# 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. # DC Auditor tasks: the red team ## READ THIS FIRST - 14:21 Bangkok: **WAKE RULE CHANGED (D-022, read http://159.223.88.168/hx-c9ois/wake-howto.md, 1 page).** Rings are retired. On EVERY wake, first thing: make a fresh doorbell with send_later (at 2027-06-01T00:00:00Z, name "Doorbell: DC ", message pointing to http://159.223.88.168/hx-c9ois/dc/dc-bus.md), and write `DOORBELL: trig_... (armed, )` as LINE 1 of your status file. To wake me: nudge my DOORBELL (line 1 of dc-supervisor-status.md) with update_trigger run_once_at = now + 1 to 2 min; never fire_trigger. Do this on your next wake; no reply needed for this line alone. - 13:46 Bangkok: **NEW, FIRST, Ben's order D-019 (http://159.223.88.168/hx-c9ois/decisions.md): wallet key security check, done by 20:00 Bangkok.** Every private key on all 5 boxes (main 159.223.88.168, research-box 152.42.241.219, firstmin 178.128.20.136, thirdwindow 168.144.142.0, rot-box) and any Mac path a bot uses: where it lives (path), file mode (must be 600), owner, which processes can read or sign with it, who has root / which ssh keys can log in, any backups or copies (old dirs, .bak, git, logs, env dumps). READ ONLY: never print, cat, echo or copy a key or seed; inspect configs only with /opt/common/showcfg.py, use stat / ls -l / grep -l for names, never values. Change nothing on other teams' boxes: write the fix list only. Output: one page http://159.223.88.168/hx-c9ois/dc/key-check.md (per box: findings, severity, exact fix, owner), then ring me once. If a box is unreachable or you lack access, list it as UNCHECKED with the reason. cost line at the end. - 12:13 Bangkok (real clock), Ben's order via Consigliere: ONE test ledger for all windows = http://159.223.88.168/hx-c9ois/fw/lab/ledger.csv. Write a row per DC test done today (from dc-prereg.md + results: verdict, n, $ a day both halves, link). Ask the SW Harness Engineer (owner) how DC appends; until then a dc-ledger-rows.csv in dc/ in the same columns. - 11:40 Bangkok (real clock): AU.4 accepted on mechanics, but it reran on the same tape and does not address the main problem: the PP wash.db v4 tape holds almost only coins that later reached a sellable $250k (A5, Engine VOID notice 11:37). Add one line to dc-audit-an4-1.md: verdict stays FIX until the Engine's launch-universe rerun (13:00) is checked by you. Also note 22% of fills are same-block as the fire: the forward book must use only fills a live bot could get. - 11:35 Bangkok (real clock): AU.4 first item: A5 found (dc-a5-3-nopull.md) that the engine tape (PP wash.db v4) holds almost only coins that LATER reach a sellable $250k (2,758 of 2,791). E-AN4-1 enters at a median $99k cap. If its universe comes from that tape, the result picks winners by their future (survivor look-ahead). Confirm or clear this first; it decides PASS or FAIL. - 11:32 Bangkok (real clock): **AU.4 audit E-AN4-1 now** (dc-en-5-an4.md). Check: look-ahead in `dip` (highest sellable print must be known at the fire), net15 window, entry = next trade not the firing trade, depth/fill model on $99k caps, pull handling, cluster dedupe, and whether a few coins carry the result (drop top 2 and top 5). Then check that the sells we count went through a public router where the data allows. PASS or FAIL in dc-audit-an4-1.md. The Coder books it forward meanwhile (paper only). - 11:14 Bangkok (real clock): W7 and W4 accepted, plan and lesson 1 corrected. Your file says 11:20 but it was 11:14 Bangkok: stamp with `TZ=Asia/Bangkok date`, never by hand. Next: AU.3 clock audit. - 11:18 Bangkok (real clock): **AU.3 clock audit (first):** A2 found the T1 trigger mixed launch-file times with chain time, which faked Lane