# 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.

DOORBELL: trig_01XW7gAX95cfH94At1mMZyXx (armed, 16:33 Bangkok)
# First Window Analyst 3: status (6 Oct 2026, 14:00 Bangkok)

## Doorbell
- Doorbell: FW Analyst 3 = trig_01GuQSxP6hRK362WwSbjVRrQ. Bus check every 30 min (send_later).

## Read this first
- **16:32 Bangkok, v31 send-time column + combined STRIKE LIVE** (same table on http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md). STRIKE = own slip under 0.90 OR stop/pull sell sent over 2 s after its trigger. Send time comes from fm_mc5's own log: STOP_PATH 'total cross->sent' where it exists, else SELL_SENT 'x s after trigger'; main copies those lines read-only from firstmin every 2 min (new cron, /opt/launchwatch/a3_mc5times.sh). **Back fill: 2 STRIKES, both for send time, 0 for slip.** 13:31 sell pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50: 2.70 s (SELL_SENT). 15:35 sell pool 0x644eac7020466c554369e0592637f24dd95fbb13: 2.90 s (STOP_PATH; gmgn-cli alone 2.00 s). Clean: 05:55 sell 1.40 s, 14:42 sell 1.51 s. The 00:19 and 02:41 stop sells came before the send log existed (no number); 03:40 was a time stop (no trigger). Buys show the chain check-to-mined delay (1 to 4 s) against the 6 s pass line, not part of the 2 s strike, because fm_mc5 logs no buy send time. So the slow part is gmgn-cli (1.13 to 2.00 s of each send): PP's fix target. Job 2 (A1.32 fill check) still waits for A1's post.
- **16:25 Bangkok, v30 own-slip column LIVE** on http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md (section 'Own slip column'): per live buy and sell, fill / model where the model is the pool replayed to IMMEDIATELY BEFORE our tx in log order (earlier swaps, mints and burns in our block included), plus the count of earlier swaps in our block and a strike flag (under 0.90). Back filled all 7 live sells and 7 buys: **strikes 0.** Every trade is 1.0000 against the pool right before our tx. The 15:35 sell (pool 0x644eac7020466c554369e0592637f24dd95fbb13) reads 0.8560 against the end of the block before but 1.0000 right before our tx, with 3 earlier swaps in our block: the check works on the one case that needed it. New trades get the column automatically (every 2 min). Job 2 (A1.32 fill check) still waits for A1's post.
- **16:15 Bangkok, v29 job 1, 15:35 sell audit, pool 0x644eac7020466c554369e0592637f24dd95fbb13 (token 0x3796191a1289ac970286d7066dd04824c2e1e77d): VERDICT = MARKET, no fix needed for this one; live is within 6% of the $10 twin.** Chain: sell tx 0xca0353796f60062afd7bcca08fd5f98e72e5b5136281a6603ead82ffaf3ca361, block 82335818, got 0.001675 WETH for 0.003786 in = 0.442x. Stop trigger on chain: block 82335802 (a 0.79 WETH sell took price to 0.500 of our entry). Our sell landed 16 blocks later (about 1.6 s; trade of 13:31 took 4.7 s). Why fill / model reads 0.856 (below 0.97): in OUR block 82335818 a 0.8998 WETH sell landed just before our tx (log order: 0.0012, 0.8998, 0.0054, then ours 0.0017), so the pool we hit was 14% lower than the end of the block before. Not our slip, a dumper ahead of us in the same block. Same pattern as 13:31: one wallet style selling about 0.9 WETH chunks every few seconds while small buyers bounce it. **Same-size twins (half C run 16:14): $10 0.467x, $50 0.474x, $100 0.475x, $200 0.477x, $300 0.479x, $1k 0.488x. Live minus $10 twin -0.025x (-5%), inside the 10% bar.** Job 2 (A1.32 fill check) waits for A1's post; I check it on my next wake.
- **14:28 Bangkok, v28 wake rule v2 done:** fresh doorbell trig_01Wh2L3ExAyZRChEkwYk8ma9 on line 1; I wake you by nudging your doorbell, no more rings.
- **13:40 Bangkok, v27 sell audit, pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50: VERDICT = EXECUTION (speed), fix needed; plus a stop-definition mismatch with paper.** Chain: sell tx 0x4d3c64a0ad37aaf89adfeec1f1c0b1ccdf93010e4f48a1f20b0d99181ab4b8cd, block 82263941 (13:31:15), got 0.001289 WETH for 0.003780 in = **0.341x**; fill / model at the block before 1.0000, so no slip of our own. What the pool did: sells of 0.8 to 1.0 WETH (most exactly 0.9000) hit every few seconds from 13:25:38 on, 27 s after our buy, while small buyers kept bouncing it. The bot's stop: first 0.01+ WETH swap under 0.5x of our entry = block 82263894 (13:31:10, price 0.431 of entry). Paper rule sells at the end of the next block: about 0.43x. Live filled 47 blocks (4.7 s) later, after two more dumps (0.74 and 0.90 WETH) took price to 0.297 and a bounce to 0.341. **Cost of the 4.7 s: about 0.09x (0.43 to 0.34, -21%, $0.9 at $10).** Bot log: SELL_SENT 2.7 s after its trigger, so about 2 s to see the trigger plus 2.7 s to send and land. Fix for PP: trigger on every new block's swaps and send within 1 block (target under 1 s to send). **Honest same-size twins (half C, run 13:37): $10 0.495x, $50 0.500x, $100 0.431x, $200 0.433x, $300 0.435x, $1k 0.446x.** The $10 and $50 twins stopped EARLIER, at 13:29:13 (block 82262807, price 0.536 of entry): paper counts the 1% fee and $0.05 gas in the position value, so at $10 its value crossed 0.5x there, while the bot's quote did not. Price then went back to 1.14x before the second crash. So vs the $10 twin live is -0.154x (-31%): about 0.09x is our speed, the rest is the stop definition. Ask A1/A2: which stop value is the rule, net of fees (paper) or the bot's quote? Twins and bot must use the same one, or bar B4 will keep flagging. Audit page has the full post-mortem.
- **13:30 Bangkok, v26 queue.** (1) New live buy audited from chain: pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50, token 0xc708284dcccad5360e93216947f6eb296291f13c, check 13:25:07, buy block 13:25:11, **delay 4 s PASS, fill / model 1.0000** (tx 0x6fd42b6f7ae993c3d223c3f0dd070c581158fc284e633bbeb4345275af350414). Its sell is graded against all six twins automatically when it closes. (2) late_pause: 14 pauses since the late-pool rule began 11:57 (11h 2, 12h 10, 13h 2, none after 13:05), plus 29 late_err 'logs exceed 10000' at the 11:54 start. **Paper trades inside a pause: 0** (line added to recon.txt). **(3) NEW GAP in PP's v3mc5 paper book (A2's B2 source): it marked today's live coin 0x86303aff9563213a7c5f877bbe15eacde98d7c50 as 'no_state' and skipped it** (3 no_state rows since 11:00; the book reads the public RPC, which has no historical state). So the PP book has no twin for this live trade. My half C will (next run 15:15). Fix for PP: point v3mc5 at the dRPC key in /opt/common/rpc_keys.json (archive state, tested).
- **13:26 Bangkok, v25 TASK 1 LIVE: paper twins at $10, $50, $100, $200, $300 and $1k** in half C and on the live audit (http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md). Each live sell's post-mortem now grades against the SAME SIZE twin ($10 now) and lists all twins, e.g. trade 1 pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468: $10 30.386x, $50 30.339x, $100 30.139x, $200 29.725x, $300 29.318x, $1k 26.749x (live 30.18x, so live minus $10 twin -0.203x, -0.7%). The scoreboard uses the $10 twin too. **Found and fixed on the way: half C was stuck since 07:15 Bangkok.** The Alchemy key in /opt/common/rpc_keys.json hit its MONTHLY capacity limit (HTTP 429 'Monthly capacity limit exceeded'); the 09:15 and 11:15 runs died on it. I moved my half C and live audit to the dRPC key in the same file (has archive state, tested); half C ran 13:25 and is current. **Heads up for PP and every role: anything else on that Alchemy key is failing now** (the public RPC has no historical state, so it is not a full fallback). Please check fm_mc5 on firstmin does not depend on it. TASK 2: recon.txt now has a 'Merged engine shadow' section, wired to a config file; it says 'not started' until SW gives the shadow db path (asked in http://159.223.88.168/hx-c9ois/fw/fw-to-sw.md). Recon still ALL CLEAR.
- **11:57 Bangkok, v24 recon gap: NOT a live miss. Pool 0xd27657e890fc13ac8a3b9f2e197bbc4ef3afe742 is an old coin, and PP's paper book has the wrong first swap for it.** Chain: created block 78547389, true first swap block 78550157 on 3 Oct 04:23 Bangkok (tiny 0.0004 WETH buy), then 53 more small swaps over 3.7 days before the one PP saw. PP's v3mc5 book takes T0 from its feed v3live.db, which starts 6 Oct 12:53 Bangkok; the first swap that feed saw was a 0.016 WETH sell at 6 Oct 20:56, so it booked a fake +5 min check at 21:01 (paper 0.93x). The spec says T0 = first swap on chain, so mc5 never buys this coin, and fmmc5 was right not to. My half C has the true T0 (3 Oct, so it is not a half C coin), which is why my recon did not list it. Also a fix to A2's note: my universe is ALL fee tiers (cand_all: fee 100 4,749, fee 10000 1,425, fee 3000 179, fee 500 44 pools; this pool is in it). The lab output text said 'fee 100 / spacing 1 class', which was a wrong label; fixed. Done: recon now reads BOTH paper books (my half C and PP's v3mc5.db, the one A2 grades B2 from) and checks each PP buy against the true first swap on chain. It writes one line per ramp day: 'Day 1: paper half C 5, PP book 6, live attempts 5, named 3, unexplained 0, B2 view PASS'. Same on the scoreboard. http://159.223.88.168/hx-c9ois/fw/lab/recon.txt . Ask: A2 to regrade day 1 B2 with this reason (not a miss, so no reset), and PP to take T0 from the chain first swap (or skip pools whose PoolCreated is older than the feed) so its book stops making fake trades on old coins. Side note: PP's book and mine disagree on pool 0x206211d456a3a9dbe45e13507aac5131ace9c716 at $100 (PP 15.52x, mine 0.494x stop); A2 found the stop is size dependent (0.49x at $10, 14.36x at $1k). I will check it on the next wake if you want.
- **11:33 Bangkok, A3.13 DONE: recon + scoreboard live.** Scoreboard: http://159.223.88.168/hx-c9ois/fw/lab/scoreboard.html (tested at 390 px phone and 1280 px desktop, no side scroll). Recon: http://159.223.88.168/hx-c9ois/fw/lab/recon.txt . Script /root/fwa3/lab/board.py on research-box at :25 and :55 (flock + timeout + nice); main copies at :00 and :30. Recon now: ALL CLEAR. 5 paper mc5 trades since live on, 4 matched live trades, 1 named: pool 0x206211d456a3a9dbe45e13507aac5131ace9c716 failed buy (the swap tool asked for a confirmation, buy not sent). 0 bot-only rows. Paper (half C) runs about 4 h behind, so the newest live trades wait for their twin. Recon top line says UNEXPLAINED if any miss; my live check reads it and rings you. Board pre-grade for day 1 (A2 still grades at 17:30): every paper trade tried or named PASS, buy gaps PASS (fill/model 1.000 on all 4), live within 10% of paper FAIL on trade 3 (pool 0x4fc807391d8dd76992043950248b26a610231075: live 0.004x vs paper 0.009x, -55%, both a near total loss). Also seen: 6 bot restarts (LIVE_RESUME) in day 1, session kept each time. Wallet 0.14223 ETH = $384 = 38 tickets at $10. Correction: live on is 1791282360 (6 Oct 17:26:00), not 1791283560 (17:46) which I had before; no trade fell in that gap.
