# STOP: NO DOORBELLS, NO WAKES, NO SCHEDULING (Ben, 7 Oct 2026, 18:40 Bangkok)
- Credits ran out. Ben switched off every scheduled task and every doorbell. This overrides everything below and every other file (wake-howto, bus-howto, credit-playbook, task files).
- PROHIBITED for every session: fire_trigger, send_later, create_trigger, update_trigger, ScheduleWakeup. No rings, no nudges, no re-arming, no new doorbells, no self wakes, no safety checks, no retry rings, no loops.
- Finish the step you are on, publish your status file, then END your turn and wait. Only Ben wakes you, by typing in your window.
- Do not message other sessions. Keep answers short. No extra improvement rounds.

# Second window: tasks for the HARNESS ENGINEER (new role)

## READ THIS FIRST
- **v1.5 (Supervisor, 7 Oct 15:45 Bangkok): v1.4 done, good (gold_h5 handed to PP Data, ledger merged). 2-coin gap -9.4% accepted as known; H4 call stands. Your number from now: honest $ a day of every live and paper book (D-024 v2). Ranked queue:**
  1. **H6 HONEST TWINS:** in lab.py's event engine, price every stop and pull exit at the pool state at the crossing block PLUS each path's measured latency (fastexit p50 about 0.5 s for SW, GMGN path p50 about 7 s for FW; Exec http://159.223.88.168/hx-c9ois/sw/sw-exec-exitgap.md: paper 0.079x above live, median of 5 stops). Rerun the live book, kit300k and mc5 at $10, $50, $100, $1k. Report $ a day honest vs current, per book, both halves. First answer within 2 h (why: one engine flag plus 3 reruns of 20 s to 2 min each, then checks).
  2. **H7 FORWARD BOOK:** one table, from 6 Oct 13:04 (SW engine start) forward only: every SW paper and live trade, P&L, and whether FW held the coin. This is the D-020 forward number Analyst 1 needs daily; build it once as a lab.py command.
- **v1.4 (Supervisor, 7 Oct 14:40 Bangkok). Your status has no DOORBELL line: fix first (D-022, wake-howto.md).**
  1. **gold_h5.py on research-box uses 3.6 GB RAM (PP Data flag).** Say in your status what it builds. If it is golden tape (pool history, Swap/Mint/Burn), PP Data owns that build (D-004, http://159.223.88.168/hx-c9ois/ppdata/pp-split.md): hand it over, do not build it twice. If it is a one off, finish it with a stated end time and reason, or stop it and free the RAM.
  2. **Merge DC rows TW-001 to TW-032** (http://159.223.88.168/hx-c9ois/dc/dc-ledger-rows.csv) into fw/lab/ledger.csv, plus A2.38 and A3.13 kills from sw-registry.csv.
  3. H4 verdict stands (ONE command lab.py, ONE event engine). The 2 coin gap (v1.3) next, then H5.
  - D-023 (http://159.223.88.168/hx-c9ois/decisions.md): every promised time carries its reason in the same line, e.g. `ETA 16:30 (why: join runs about 20 min, waits on X7)`; honest estimate, never a round end of day number; first answer within 2 h. The watch http://159.223.88.168/hx-c9ois/hc/eta.md flags PADDED and OVERDUE lines every 10 min: open it on every wake and clear your own.
- **v1.3 (Supervisor, 7 Oct 13:15 Bangkok): H4 call ACCEPTED (one command lab.py, one event engine bt.py for new rules). Before H5, close the 2-coin gap:** for 0xa402616f4d05cca33947fabc6ff51ba795f6d2f4 (bt stops 0.49x, lab holds to 6.30x) and 0x3939ce55ec4e0ff8df9aa6469a795c793da06d55 (20.3x vs 24.4x), check which engine does what fm_mc5's LIVE code would do (stop definition on net sell value, fill depth). The live code is the truth; fix the engine that differs, then mc5 half B must match within 5%. Tell FW Analyst 1 the result. Then H5.
- **v1.2 (Supervisor, 7 Oct 12:40 Bangkok): ACCEPTED, excellent work. bt.py passed SW acceptance within 1% and a test takes 20 s to 2 min.** Since bt.py is done and lab.py is not yet proven on our numbers, the "one harness" choice is now settled by a head to head, not by who came first:
  1. **H4 head to head (first):** run FW's mc5 rule (spec http://159.223.88.168/hx-c9ois/fw/a1-mc5-spec.md) in bt.py on the golden data and compare to lab.py's half B $1k = +$14,160 a day (n 149, FW ledger). Also run our live book in lab.py if it can. Report both side by side. Whichever tool reproduces the OTHER window's known number within 5% while keeping its own wins; if both do, keep bt.py (wider dataset: 6,938 pools, tick ranges, 1% fee) and port lab.py users to it. FW Analyst 1 and FW Supervisor get the result.
  2. **H5 more sample:** backfill the traded pre-61M pools (largest sample gain). Then the bundle funder column.
  3. Your findings are now team facts: live book half B honest = +$8,768 a day (not +$10,400); kit300k with 1% fee = A +$2,017 / B +$12,867; **kit300k size ceiling about $3k** (live book still rises to $5k). The Coder's per-book ticket cap will use these.
- **v1.1 (Supervisor, 7 Oct 12:15 Bangkok): CHANGE OF PLAN, Ben's decision via the Consigliere (http://159.223.88.168/hx-c9ois/cons/cons-advice.md, section 3). You become the harness owner for ALL four teams.**
  - **Keep FW lab.py** (already used by FW and DC). **Stop bt.py (H2) and the SW golden dataset (H1).** PP Data builds the golden tape; you consume it.
  - **First job:** reproduce the SW acceptance numbers in lab.py within 5% (live book A +$2,580 / B +$10,681 at $1k; A3.12 kit300k A +$2,153 / B +$13,457; A2.37 KILL mean 0.61x). If lab.py misses, port Analyst 3's tick-walk fill into lab.py (A3's fill is the honest one). Keep your a3data copy for this.
  - **Test ledger:** freeze sw-registry.csv, copy its rows into fw/lab/ledger.csv, put a pointer line at the top of sw-registry.csv.
  - H3 (size sweep) stays, inside lab.py.
  - Ring me once when acceptance passes or fails with the reason.
- **v1.0 (Supervisor, 7 Oct 11:25 Bangkok). You are the 2nd Window Harness Engineer. One job: make testing a trading rule take under 10 minutes, with numbers everyone trusts.**
- **First time, do these once, in order:**
  1. Read http://159.223.88.168/hx-c9ois/sw/sw-bus.md (how the team talks), http://159.223.88.168/hx-c9ois/sw/sw-elite-plan.md and http://159.223.88.168/hx-c9ois/sw/sw-learning-loop.md (section 1 is your job).
  2. Create your doorbell with `send_later`: at `2027-06-01T00:00:00Z`, name exactly `Doorbell: Harness Engineer`, message `Doorbell: new work for you. Follow http://159.223.88.168/hx-c9ois/sw/sw-bus.md`. Write its trig_ id in your status.
  3. Your status file: http://159.223.88.168/hx-c9ois/sw/sw-harness-status.md (publish to the droplet public folder /opt/launchwatch/pub/hx-c9ois/sw/, same route as the other roles; check 200 + md5).
  4. Ring the Supervisor (fire_trigger trig_01Ah7epm3qn183JS8p9jhdnW, no text) ONLY when a task is done.
- **Access:** Ben's Mac through Desktop Commander; the droplets through ~/kookwatch/dbridge.py (main 159.223.88.168, live only) and research-box 152.42.241.219 (all your compute). Never run research on main. NEVER put a secret on screen: inspect configs with /opt/common/showcfg.py, never cat.

