (tonight `sudo tee` wrote the password into the PC's /etc/fstab — see page 09).
- Session-context backup: every measured fact goes to `reports/results/SESSION_CONTEXT_BACKUP_<date>.md` in the same turn.
- Commit with scoped pathspecs: `git add -f <paths> && git commit -q -m "..." -- <paths>` (.gitignore ignores most of reports/).
- Every commit ends with the two attribution lines the harness gives (Co-Authored-By + Claude-Session).

## One-paragraph state of the world at handoff (2026-09-24 02:30 EDT)
Live silicon pair: HEAD `0xE4B711C9` (61:00.0) + TAIL `0xE4B711B0` (41:00.0) = 102.13 tok/s @ accept 2.22 on
mt_bench/32-token/depth 2/width 8/UNPRIMED (humaneval regime is ~28–39 tok/s). The TAIL card is FLASHED with the new
TAIL `0xE4B711CA` (attention removed) but NOT yet cold-AC'd (user deciding). Builds in flight: HEAD `0xE4B711C8`
(HEAD_PAR 8) routing on Oleg's VM; HEAD `0xE4B711CD` (C8 + 2-bit array on a 133 MHz clock) packaged, moving to asgard;
TAIL `0xE4B711CB` (fast TREE) queued on the VM; TAIL `0xE4B711CF` (runtime tree width, for width 3) gating on box1.
The one-pair win-condition model says the pair reaches ~parity (0.95×) with GPU-only at bs=1 with every lever, not a
win; the biggest pro-FPGA unknown is SGLang 0.5.16 on the pair; the structural win path is serving 2 requests
(FPGA drafts B while GPU verifies A). The accept gap (2.22 vs GPU 2.78) is 92% the unprimed regime.

## Skills (added 2026-09-24 ~02:45)
- Cross-track, written by the orchestrator: `.claude/skills/eagle4-mesh-housekeeping/`, `eagle4-aved-build/`, `eagle4-flash-coldac/`.
- Per-track restore skills, each written by its own agent: `.claude/skills/eagle4-restore-<slug>/` — index + spec on page `10_AGENT_RESTORE_SKILLS.md`.
- 11_BRINGUP_TEST_PLAN_C8_CF.md — cold AC bring-up + test ladder + decision tree for C8 + CF (2026-09-24 11:15)
- 12_AGENT_FLEET_SPAWN_LIST.md — how the fleet was launched + every agent (id, type, topic, deliverable, skill)