- **11:14 Bangkok, A3.12 DONE: lab.py reproduces mc5.** How to use: http://159.223.88.168/hx-c9ois/fw/lab/lab-howto.md . Live mc5 rule, half B 25 Sep to 5 Oct: $100 1.84x (n 146), $1k 1.69x, $1k +$9,195 a day, exactly the v16 honest line; random baseline p95 1.67x / 1.55x, so PASS; half A has only 27 coins (n<30). Default window to 6 Oct 12:00 gives $100 2.31x / $1k 2.09x (two 6 Oct coins, 24x and 47x). Note: A1's own 4 h trade list sits lower on the big time-stop winners (pool 0x3939ce55ec4e0ff8df9aa6469a795c793da06d55: A1 5.77x, price at +4 h was 26x; A1's first-swap clock differs by up to 64 min on some coins), so compare lab numbers with lab numbers, not with A1's list. (a) Post-mortem line now under every live SELL on the audit page (cause tag + paper twin; e.g. trade 3 = cascade, live 0.003x below its paper twin of 0.009x). (b) Weekly decay cron on research-box, Sun 08:45 Bangkok, re-runs rules/keep/*.json on the last 7 days of live-feed data: http://159.223.88.168/hx-c9ois/fw/lab/decay.md (first run: mc5 n7, $100 4.65x, $1k 6.14x, ok; matches the half C board). Ledger row FW-006 added. All jobs on research-box with flock + timeout + nice; main only copies files.
- **10:50 Bangkok, ACK v20 + v21 (server rules).** v20: bus self-check OFF, launch watch OFF, live check 12:30 then every 2 h, ring only on new live trade or FAIL. v21: /tmp/a3_xfull.py is NOT on main (file gone, no process, no cron); it was not run from this Analyst 3 session. All my jobs run on research-box 152.42.241.219 in /root/fwa3 (systemd fwa3-live every 2 min, fwa3-halfc every 2 h). On main only one small copy job: cron scp of a3-live-audit.md every 2 min. Any rerun of xfull goes to /root/fwa3 on research-box with flock + timeout + nice/ionice, never a URL piped into python. No new live trade since 05:55.
- **10:11 Bangkok, hourly: no new live trade since the 05:55 sell.** mc5.db mode live 5 rows, 0 open, matches the chain audit (page fresh 10:06, timers active). Launch watch DOWN for the 2nd hour: lwatch.py hangs with no output even with 170 s run time (RPC calls stall; Alchemy gave null blocks last hour). Live audit and Half C are NOT affected. I will look at the stalled call next hour; if you need launch counts sooner, say so.
- **09:10 Bangkok, hourly: no new live trade since the 05:55 sell.** mc5.db mode live still 5 rows, 0 open, matches the chain audit (page fresh 09:06, timers active). Launch count NOT available this hour: lwatch.py got a null block back from the Alchemy RPC (twice), then timed out at 55 s. I added a public RPC fallback for that one call in my own lwatch.py (backup lwatch.py.bak); will retry next hour. The live audit page is unaffected.
- **08:05 Bangkok, hourly: still no new live trade since the 05:55 sell.** Bot db mc5.db mode live: 5 rows, 0 open (4 closed trades plus the 19:04 failed buy), matches the chain audit. Launches last 2 h: c500 PASS 7 (last 07:22), pond PASS 0, WETH pond 0; 385 launches. Half C 07:15: 7 closed, $100 mean 4.65x, $1k mean 6.14x, not below H2 p5, no flag. Timers active, audit page fresh 08:04.
- **07:05 Bangkok, hourly: no new live trade since the 05:55 sell.** Live book still 4 closed (0.485x, 0.005x, 30.18x, 0.567x), about $40 in, about $313 out, all fills 1.0000. Launches last 2 h: c500 PASS 13 (last 07:01), pond PASS 0, WETH pond 0; 356 launches. Timers fwa3-live and fwa3-halfc active, audit page fresh 07:04. Half C refresh 07:15.
- **06:04 Bangkok, 4th LIVE TRADE, both fills PASS, stop 0.567x:** pool 0xa6caf95634d4bcffdfb535eec889c9c928ff13d5, token 0x1ec25a196c49dd3445f1fcc53c37c513af92839c. BUY 05:53:48, tx 0xc062dc07c4e1c03b67ed17ef0d6fa0454cc0288926f6e0bdcf3a7c457c957e4d, block 81997741, check 05:53:46, delay 2 s, 0.003672 WETH, fill/model 1.0000, signer the wallet. SELL 05:55:23 on stop 0.5x, tx 0x973ca446657534c99eced5ca71a5ee1752b631d42bf1e71b022d76f232bc2403, block 81998653, 0.002082 WETH vs model 0.002082 (1.0000), held 1.6 min, chain x 0.567. Bot log: SELL_SENT 1.4 s after trigger (new log line, faster sell path). Live book: 4 closed (0.485x, 0.005x, 30.18x, 0.567x), about $40 in, about $313 out; all 4 buys and 4 sells fill/model 1.0000. Hourly: c500 PASS 19 last 2 h (last 06:03), pond PASS 1 (04:08), WETH pond 0; 471 launches.
- **05:18 Bangkok, Half C 05:15 refresh, live vs paper on the 30x coin:** pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468: paper $100 final 30.14x (time stop) vs live 30.18x, live/paper 1.001. Live vs paper on all 3 closed live trades: 0.485 vs 0.492 (0.986), 0.005 vs 0.009 (cascade, both near total loss), 30.18 vs 30.14 (1.001). The honest model matches live fills. Half C: 6 trades all closed, $100 mean 5.35x, $1k mean 7.09x, both above the H2 p5; no flag. v19 unchanged, no new live trade.
- **05:03 Bangkok, hourly watch:** launches last 2 h: c500 PASS 17 (last 04:55, 11 in the 04:00 hour), pond PASS 3 (last 04:07), WETH pond 1 (03:50); 532 launches. Live book unchanged: 3 closed (0.485x, 0.005x, 30.18x), none open, no new buy since 02:38.
- **04:02 Bangkok, hourly watch:** launches last 2 h: c500 PASS 12 (last 03:56), pond PASS 2 (last 03:16, back after none since 00:35), WETH pond 1 (03:49); 692 launches. Live book: 3 closed (0.485x, 0.005x, 30.18x), none open, no new buy since 02:38.
- **03:45 Bangkok, LIVE SELL, BIG WIN, fill PASS: 30.18x.** Pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, token 0x717d38587504b79e2e888c860bd1569c0d5984af, sold 03:40:27 on the 4 h time stop, tx 0xdebc14f88d03ce9b32bd5646a16df68005c45f2206babf525b37795ee8948db7, block 81919646. Got 0.110642 WETH vs model 0.110642 (fill/model 1.0000) for 0.003666 WETH in; held 240.1 min; chain x 30.18 before gas (bot mc5.db 29.43x incl. its own cost math). About $10 to about $300. Live book closed: 3 trades, 0.485x / 0.005x / 30.18x, mean 10.2x, about $30 in, about $307 out. All 3 buys and all 3 sells fill/model 1.0000. Correction: I said the time stop was about 04:20; it was 03:40 (4 h after the 23:40 buy). The 04:20 figure was wrong.
- **03:17 Bangkok, Half C 03:15 refresh:** 6 trades (1 open). $100 mean 4.60x, $1k mean 6.45x, both above the H2 p5; no flag. Live vs paper on the cascade coin pool 0x4fc807391d8dd76992043950248b26a610231075: paper $100 0.009x (sell at the end of the next block) vs live 0.005x (sold about 190 blocks later, after the price slid further). Both are a near total loss; the gap is the 20 s of RPC 429 errors in the bot, worth about $0.04 at $10. Open live coin pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468 now marks 25.6x on paper at $100 (22.9x at $1k), time stop about 04:20. No new live trade since 02:41.
- **03:02 Bangkok, hourly watch:** launches last 2 h: c500 PASS 18 (last 03:00), pond PASS 0 (none since 00:35), WETH pond 0 (none since 00:38); 798 launches. Live book: see the 03:01 line (3 buys PASS, closed 0.485x and 0.005x, 1 open pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, time stop about 04:20).
- **03:01 Bangkok, LIVE SELL, fill PASS, trade nearly a total loss (0.005x):** pool 0x4fc807391d8dd76992043950248b26a610231075, token 0x313e89b7798d0ec8b755fd46a9da9a14a4ece708, sell 02:41:49, tx 0x2cf3c8f72295d561ee72ba0e453b64ad28e673bba0cc5dc3ff0c62fe3a08a5bf, block 81885268. Got 0.000020 WETH vs model 0.000020 (fill/model 1.0000), held 3.3 min. Why: a cascade dump, not execution. From chain swaps: price was 0.60x at block 81885070, then 118 sells in blocks 81885076 and 81885077 (about 11.3 WETH out) took it to 0.009x in 2 blocks. Even a sell at the end of the next block would have got about 0.01x. Then a 100% LP burn at 02:42:33 (after our sell). Also in mc5 log: walk_err HTTP 429 at 02:41:32 (RPC rate limit), but it did not change the result here. Before the dump the pool had a repeating pattern: one wallet 0xe039e21687e99744a93a9caf4fb3a716309d78b9 selling about 1 WETH every 20 to 100 blocks into a steady line of 0.2 to 0.4 WETH buys. Live book: 3 buys PASS, 2 closed (0.485x, 0.005x), 1 open (pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468).
- **02:42 Bangkok, THIRD LIVE BUY, PASS:** 07 Oct 02:38:31, pool 0x4fc807391d8dd76992043950248b26a610231075, token 0x313e89b7798d0ec8b755fd46a9da9a14a4ece708, tx 0x89da0e7097e8a2fea2b1287f38262acb0d7cd9b122f0d1e4a416e0d8a89272aa, block 81883330. Check 02:38:30, delay 1 s, paid 0.003684 WETH, fill/model 1.0000, signer is the wallet. All 3 live buys PASS (1 s, 4 s, 1 s; fill/model 1.0000 each). Open now: 2 (this one and pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468).
- **02:02 Bangkok, hourly watch:** launches last 2 h: c500 PASS 22 (last 01:49, 12 in the 01:00 hour), pond PASS 8 (none since 00:35), WETH pond 3 (none since 00:38); 619 launches. Live book unchanged: 2 buys PASS, 1 closed 0.485x (pool 0xe507152250372fadb2c6cb50f2b5ff5e89dd8d21), 1 open (pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, paper mark 21.7x at $100, time stop about 04:20), 1 failed buy (19:04).
- **01:24 Bangkok, Half C 01:15 refresh + first live vs paper:** closed live trade pool 0xe507152250372fadb2c6cb50f2b5ff5e89dd8d21: live 0.485x vs my honest paper $100 0.492x (stop), live/paper 0.986. The open live position pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468 marks 21.7x on paper at $100 (19.6x at $1k), still open, 4 h time stop at 04:20 Bangkok; mark only, not banked. Half C (175 pools, 12 candidates, 5 trades incl. 1 open): $100 mean 4.72x, $1k mean 7.06x, both above the H2 p5; no flag.
- **01:02 Bangkok, hourly watch:** launches last 2 h: c500 PASS 15 (last 00:55), pond PASS 11 (last 00:35), WETH pond 9 (last 00:37); 624 launches. Live book (mc5.db + chain): 2 buys both PASS (1 s, 4 s, fill/model 1.0000); 1 closed, pool 0xe507152250372fadb2c6cb50f2b5ff5e89dd8d21, stop 0.5x, chain 0.485x, sell fill/model 1.0000; 1 open, pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, no sell yet; 1 failed buy (19:04). Half C next refresh 01:15.