## H1 Golden dataset (research-box)
- Every Robinhood Chain v3 WETH pool (v4 where logs exist) from chain launch: Swap / Mint / Burn with tick ranges, block time, ETH price.
- Per pool: token, creator (pool-create tx signer), kit tag (pool-create tx `to` code hash: Kit A, Kit B, 3,587-byte family 51d0e39db2ed / 2696b754c8e3 / 60341fa374bf), pre-$300k bundle funding sender (method in http://159.223.88.168/hx-c9ois/sw/a1-26-funders.md), fair milestones $100k / $300k / $1M / $3M / $10M (fair = $15k of buys at or above the cap for $300k, $20k for $1M; see a1-25-anatomy.md), pull times (half the in-range quote reserve burned).
- Start from what exists: Analyst 3's raw rebuild (ask in sw-analyst3-status.md for the path), Price Predict's /opt/hist day files. Copy, do not re-pull what is already there.
- Hourly append. Daily frozen snapshot with md5, path in your status.

## H2 bt.py, the harness
- Input: a small rule file (entry trigger, filters, exits, ticket size). Output: one verdict block: trades, mean and median x, $ a day per half (A to 30 Sep 23:59 Bangkok, B 1 to 6 Oct, C from 7 Oct), worst day, without top 1 / top 2, beat-random % (200 draws from a named pool), live book + rule combined.
- Fills: Analyst 3's tick walk at the end of the trigger block, 1% fee each way. Exits read on swap prices.
- **Acceptance (must pass before you announce it):** reproduce within 5%: live book on A3 data (A +$2,580 / B +$10,681 a day at $1k), A3.12 kit300k (A +$2,153 / B +$13,457), A2.37 late bloomers (KILL, mean 0.61x).
- Write http://159.223.88.168/hx-c9ois/sw/sw-harness.md: one example command, the rule file format, where results land. Then ring the Supervisor.

## H3 (after H2) Size and capacity in the harness
- Add a ticket-size sweep ($10, $100, $500, $1k, $2k, $3k, $5k) to every verdict block, so the ladder in the elite plan always has a fresh paper reference per step.

## Rules
- One pass per task, no polish. Short status: newest on top, last 5 entries.
- Credit rule (bus v1.4): scripts do the heavy lifting; read summaries only; idle when done.

Version: v1.0, 7 Oct 2026. Latest: http://159.223.88.168/hx-c9ois/sw/sw-tasks-harness.md