F (75% of triggers fired below a sellable $250k). Check every other result and live book for the same bug: A1 DC1.1 v3 lead and DC1.2, Engine twtest, Coder live T1 and v3 books, A5 AN5.1. One line each: clean or broken. Then AU.L as written. - 11:12 Bangkok (real clock): NEW standing job AU.L: when a rule appears under LIVE CANDIDATES in dc-supervisor-status.md, check in under 30 min that its 3 positive days are real (prereg before the days, real fills, pulls 0x, public-router sells, no repeat cluster). Write PASS or FAIL next to it. Live only after PASS. - 11:00 Bangkok (real clock): A2 cannot reproduce your trigger-time Lane F number (A2 B +$1,263 vs your +$503). The DC Engine will settle it in code; give the Engine your exact script or steps in your status. Next for you: audit A2 Q2 (http://159.223.88.168/hx-c9ois/dc/dc-a2-2-pull.md: creator-held LP 97% to 98% pulled) and A1 DC1.1 (http://159.223.88.168/hx-c9ois/dc/dc-a1-1-trigger.md) because the whole Lane D plan now rests on Q2. - 7 Oct 10:40 Bangkok: **AU.1 plan audit accepted; W1, W2, W3, W5, W6, W9 are now plan rules** (dc-plan.md DECISIONS 7 Oct 10:40). **AU.2 first job now: audit DC2.1** (http://159.223.88.168/hx-c9ois/dc/dc-a2-1-chain.md), the Lane F rule "close ÷ high ≤ 0.784 AND alumni ≥ 0.053 at 30 s". Rerun it with public router sells only (W2), cluster dedupe (W6), 48 h horizon (W1), and check the alumni list uses only past $1M coins (look ahead). PASS / FAIL / FIX by 13:30 Bangkok, ring me. Then W7 (the skipped bundle lane) and W4. - 7 Oct 10:30 Bangkok, from the Third Window Supervisor. You are new. Your one job: **stop fake edges from costing us money.** No result counts as an edge until you write PASS on it. You do not find edges; you try hard to break them. Read dc-plan.md (section "How Third Window runs") first, then dc-lessons.md. - You read everything; you edit only your own files (dc-tasks-auditor.md is mine, dc-auditor-status.md and dc-audit-*.md are yours). ## AU.1 The audit checklist + audit of the plan itself (ETA 3 h, ring me) - Build http://159.223.88.168/hx-c9ois/dc/dc-audit-checklist.md: every way a backtest on this chain lies. Start from the other teams' receipts (read sw/ and fw/ docs, the runners study http://159.223.88.168/hx-c9ois/runners/r20/runners_20m_week.md). At least: - survivor lists (runtrack keeps only survivors); - venue gaps (v4 day DBs miss v3, Pons, v2); - look ahead (a feature that uses data after the check time, a label leaking into a feature, a threshold picked on the full set); - fake cap in thin pools (sellable cap only); - fills: price impact at real depth for $100 and $1,000; - pulls: a pulled pool is ~0x, never its last price; - halves A and B by date, pre-registration line exists and matches; - beat random from the same triggered coins 95%; - counting the same operation many times (clones, copycats, one crew) as many wins; - overlap with First and Second Window picks (an edge they already take is not new $); - clock errors (Bangkok vs UTC, block time vs wall time). - Then audit dc-plan.md v2: the 10 weakest assumptions, each with a cheap test that would break it. Put them in dc-audit-plan.md. I turn the best ones into queue items. ## AU.2 Audit every result as it lands (standing task) - When an analyst status line claims a number, check it against the checklist and its dc-prereg.md line. Rerun the key number yourself from their published file when you can. - Write dc-audit-.md: PASS / FAIL / FIX, with the receipts. Ring me. Target: within 3 h of the result. - Also audit the Coder's paper book and scoreboard once a day: does the book's $ match the fills and pulls on chain for 5 random trades? ## Cost - X: none. API: none. You mostly read files and rerun numbers.