- **00:35 Bangkok, FIRST LIVE SELL, fill PASS, trade a loss:** 07 Oct 00:19:04, pool 0xe507152250372fadb2c6cb50f2b5ff5e89dd8d21, token 0xd273ef9fba402603d9f583e848a0708ed0efa55b, tx 0xe98f2f5937a1e5d4abb970c4e2286b1a1dcaf65cb3bcbaa952cfad92ee0e2e1a, block 81801297. Got 0.001779 WETH vs model 0.001779, fill/model 1.0000. Held 6.5 min, chain x 0.485 before gas (bot mc5.db: stop_0.5x, 0.471x). About -$5 on $10. The sell executed exactly as modelled; the loss is the rule's stop, not the execution. Live book now: 1 closed (0.485x), 1 open (pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, no sell yet), 1 failed buy (19:04). Not a FAIL by the delay/fill test, so no ring.
- **00:19 Bangkok, SECOND LIVE BUY, PASS:** 07 Oct 00:12:33, pool 0xe507152250372fadb2c6cb50f2b5ff5e89dd8d21, token 0xd273ef9fba402603d9f583e848a0708ed0efa55b, tx 0xaa8f3ee5f29518ce06b400d0b1017f084a26600f20c24f0dbd797e2f16ab1605, block 81797488. Check 00:12:29, delay 4 s (pass line 6 s), paid 0.003671 WETH, fill/model 1.0000, signer is the wallet. Both live buys so far PASS (delays 1 s and 4 s, fill/model 1.0000 both). First position pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468 still open, no sell on chain yet. Timers fwa3-live and fwa3-halfc active.
- **00:00 Bangkok, hourly watch:** live position pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468 (bought 23:40:24, PASS 1 s, fill/model 1.0000) still OPEN, peak 2.51x per mc5.db; no sell on chain yet. Live attempts total 2 (1 filled, 1 failed 19:04). Launches last 2 h: c500 PASS 13 (last 23:59), WETH pond 6 (all in the 23:00 hour), pond PASS 4 (last 23:58); 823 launches. Half C (to 23:15): 3 closed, $100 mean 0.48x, $1k 5.11x, both inside the H2 range; next refresh 01:15. Hourly check now reads mc5.db (live table), not lane1.db.
- **23:46 Bangkok, FIRST LIVE FILL: PASS on both tests.** BUY 23:40:24 Bangkok, pool 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468, token 0x717d38587504b79e2e888c860bd1569c0d5984af, tx 0xa8294408cb8366814caa4b4db1fc7c6e9a9b13b9ccff17e7321c09fc68a91cbe, block 81778607, signed by the wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286. +5 min check 23:40:23 (first swap block 81775635), buy block 23:40:24: **delay 1 s** (pass line 6 s; block times from my 25k-block marks, about 1 s resolution; the bot's own t_check is also 23:40:23). Paid 0.003666 WETH (about $10), got 14,100,423,173,104,668,571,920 token units; my tick walk at the block before says 14,100,423,173,096,943,386,624: **fill / model 1.0000**. Cap at check $688k. Sell will be audited the same way when it happens. Page: http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md
- **23:26 Bangkok, v19 done. Correction to my 23:22 line: fmmc5 DID see pool 0x206211d456a3a9dbe45e13507aac5131ace9c716 and tried at 19:04:26; it failed (gmgn-cli confirmation, no --yes), so nothing was signed.** I said 'never saw'; wrong, because I read lane1.db, not mc5.db. (1) The audit page now has a 'Bot attempts' section fed from mc5.db (c table mode live plus LIVE_* log lines, read-only, pulled every 2 min from the Mac and pushed to research-box). Each attempt shows ATTEMPTED FILLED, or ATTEMPTED FAILED with the reason, so a failed try never shows as 'never saw'. (2) Half C board now has a 'live bot' column and exit reasons at both sizes. This coin at the live size ($10) was stopped on paper at 0.494x, so the failure cost about -$5, not the $1k row. Second correction: the 11.2x at $1k I posted was a mark at the end of the data (21:15), not a closed trade; the 23:15 refresh closes it at the 4 h time stop at **14.36x**. Half C $1k mean 5.11x, $100 mean 0.48x (n=3, both inside the H2 range). Page: http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md
- **23:22 Bangkok, LIVE GAP: paper mc5 says BUY at 19:04 Bangkok, live bot did nothing.** Pool 0x206211d456a3a9dbe45e13507aac5131ace9c716 (token 0xb18a4bcf9018bedb1fcb90ed80f87087151c09f4, fee 100), first swap 18:58:53, cap at +5 min $888k (inside $250k to $1M), dust guard passed. Wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286: last signed tx 12:59 Bangkok (Blockscout, nonce 8), lane1.db 0 live rows. So the live bot never saw or never acted on this coin. Please check whether fmmc5 watches fee 100 pools / this venue. Outcome on paper: at $100 and $500 the 0.5x stop fired by a hair (0.494x, 0.499x, at 19:12); at $1000 it did not (our bigger buy keeps more depth) and the coin rode the 4 h time stop to **11.2x** (exit 23:04). That one coin lifts half C $1000 mean to 4.07x (n=3); $100 stays 0.48x. The board's 'why' column shows the $100 exit only; fixing that label next. Both still inside the H2 range.
- **23:00 Bangkok, hourly watch:** fmmc5 wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 still has NO live trade on chain, and lane1.db has 0 live rows. Launches last 2 h: c500 PASS 12 (5 in the 21:00 hour, 7 in the 22:00 hour, last 22:59), so c500 is clearly back. Pond PASS 2 (last 22:55), WETH pond 0. 971 launches in 2 h, busiest fee 10000 / spacing 200 ETH (97). Half C (to 21:15): still 3 trades, mean 0.48x, inside the H2 range; next refresh 23:15. Audit page fresh (22:58).
- **22:00 Bangkok, hourly watch:** fmmc5 wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 still has NO live trade on chain, and lane1.db has 0 live rows. Launches last 2 h: c500 PASS is BACK, 4 in the 21:00 hour (last 21:57; it was 0 since 17:05). Pond PASS 5 (last 21:37), WETH pond 1 (20:52). Launch pace up: 940 launches in 2 h (633 at 20:57), busiest fee 10000 / spacing 200 ETH (91). Half C (to 21:15): 106 pools, 7 candidates, still 3 trades, mean 0.48x, inside the H2 range. Audit page fresh (21:56).
- **20:58 Bangkok, hourly watch:** fmmc5 wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 still has NO live trade on chain, and lane1.db has 0 live rows. Launches last 2 h: pond PASS 6 (last 20:29), WETH pond 1 (20:52), c500 0. Busiest class fee 10000 / spacing 200 ETH (71). Half C unchanged (to 19:15: 3 closed, mean 0.48x, inside the H2 range); next refresh 21:15. Audit page fresh (20:54).
- **19:57 Bangkok, hourly watch:** fmmc5 wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 still has NO live trade on chain, and lane1.db has 0 live rows. Launches last 2 h: pond PASS 4 (last 19:52), c500 0, WETH pond 0. Busiest class fee 10000 / spacing 200 ETH (51). Half C board (to 19:15): 3 closed, mean 0.48x at $100, inside the H2 range. Both audit timers active, page fresh (19:56).
- **18:55 Bangkok, hourly watch:** fmmc5 wallet 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 still has NO live trade on chain (audit every 2 min, http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md). Launches last 2 h: pond PASS 5 (last 18:43), WETH pond 6 (17:00 hour), c500 1 (17:05). Busy: fee 10000 (38), fee 2500 / spacing 50 (22), fee 100 / spacing 200 ETH (19). Half C board fix: a trade now shows closed once its exit happened (it showed open by clock).
- **17:52 Bangkok, v18 running: live fill audit + half C board are LIVE at http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md** (audit from chain every 2 min, half C every 2 h). No live trade by 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286 on chain yet. Audit model checked on a test wallet: buy and sell fills match my tick walk to 1.0000. Half C so far: 3 candidates, 2 trades (1 closed 0.48x, 1 open), far too few to judge. Launch watch 17:52: pond PASS back (2 in the 17:00 hour, last 17:47), WETH pond 5 (17:00 hour), c500 7 in 2 h.
- **17:27 Bangkok, v17 FILL DELAY PASS LINE: fill within 6 s of the +5 min check.** 2, 4 and 6 s are one flat plateau; at 10 s every column loses more than 5% against the best of 2 to 6 s (H1 -12% to -13%, H2 -9% to -15%, $1k and $2k). 30 s keeps 85% to 95%. Section 15.
- **17:23 Bangkok, v16 HONEST MONEY LINE for mc5 done: the edge survives every friction.** Fill 6 s after the check, every exit sold at the end of the NEXT block, real tick depth, own buy kept, 1% + gas each way, 4 h. Through 5 Oct, H2: $100 1.84x +$1,118 a day; $1k 1.69x +$9,195; $2k 1.58x +$15,524. Fill at 30 s instead of 6 s costs 13% to 24% of that ($1k +$7,965). Max drawdown about 14 tickets (H2), H1 under 4. Without the top 3 coins still positive at every size. 19 of 116 stops filled under 0.3x (named in section 14). Section 14.
- **17:17 Bangkok, v15 done. (1) Re-entry after our 0.5x stop = NO EDGE, killed.** Every rule lives on one coin per half. H1 pick (buy flow back, 2 h) fails H2: 0.90x. Best rule in both halves (cap back to $500k within 2 h: H1 6.72x n7, H2 1.74x n38, +$333 / +$234 a day) drops to 1.38x / 1.05x without its best coin, and its H2 money is 2nd window coins (3.84x on 11 w2 coins, 0.88x on the other 27) = double exposure, not new money. **(2) Supply look-ahead fixed: changes nothing** (A1's own code with supply at the +5 min block: 200 trades vs 201, 1.89x / 1.58x, +$12,887 vs +$12,826). Section 13.
- **17:08 Bangkok, v13/v14 mc5 rebuild DONE: A1's edge is REAL on my own code and my own chain data.** Same coins 190 of A1's 201 (11 A1 only, 1 mine only). 2 h stop: H1 1.96x (A1 1.86x), H2 1.63x (A1 1.58x), +$13,016 (A1 +$12,826). 4 h stop, 13 Sep to 5 Oct: H2 1.66x (A1.18 1.67x), +$937 a day in H2 (A1 $946). Same 5 biggest winners. No look-ahead found that matters (one small one: A1 reads token supply at today's block, not at +5 min). Warning: it is a lottery (median 0.48x, 1 coin in 4 wins). Full detail section 12.
- **16:43 Bangkok, hourly launch watch:** c500 is BACK: 5 c500 PASS coins in the 16:00 hour (last 16:31), after nothing since 10:14. Pond PASS 0, WETH pond 0 in the last 2 h. Busy: fee 10000 (67), fee 2500 / spacing 50 (28), Doppler spacing 8 (17). Live test: still 0 live trades in /opt/fm/lane1.db (131 watch rows), no paper vs real yet. v13/v14 mc5 rebuild: own chain data is downloading (441 candidate pools, Mint/Burn/Swap to first swap + 4 h 15 min), results with 2 h AND 4 h time stop in the next update.
- **14:58 Bangkok, hourly launch watch:** pond PASS 0 and c500 PASS 0 in the last 2 h (last c500 10:14). WETH pond 2 (both 14:00 to 14:59 hour, last 14:38). Busy: fee 10000 (57), fee 2500 / spacing 50 (32), Doppler WETH spacing 8 (17), hook 0x6404ddb4f33b4d974e570f66288e2a7ca469a840 (12). Live test: still 0 live trades in /opt/fm/lane1.db (129 watch trades closed), so no paper vs real comparison yet.
- **14:41 Bangkok, sup-to-a3 v12: CORRECTION. The +144% and the "-91% other launches" split are WRONG; the right number is +61% a trade at $20, median -100%.**
  - (a) My split used the "factory" column of the day-file launches table, not the launch tx. Analyst 2 is right: all WETH pond pools launch through 0x58daec3116aae6d93017baaea7749052e8a04fa7. The 825 "other" pools are pools whose launches row was never filled with metadata (factory, creator, name and time all empty; same tx and block as the pool). That field is filled later by the rebuild and clearly tracks the outcome, so it leaked the result. Dropped.
  - (b) My books do set pulled coins to 0 (pull before our sell fills = 0). So my +61% all-pool number and Analyst 2's +55% (+51% without fees) are the same result. +206% (pulled coins at last price) is not ours.
  - Fixed grid, all 2,330 pools, $20, fees and depth in: sell at 5 min +41%, 10 min +50%, 15 min +61%, 20 min +57%, 30 min +42%; "3x first" +18% to +25%; "5x first" +28% to +35%. 15 min stays best: 16 of 16 days up, wins 22% of coins, median -100%. Walk forward still picks 15 min: test 28 Sep to 5 Oct +74%, 8 of 8 days.
  - Gap check (v11) restated without the bad split: all 58 wallets real +27% a wallet, our 15 min exit on their own fills +128%, sim for the same coins +102%, fills 0.7 s after the first trade at 0.97x the sim price. Conclusion unchanged: the sim fills are right; real wallets lose the edge by selling at the wrong time (never, or too early).
  - Sections 10 and 11 still show the old PosM-only numbers; read them through this correction.
- **14:40 Bangkok, sup-to-a3 v11 done: the sim vs real gap is explained, the sim fills are right.** Detail in section 11.
  - The 58 wallets ($15 to $35 first buy, first 5 s): per-wallet average real +27% (dollar-weighted -15%). 22 of them bought coins NOT launched through 0x58daec3116aae6d93017baaea7749052e8a04fa7: real -42%, sim -81%. The sim's +144% is PosM only, so part of the gap is just a different coin set.
  - The 36 PosM wallets: real +69%; OUR exit (sell all 15 min after THEIR fill, through the pool) on their own fills +255%; sim for the same coins +215%. Their fills match the sim: median 0.8 s after the first trade, paying 0.95x the sim entry price. So not (a) worse entry and not (d) wrong sim fills.
  - The gap is (b) exit timing, both ways: 10 of 36 never sold and lost 100% to the pull; most of the rest sold early for 0% to +33% on coins that were up 4x to 33x at 15 min. (c) pulls are already inside both numbers (72% of these coins were pulled within 15 min of the wallet's buy).
  - Shape of the edge: our 15 min exit wins on only 10 of 36 coins (median -100%); the average comes from the 1 in 3 that runs 4x to 33x. Best coin removed: +168%. It is a lottery that pays only with many coins a day; at today's 3 launches a day it is mostly luck.
  - Clock: earlier lines said 14:50 and 15:15 Bangkok; real times were 14:22 and 14:31 (fixed in v5). Now taken from the server clock.
- **14:40 Bangkok, sup-to-a3 v10 done (WETH pond).** Detail in section 10.
  - (1) WHO: the same pond crews. Top roots: 0x0ec66d31aab7e387b2f4cda075e39f7a6a0f541b (391 coins), B/C 0x4757fc41278bd0f1b028c6dae71d381dc4558d98 (273), A 0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92 (235), 0x20e7a901982b0acbb8cccdc9148373673e2bf6b1 (192), 0xca24b6f1b00a7d336ef2984239daa836f17cb6c1 (134), 0x0427e497f8075524fdc6a73ed6dfe1be958a2a56 (123); 334 unmapped. They run ETH pond and WETH pond side by side. Where they went: on 6 Oct their mapped deployers launched only ETH pond (24 coins), zero WETH pond and nothing in any other class. They slowed down; they did not move (as far as the crew map can see; new unmapped wallets are possible).
  - (2) Pull test, outside buys only (Analyst 2's pt2.py, REPN 5), all WETH pond 20 Sep to 5 Oct: ratio within 5 s / 30 s / 60 s: $5 1.0 / 1.9 / 1.3; $20 0.7 / 1.0 / 0.9; $50 0.8 / 0.7 / 0.6; $100 0.5 / 0.7 / 0.7. Passes at $20 to $100. Watch crew 0x20e7a901982b0acbb8cccdc9148373673e2bf6b1: $5 buys pulled 42% (3.8x).
  - (3) Time-stop grid (PosM-launched, entry first trade + 4 s, fees and depth in): 5 min +115%, 10 min +128%, 15 min +144%, 20 min +137%, 30 min +115% a trade at $20; "3x first" +79% to +91%, "5x first" +94% to +106%. Every row 15 of 15 days up. Walk forward picks 15 min on 20 to 27 Sep: test 28 Sep to 4 Oct +162% a trade, 7 of 7 days, n 486. Size barely matters ($100 +141%).
  - WARNING: real outside wallets in this class did much worse than this simulation: first 5 s at $3 made +32%, at $20 made -15% (58 wallets). The simulation sells everything at one fixed moment through the pool; the real gap is not explained yet. Treat +144% as an upper bound until a live paper test runs on new launches.
  - (4) Launch watch now has a WETH pond line: 3 launches in the last 3 h (11:00, 12:00, 14:23 Bangkok). Pond PASS 0, c500 PASS 0.
- **14:31 Bangkok, sup-to-a3 v9 done: Doppler stays DEAD at every entry time.** Detail in section 9.
  - (1) The honest "+25%" used the token amounts in the Swap event and marked unsold coins at the last price. What the wallets really got back from their own sells was -3%. So no, it was not real.
  - (2) Entry after the fee window, best exit each time (sell at 1.5x or at 15 min): +15 s -17% at $20 / -28% at $100; +30 s -15% / -27%; +1 min -15% / -27%; +5 min -13% / -25%. 0 of 16 days up in every row. Walk forward test -14% to -35%. Farm filter: still -11% at $20 / -24% at $100.
  - (3) By extra fee hook: 0x9982538f41f2ae29ddb9d3d9307010052984fdbb has 1,305 of 1,344 traded pools, -27% at $100. 0x6f02324d20cc679d0e585290caa6b16bacbc0f77 (long.xyz) -20% (n 10). 0xe2aebc987592593b667ec29178d0a83929db78b6 -1.5% (n 12, 2 of 7 days up). No hook: 11 pools, too few.
  - Assumed fees: pool LP 0.2% + extra hook 1.5% each way after the window (Analyst 2's range 0.4% to 1.75%), GMGN 1% each way, gas. Buying inside the 80% window was never modelled; it can only be worse.
- **14:22 Bangkok, sup-to-a3 v8 done. Doppler KILLED, WETH pond is the new lead (but drying up).** Detail in section 8.
  - Doppler (hook 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544): the +25% was paper. Real sells of those 702 wallets: -3%; the rest was unsold coins marked at the last price. Our fixed exits from a first-5 s entry lose on every day: -11% at $3, -13% at $20, -24% at $100 (0 of 16 days up). Walk forward -28%. Farm filter (-22%) and launch source (Airlock -21%) do not save it.
  - Fee 100 WETH pools, no hook ("WETH pond"): the +32% is real sells. Best fixed rule: buy at first trade + 4 s, sell at 15 min, no target: +61% a trade at $20, 16 of 16 days up. Walk forward (exit picked on 20 to 27 Sep, tested 28 Sep to 5 Oct): +74% a trade, 8 of 8 days up.
  - Only coins launched through the PositionManager 0x58daec3116aae6d93017baaea7749052e8a04fa7 work: +144% a trade at $20, 15 of 15 days. Other launches: -91%.
  - Catches: 98% of these are crew coins (farm check flags them; the 48 non-farm coins +53%). Holding 1 h instead of 15 min loses (-6%): crews pull between 15 and 60 min. Launches are drying up: about 140 a day late Sep, 16 on 4 Oct, 3 so far on 6 Oct.
  - For Analyst 2: Doppler swaps are in the swaps table of /opt/hist/rh/DAY.db on research-box (pid = pools.rowid). 87% of Doppler pools never trade at all (checked on chain: 0 swaps in their first hour), so "no swaps" is real, not missing data.
- **14:20 Bangkok, tasks 1 + 2 done (honest map, outside wallets only, crew wallets removed with the pulltrig/A1.9 rule):** detail in section 7.
  - Pond native (fee 100, ETH): outside buyers of $20 lost 96% (5 to 15 s entry, 34 wallets, 97% pulled). Paper pond lanes do not survive for outsiders.
  - c10k (fee 10000): outside $20 buyers lost 65% (1,739 wallets, 99% pulled). Dead, as found.
  - c500: outside buyers won +21% at $20, +35% at $50, +33% at $100 (5 to 15 s), but almost all of it on 21 to 26 Sep. Trap ratio (pull within 5 s): $3 9.2x, $10 5.9x, $20 2.4x, $50 1.1x, $100 0.3x. Small buys get pulled, bigger buys do not.
  - Best outside cell that never gets pulled: hooked launchpad 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544 (WETH, spacing 200), $100 in the first 5 s: +25%, 11 of 16 days positive, 702 wallets, 0 pulls ever. Small $3 tickets there are a lottery (+2,169%, but carried by a few days).
  - fee 100 WETH pools (no hook): $3 in the first 5 s +32%, 14 of 16 days positive; trap ratio at $20 = 1.0 (no trigger). $20 tickets there: mixed (-15% to +34%).
- **14:13 Bangkok, launch watch:** still zero pond and zero c500 launches (last c500 10:14). Busy now: fee 10000 (44 in 1.5 h), fee 2500/spacing 50 (25), new hook 0x6404ddb4f33b4d974e570f66288e2a7ca469a840 (19).
- **14:10 Bangkok:** on the bus. Read sup-to-a3 v7. Working now: task 1 honest map (outside wallets only, split from crew wallets, A1.9 method + /opt/sw/pulltrig.py), then task 2 (trap ratio at $20 column), task 3 launch watch (hourly). Run order accepted: v3 runner lane first, pond A and B/C cannot run while their crews are off.
- **Paper vs real (curve):** a sell of our size moves the price very little. Lane1 watch book ($5): real is 0.06% under paper. Lab books ($20): 0.1% to 2% under. Only fast10_real ($200) loses 13% and c500 at $20 loses 13% of its profit (thin pools). The lab "worst print" paper is already below what a real sell gets (broad20: paper +$561, real-depth +$636 on 10 trades).
- **Paper vs real (live):** zero live buys yet. No pond launch since 09:40 Bangkok, so the live test has nothing to buy. Rerun when it has 5 closed trades.
- **The real risk is pulls, not slippage.** c10k: any $1 to $30 outside buy is followed by a pull within 3 s 95% to 100% of the time (fail, matches Analyst 2).
- **c500 did not move, it stopped.** It is ONE operator: the same 4 token code variants (4655 bytes, always about 40/20/20/20 mix), a new crew root every 1 to 2 days. On 6 Oct its wallets launched 41 c500 coins and 1 coin elsewhere (fee 2500 / spacing 25, code 7154 bytes). Last c500 pass 10:14 Bangkok. Nothing as of 13:55.
- **New hook class 0x6404ddb4f33b4d974e570f66288e2a7ca469a840 is a different operator** (token code 2390 bytes, quote token 0x5fc5360d0400a0fd4f2af552add042d716f1d168, launched through 0xc35d4eb5eedbdb7531306f545f39dab3ef51c560, deployers not in the crew map). Started 12:48 Bangkok, 14 launches in 4 h.
- **L-BROAD20 luck test:** cannot be rerun as asked. The 1,499 idea ledger and the 8 model ensemble live in another session's cloud lab (/home/claude/lab/iter), not on any box. Proxy on 384 to 576 lane ideas: picking the best idea on past days only held up, 11 of 11 next days positive, winner shrank 0% to 19%. Ask PP to publish the per-day result of every lab idea so the real test can run.
- **Run order:** 1. pond A at $5. 2. pond B/C at $5. 3. v3 runner lane. 4. c500 only after Analyst 2's pull test passes AND launches come back. c10k off.

## 1. Whole book (HOT regime, real fills through pool depth, fees and gas in)
| lane | pull test passed at | coins a day | profit a coin | $ a day | worst day | days |
|---|---|---|---|---|---|---|
| pond A (0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92) | $3 to $5 (Analyst 1) | 14 | +$8.96 at $5 (+179%) | +$129 | +$49 | 5 |
| pond B/C (0x4757fc41278bd0f1b028c6dae71d381dc4558d98) | $3 to $5 (Analyst 1) | 19 | +$3.76 at $5 (+75%) | +$72 | +$19 | 5 |
| pond other crews | open (Analyst 1) | 17 | +$1.56 at $5 (+31%) | +$26 | -$61 | 5 |
| c500 | untested (Analyst 2) | 170 until 5 Oct, ~0 now | +$7.83 at $5 / +$28.41 at $20 | +$1,334 at $5 | +$389 | 14 |
| c10k 3 ETH | FAILED | 0 | was +$4.79 at $5 | 0 | | 10 |
| c10k 5 ETH | FAILED | 0 | was +$4.05 at $5 | 0 | | 10 |
| v3 runner lane | passed (second window) | see second window docs | not rebuilt here | | | |
- Live paper last 6 h (supervisor): c10k 3 ETH +162% (26), c10k 5 ETH +115% (14), pond badge +87% (12), c500 0.
- Same-day losses: no day with 2+ lanes losing (14 days). Most positions open at once, all lanes: 6 ($30 at $5, $120 at $20). Median hold 1.5 to 4 min.
- Data: pond pass only exists 30 Sep to 4 Oct (5 days). Day files stop at 4 Oct (see section 6).

## 2. Paper vs real fills, detail
- Method: pull every Swap of the pool from chain, take the pool's live liquidity from each Swap event, sell our exact token amount through it at the exit time (constant product inside the active range). Not eth_call (no quoter on this chain); same math the pool runs.
- Lane1 watch book: 55 trades with a sell. Bot paper 0.3134 ETH, tape mark 0.2988, real 0.2986. Bot is 5% above the tape, mostly on 10x_rest trades (leg timing).
- Lab picks, real-depth vs paper: broad20 +$636 vs +$561 (10), broad20_crew +$102 vs +$76 (4), pond20 +$158 vs +$146 (20), fast10_real ($200) +$516 vs +$513 (2, depth cut 13% off the mark).
- Historic lanes at $20, depth gap on profit: pond 0%, c10k 1%, c500 13%. Entry slippage median 0.1% to 0.5%.

## 3. Pull trigger, coarse cut (all buys in the first 60 s, before any pull)
| lane | $1 to $4 buys pulled within 3 s | $4 to $30 | $30 to $80 | base (< $1) |
|---|---|---|---|---|
| c10k 3 ETH | 98% to 100% | 99% to 100% | 69% | 21% |
| c10k 5 ETH | 95% to 100% | 98% to 100% | 30% | 9% |
| c500 | 34% to 38% | 1% to 3% | 1% | 2% |
| pond A | 0% | 0% to 1% | 3% | 0% |
| pond B/C | 1% to 3% | 0% to 13% | 1% | 0% |
- Caveat: this data cannot tell crew buys from outside buys. c500's $4+ bins are probably the crew's own wash buys, so c500 stays untested until Analyst 2 splits by wallet. The $1 to $4 c500 result is a warning.

## 4. c500 decline
- Pass coins a day: 265, 200, 192, 219, 238, 233, 204 (28 Sep to 4 Oct), 170 (5 Oct), 41 (6 Oct, last 10:14).
- Crew roots by day (one operator, roots rotate): 1 Oct 0xc4a2ffded2e39b6beb595e78950870dc5ec352d5 and 0x6d27a4b18bdc5f2de2e10ae185147b6928fa90f6; 2 to 3 Oct 0xa9e19a0183246ce4adfac43f627de34ee894a2d1; 4 Oct 0x83fbc2e09cad486fb792994f4887ddd73b214758 and 0x0a30f79805103739a9060a2e037c25ba6702643c; 5 Oct 0xebe97ee22d53e806644be18294bd7836bd5e74ad; 6 Oct 0x11f21754f96359e55b662b299ab0540f492e4ff3.
- Where they went: nowhere visible yet. Their wallets made no launches in any other busy class. A side habit: about 6 to 15 a day of the same coins on a per-pool vanity hook (address ending 0840), same fee 500.
- The 4655 byte token code also shows up in odd-fee pools quoted in other tokens (mostly 4 Oct): other users of the same template, not this operator.
- Whole chain is slow at midday too: fee 10000 class 25 an hour vs 47 an hour last night.

## 5. L-BROAD20 luck test (proxy)
- Grid: lane (pond, c500, c10k, and their non-pass coins) x 8 exits x entry 4 s / 10 s x 6 regimes. Each day D: pick the best idea on days before D only, trade it on D.
- 21 Sep to 4 Oct: chained out of sample 11 of 11 days positive, +$7,137 a day; winner's past average +$7,094 a day (no shrink). Without c10k: +$6,061 a day, 14% shrink. Ranked by $ a trade: 19% shrink.
- Most picked: c500, sell all at 5x, entry 10 s, HOT.
- Live broad20: 10 closed, +$561 paper (+$56 a trade), best trade removed still positive. Too few trades to call.
- Needed for the real test: PP publishes per day, per idea results of the lab ledger.

## 6. Problems found
- dRPC paid key: "User balance exceeded" (HTTP 403) since about 12:45 Bangkok. Research and backfill calls fail. The live bot log also shows HTTP 429 errors.
- Day files stop at 4 Oct: hist-rh is paused, "under 12 GB free on disk" on 159.223.88.168 (91% full).
- Launch watch: hourly check running from this session; next report in this file.

## 7. Honest map: what outside buyers really made (task 1 + 2)
- Data: day files 20 Sep to 5 Oct, every pool quoted in ETH or WETH. One row per outside wallet per pool. Outside = not a crew wallet, not first-funded by a crew distributor, not buying 3+ pools of the same crew that day (same rule as /opt/sw/pulltrig.py). Result = ETH back from its own sells in the first 30 min + tokens still held at the last price (0 if pulled). Cell = entry second and size of the wallet's first buy. Kept cells: 30+ wallets, 10+ pools, 3+ days.
- Caveat: held tokens are marked at the last price, not sold through depth; fair for small tickets, optimistic for $500+ (the $500+ cells are probably undetected insiders too, so they are left out below).

| class | entry | size | wallets | return | best pool removed | days up | trap ratio at $20 (pull in 5 s / 60 s) |
|---|---|---|---|---|---|---|---|
| hook 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544, WETH, s200 | 0-5 s | $100 | 702 | +25% | +19% | 11/16 | never pulled |
| same | 0-5 s | $20 | 200 | +2% | | 8/16 | never pulled |
| same | 0-5 s | $3 | 328 | +2,169% | +1,502% | 9/15 | never pulled |
| fee 100, WETH, no hook | 0-5 s | $3 | 794 | +32% | +30% | 14/16 | 0.9 / 1.0 |
| fee 100, WETH, no hook | 5-15 s | $20 | 62 | +34% | | 6/11 | 0.9 / 1.0 |
| c500 (fee 500, ETH) | 5-15 s | $20 | 728 | +21% | +19% | 5/11 | 2.4 / 0.8 |
| c500 | 5-15 s | $50 | 298 | +35% | +30% | 9/12 | 1.1 / 0.5 |
| c500 | 5-15 s | $100 | 1,354 | +33% | +32% | 9/12 | 0.3 / 0.3 |
| c500 | 15-30 s | $50 | 258 | +45% | +42% | 7/11 | 1.1 / 0.5 |
| fee 2500, s25, ETH | 5-15 s | $20 | 81 | +22% | +12% | 4/14 | almost never pulled |
| c10k (fee 10000, ETH) | 5-15 s | $20 | 1,739 | -65% | | 1/16 | trap (Analyst 2) |
| pond (fee 100, ETH) | 5-15 s | $20 | 34 | -96% | | 0/5 | |
- Full ranked list (470 cells): /root/fwa3/honest_cells.json on research-box. Rows: /root/fwa3/honest_rows.jsonl. Scripts honestmap.py, honestagg.py, honestdetail.py, trap tables /root/fwa3/trap/*.txt.
- What it means: outsiders lose in every class where crews pull (pond, c10k). Money for outsiders is in classes that cannot be pulled (launchpad hook 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544) and in c500 at $50 to $100 (bigger buys do not trigger the pull; small ones do).
- Next: realized-only split (sold vs still held) for the launchpad cells, and a walk-forward check of the top cells by day.

## 8. Fixed-exit rules (sup-to-a3 v8)
- Script /root/fwa3/doppler.py, rows dop_rows.jsonl, report dopana.py (research-box). Entry: pool state at first trade + 4 s. Exit: first print at or above k x entry, filled at the pool state 3 s later, else the time stop. Pull before the fill = 0. Through pool depth. Fees: Doppler pool fee 0.2% (read from Swap events of a sampled pool), fee 100 pools 0.01%, GMGN 1% each way, gas $0.03 a side.
- How the honest +25% was made (Doppler, $100, first 5 s, 702 wallets): in $120,843, sold $117,295 (-3%), sold plus still held $151,588 (+25%). Fee 100 WETH $3: in $2,558, sold $3,367 (+32%), all real sells.

| class | size | exit | $ a trade | days up | walk forward test (28 Sep to 5 Oct) |
|---|---|---|---|---|---|
| Doppler | $3 | 1.5x or 15 min | -11% | 0/16 | -13%, 0/8 |
| Doppler | $20 | 1.5x or 15 min | -12% | 0/16 | -16%, 0/8 |
| Doppler | $100 | 1.5x or 15 min | -24% | 0/16 | -28%, 0/8 |
| Doppler | any | 2x, 3x, 5x, none; 15 min or 1 h | -11% to -29% | 0/16 | |
| fee 100 WETH | $20 | 15 min, no target | +61% | 16/16 | +74%, 8/8 |
| fee 100 WETH | $20 | 5x or 15 min | +32% | 14/16 | |
| fee 100 WETH | $20 | 3x or 15 min | +20% | 13/16 | |
| fee 100 WETH | $20 | 1 h, no target | -6% | 6/16 | |
| fee 100 WETH | $3 | 15 min, no target | +60% | 16/16 | +73%, 8/8 |

- Fee 100 WETH, 15 min rule, $20, launched through 0x58daec3116aae6d93017baaea7749052e8a04fa7, coins / $ by day: 20 Sep 130 / +3,593; 21 Sep 146 / +3,625; 22 Sep 140 / +3,991; 23 Sep 121 / +4,390; 24 Sep 91 / +3,418; 25 Sep 139 / +2,537; 26 Sep 137 / +3,869; 27 Sep 115 / +2,233; 28 Sep 143 / +1,503; 29 Sep 137 / +5,139; 30 Sep 62 / +3,756; 1 Oct 65 / +935; 2 Oct 44 / +1,740; 3 Oct 19 / +663; 4 Oct 16 / +2,030.
- Launches of fee 100 WETH (all sources): 28 Sep 204, 29 Sep 286, 30 Sep 118, 1 Oct 109, 2 Oct 85, 3 Oct 51, 4 Oct 43, 5 Oct 41, 6 Oct 3 (to 13:00).
- Trap ratio for fee 100 WETH at $20 (pulltrig, outside buys): 0.9 within 5 s, 1.0 within 60 s. Our buy does not trigger the pull.
- Open: which crew roots launch these (likely the same pond crews that went quiet today), and a live paper test of the 15 min rule the moment they launch again.

## 9. Doppler after the fee window (sup-to-a3 v9)
- Script /root/fwa3/doppler2.py (entries +15 s, +30 s, +1 min, +5 min; exits 1.5x, 2x, 3x, 5x or none, time stop 15 min or 1 h; $20 and $100), extra hook per pool read with getState(asset) word 2 on hook 0x4e3468951d49f2eea976ed0d6e75ffcb44a9a544 (/root/fwa3/xhook.py, dop_xhook.json), report dop2ana.py.
- 1,344 Doppler WETH pools with 2+ trades, 20 Sep to 5 Oct. Every entry time and every exit loses on every day. Best of all rows: +5 min entry, 1.5x or 15 min, $20: -12.5% a trade.
- Why: after launch the price mostly slides, and the extra hook fee (about 1.5% each way) plus GMGN 1% each way already cost about 5% a round trip; at $100 the thin WETH pools add about 10% more.
- Verdict: no Doppler first-minute rule. Doppler stays useful only as a pull-free venue for a later-stage lane (e.g. the v3 runner gate), not for launch buys.

## 10. WETH pond (sup-to-a3 v10)
- Scripts on research-box /root/fwa3: wroots.py (crew roots by day), f100.py + f100ana.py (time-stop grid, $5/$20/$50/$100), trap/f100weth_rep.txt (pt2.py REPN 5, by crew), lwatch.py (launch watch with WETH pond line).
- Launches via PositionManager 0x58daec3116aae6d93017baaea7749052e8a04fa7 by day: 20 Sep 165, 21 Sep 174, 22 Sep 167, 23 Sep 138, 24 Sep 109, 25 Sep 158, 26 Sep 150, 27 Sep 130, 28 Sep 149, 29 Sep 177, 30 Sep 74, 1 Oct 67, 2 Oct 49, 3 Oct 20, 4 Oct 18.
- Same crews' launches from 4 Oct (deployers mapped through crew_lookup): 4 Oct ETH pond 77, WETH pond 38; 5 Oct ETH pond 73, WETH pond 35; 6 Oct ETH pond 24, WETH pond 0.
- Open question for the team: why do real outside wallets earn far less than the fixed 15 min sell? Next step: replay 20 real outside wallets' fills against the simulation, coin by coin.

## 11. Sim vs real wallets, WETH pond (sup-to-a3 v11)
- Script /root/fwa3/gap.py, per-wallet rows gap.json (research-box). For each outside wallet: its real fill (block, tokens received, WETH paid), its real result (own sells + held at last price, 0 if pulled), our exit on its fill, the sim result for that coin, entry delay, price paid vs sim entry, pull time after its buy.

| group | wallets | real | our 15 min exit on their fill | sim, same coins | entry delay (median) | price paid vs sim entry | pulled within 15 min of their buy |
|---|---|---|---|---|---|---|---|
| all | 58 | +27% | +128% | +102% | 0.7 s | 0.97x | 79% |
| PosM-launched | 36 | +69% | +255% | +215% | 0.8 s | 0.95x | 72% |
| other launch source | 22 | -42% | -82% | -81% | 0.7 s | 1.00x | 90% |
| PosM, sold less than half | 10 | -100% | -49% | -52% | 0.9 s | 0.95x | 90% |
| PosM, sold half or more | 26 | +134% | +373% | +317% | 0.5 s | 0.96x | 65% |

- PosM wallets: real wins 19 of 36 (median +1%); our exit wins 10 of 36 (median -100%), mean +255%, best coin removed +168%.

## 12. mc5 independent rebuild (v13 + v14, 17:08 Bangkok)
My own code, my own raw chain data (public RPC + Alchemy), no A1 file used except for the final comparison.
- Data: every v3 PoolCreated with WETH since 12 Sep 00:00 Bangkok (7,511 WETH pools). One pool per token (earliest first swap). T0 = block time of first Swap (chain block times, not a block count guess). Cap = price of the last Swap before T0+300 s x totalSupply read AT that block (eth_call at block) x ETH/USD that hour. Then for the 441 coins with cap $150k to $1.5M: full Mint, Burn and Swap logs from pool creation to first swap + 4 h 15 min.
- Fill: $100 at the pool state at T0+306 s (6 s fill delay), liquidity taken from the Swap event itself (exact, not estimated), pool fee + 1% each way + $0.05 gas.
- Exits: stop 0.5x checked on swaps of 0.01+ WETH, sold at state +3 s, value capped at 90% of pool WETH; pull = first Burn whose liquidity units are 50%+ of the minted units of positions covering the current price (never WETH amounts) = 0; time stop 2 h or 4 h.
- Dust filter: A1's 201 trades have a hidden filter (his code needs a 0.001+ WETH swap that moved the price before +5 min). I used the same filter for the match ("moved"). The new A1.20 guard (also 0.1+ WETH in pool) shown separately.

| run | H1 n / mean | H2 n / mean | all $ | A1 |
|---|---|---|---|---|
| 2 h, 13 Sep to 6 Oct, "moved" filter | 31 / 1.96x | 160 / 1.63x | +$13,016 (191 trades) | 42 / 1.86x, 159 / 1.58x, +$12,826 (201) |
| 2 h, to 5 Oct only | 31 / 1.96x | 156 / 1.38x | +$8,960 | |
| 4 h, to 5 Oct, "moved" filter | 31 / 3.00x | 156 / 1.66x | +$16,502; H2 +$937 a day | A1.18: H2 1.67x, $946 a day |
| 4 h, to 5 Oct, full A1.20 dust guard | 22 / 4.16x | 146 / 1.77x | +$18,163 | |
| 2 h, no dust filter at all | 59 / 1.06x | 199 / 1.32x | +$6,703 (258) | 67 dust coins, almost all end near 0x |
- Drop the best 3 coins: 2 h all 1.29x, 4 h all 1.52x. Still above 1 but thin. Median 0.47x to 0.49x, wins 19% to 27%.
- 6 Oct alone matters: the 35x coin on 6 Oct lifts 2 h H2 from 1.38x to 1.63x.

Same coins check (2 h, "moved" filter): 190 coins in both. On those 190: A1 mean 1.70x, mine 1.69x. Cap ratio mine / A1 median 1.000. 33 coins differ by more than 0.25x, both ways (biggest: 0xa2508b15ab16826c720631b9afa0a90c453ae8a4 A1 6.31x, mine 0.09x stop; 0xbed6314fe7b46707aff9863c5e5838aa93397880 A1 0.50x stop, mine 4.96x). Cause: the stop fires on a different swap because A1 estimates pool depth and I read it exactly. They cancel out.

A1 only (11):
- Pulled BEFORE our +5 min buy, A1 counts them as 0x, I skip them (cannot buy a pulled pool): 0x552163be4d46cb1bc77cd12d1f90ec6d0d28901f, 0x6d823ce811023473e5d3cc9e2b1ce6a8de7c1a94, 0xc6fb2e4eb44dcd17e02ade42b403af6fa343b0b3, 0xd95cfefd94672ab811328fd4a58fe21e7bce1e18, 0xff5d0d74c79292935768b5186569cdac3fa2557c. A1 is stricter here (costs him about $500).
- Cap differs at the band edge, most likely because A1 reads totalSupply at TODAY's block (look-ahead): 0x21c5b1f2a50cf8657ad5409fdfce39ae1438782e (A1 $538k, mine $1,126k at that block), 0x8d5c42c096344e3ad4d5fcac8fe7a4b4ffca0150 (A1 $274k, mine $227k, A1 4.70x), 0x6a7799ab447670d28be81317e003884142b49fee (A1 $251k, mine $235k).
- My 'moved' filter edge or skip rule, or a different first pool: 0x07312c2b29989c47572f59f03f24e898b0dc822b, 0x8c3ffde1ff35f506adec33b6efe58a17b2c6dce3, 0xca75082b85bb7bec8325d513f615b16bda260020 (A1 pool 0x3de50629d1f5305af91c672ba6f259ba589fd6a8, my first trading pool 0x98fd3d100bb4d7fa5cd82c2a1cb8d0c33c219a53).
Mine only (1): 0x81b2ae5f1c126dacbbb50572fd0ddf2272780910 (pool 0xaacb501075d9c9792c82f80d338533f08e5419d7, mine $293k, A1 $245k, 0.40x).

5 biggest winners, 2 h (same 5 as A1, same order):
| x | token | pool | T0 Bangkok |
|---|---|---|---|
| 35.41x | 0xc38c332012a9116dcade0c7f43b5d3c71799a9f8 | 0xb4b0dfeaa2e2a838f0f7a304f005ebe32390d4ef | 6 Oct 06:24 |
| 22.50x | 0x257723230c9cccf6a2df1e7cf67cffcfcf01eeb8 | 0xeb1646dbfbeacf76483df1092c208388c029c534 | 30 Sep 08:26 |
| 21.61x | 0xba1a2d9783ebe2b76493c5c12eb9813a0e062843 | 0x6e8a6a1af8ec53f3f09c7303e5403558dee42d0a | 2 Oct 23:43 |
| 16.82x | 0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 | 0x7608c38ea8cc2082b043e886185979942cdeb58e | 4 Oct 22:21 |
| 12.02x | 0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd | 0xb6fe1035f6edfaeb7a9f7d4c19e11d35b3bd7c57 | 23 Sep 07:45 |

5 biggest winners, 4 h (to 5 Oct):
| x | token | pool | T0 Bangkok |
|---|---|---|---|
| 33.62x | 0xba1a2d9783ebe2b76493c5c12eb9813a0e062843 | 0x6e8a6a1af8ec53f3f09c7303e5403558dee42d0a | 2 Oct 23:43 |
| 20.05x | 0x55d598f8b2c3528e47062b760d6fc66cfd2a2578 | 0xc8b416104e0b4bd5e5e14489f90ac807b67300e2 | 22 Sep 08:49 |
| 19.28x | 0x257723230c9cccf6a2df1e7cf67cffcfcf01eeb8 | 0xeb1646dbfbeacf76483df1092c208388c029c534 | 30 Sep 08:26 |
| 18.75x | 0x88042d98421b4d4014910bd9ca40d3cf3a309733 | 0xc586fcefc588870aa3159b713aa813578326562c | 22 Sep 05:04 |
| 18.28x | 0xfa1824e750f9a5bc7d253342935471b8f2709d70 | 0xd958ef2cc262fdbe3b670aec434d8b9102f5ad34 | 23 Sep 17:48 |

Look-ahead checks:
- Cap at +5 min: price from the last Swap before T0+300 s in both. Supply: mine at that block; A1 at today's block (small look-ahead: 3 coins cross the band edge, net +$260 for A1).
- T0 from the chain first Swap in both. My block times from chain anchors; A1's from a block count x 0.101283 s (close enough, same 190 coins).
- Pull uses Burn liquidity units vs minted units, in-range only. Confirmed in both.
- Fill: mine at +306 s state, A1 at the +300 s state (no delay). Small.
- No data from after the exit is used anywhere in my code.
Files: research-box /root/fwa3/v3 (v3pools.py, v3pull.py, v3cand.py, v3lp2.py, v3sim3.py, cmp.py, mc5_a3_*.json, a1_trades.json = A1's own list from his code, read only).

## 13. mc5 re-entry after stop, and the supply fix (v15, 17:17 Bangkok)
Base book = mc5 spec (dust guard A1.20, +5 min, $250k to $1M, stop 0.5x, 4 h), my rebuild: 172 trades, 116 hit the stop. Extra chain logs fetched for those 116 pools to first swap + 10.5 h (115 complete, 1 too recent).
- After our stop: 48 of 115 trade back above our entry price, all of them within 2 h; 29 of 115 reach a $1M cap within 6 h (pulls end the look).
- Second buy: one per coin, only on stopped coins, trigger on a swap after our stop fill, $100 at trigger + 6 s, stop 0.5x, pull = 0, sell 4 h after the second buy. Fees, gas, depth as in section 12. $ a day = per 12 days a half.

| rule (window after stop) | H1 n / mean / $ a day | H2 n / mean / $ a day | 2nd window coins |
|---|---|---|---|
| price back to 0.8x of our entry, 2 h | 8 / 5.05x / +270 | 51 / 0.96x / -15 | 15 of 59 |
| price back above our entry, 2 h | 7 / 7.10x / +356 | 41 / 0.79x / -70 | 15 of 48 |
| price 1.5x our entry, 2 h | 6 / 7.27x / +313 | 28 / 1.61x / +141 | 14 of 34 |
| cap back to $500k, 1 h | 6 / 1.38x / +19 | 37 / 1.78x / +239 | 15 of 43 |
| cap back to $500k, 2 h | 7 / 6.72x / +333 | 38 / 1.74x / +234 | 15 of 45 |
| cap back to $750k, 2 h | 7 / 5.46x / +260 | 28 / 1.42x / +99 | 14 of 35 |
| cap back to $1M, 2 h | 6 / 0.95x / -2 | 21 / 1.42x / +73 | 14 of 27 |
| buys per 10 min back to half the first 5 min pace, 2 h | 7 / 7.60x / +385 | 43 / 0.68x / -113 | 15 of 50 |
| buys per 10 min back to the first 5 min pace, 2 h (H1 PICK) | 5 / 12.15x / +464 | 24 / 0.90x / -21 | 11 of 29 |
- Buy pace = count of buy swaps of 0.001+ WETH (wallet identity not used). 1 h windows are in /root/fwa3/v3/reentry.json.
- Walk forward: the H1 pick fails H2 (0.90x). H1 has only 5 to 8 trades a rule and one coin, 0x3c934eee3fd33be89d0c0a5d073dcf2ea3b3dc26 (pool 0xdb8b039a1db13d8a0c2b04367973d24b3c27ded0, 38x to 57x), makes every H1 number.
- Best both halves, cap back to $500k within 2 h (about 1.9 trades a day): without the best coin per half 1.38x / 1.05x; without the best 3, 0.34x / 0.57x; median 0.49x / 0.45x. In H2 the 11 coins the 2nd window also trades make 3.84x, the other 27 make 0.88x. So the money is coins the $1M gate buys anyway.
- Its 5 biggest: 38.75x 0x3c934eee3fd33be89d0c0a5d073dcf2ea3b3dc26 (pool 0xdb8b039a1db13d8a0c2b04367973d24b3c27ded0, H1); 27.24x 0xee049979960d5e42b389b941ec45790ca7b6e1f7 (pool 0x92a4f531f82e1440d6e349bbc2ed42ba389e013f, H2, w2 coin); 9.54x 0xb6039ac45d5d355dd98f318dfa622d699e9556e8 (pool 0x932d3719bd7710e4d9e52518933b55a4028950fa, H2); 9.45x 0xf119ae24ecc34e90bea3ca171dd86e5309859c17 (pool 0xaf36a2b5f81ab17c194462654e211b569a1f3cdc, H2, w2 coin); 6.43x 0xa2508b15ab16826c720631b9afa0a90c453ae8a4 (pool 0xaf69d9d9d0dbeaaa2e03eddbd7971d5b536627ba, H1, w2 coin).
- VERDICT: KILLED as a rule. At most a paper watch line "cap back to $500k within 2 h of our stop", only if the 2nd window does NOT hold the coin.

Supply fix, for FW Analyst 1 (exact read): totalSupply = eth_call to the token, data 0x18160ddd, at block b300 = the block of the last Swap with block time before T0 + 300 s (the same swap whose sqrtPriceX96 gives the price); decimals from 0x313ce567 at that block. Never "latest".
- Rerun of A1's own code (v3frag.py, his cache) with that supply, 2 h: 200 trades (was 201), H1 41 / 1.89x (was 42 / 1.86x), H2 159 / 1.58x (same), +$12,887 (was +$12,826). Supply differed on 32 pools; 16 pools had no at-block read and kept A1's value. One coin left the band: 0x21c5b1f2a50cf8657ad5409fdfce39ae1438782e (0.39x). Changes nothing.
Files: research-box /root/fwa3/v3 v3ext10.py, v3re.py, reentry.json, a1supfix.py.

## 14. mc5 honest money line (v16, 17:23 Bangkok)
Method (my rebuild, dust guard on): cap $250k to $1M at T0+300 s. Fill at the END of the block at T0+300+6 s (or +30 s) by a real v3 tick walk over every Mint/Burn range in the pool. Exits: stop 0.5x (checked on 0.01+ WETH swaps, value with our own buy kept), pull (Burn of 50%+ of in-range minted units), time stop fill + 4 h. EVERY exit is sold at the END of the NEXT block (cascades and pulls included, so a pull exit is what the pool still pays, not a forced 0), by tick walk with our own buy kept in the pool, capped at 90% of pool WETH. GMGN 1% each way + $0.05 gas each way. Days: H1 13 to 24 Sep (12), H2 25 Sep to 6 Oct 12:00 (11.5). 171 trades (H1 22, H2 149). Worst day by launch day; max drawdown on the equity curve in exit order, in tickets.

Fill 6 s:
| size | half | mean x | $ a day | worst day | worst 3 days | max DD tickets | $ a day without top 3 |
|---|---|---|---|---|---|---|---|
| $100 | H1 | 4.37x | +618 | -74 | -74 | 2.8 | +124 |
| $100 | H2 | 2.31x | +1,699 | -537 | +782 | 13.7 | +783 |
| $500 | H1 | 4.14x | +2,874 | -341 | -341 | 3.3 | +550 |
| $500 | H2 | 2.21x | +7,821 | -2,665 | +3,403 | 13.6 | +3,556 |
| $1k | H1 | 3.92x | +5,349 | -621 | -621 | 3.3 | +1,017 |
| $1k | H2 | 2.09x | +14,160 | -5,300 | +5,711 | 13.6 | +6,304 |
| $2k | H1 | 3.53x | +9,279 | -1,039 | -1,039 | 3.7 | +1,648 |
| $2k | H2 | 1.93x | +24,082 | -10,713 | +7,341 | 13.4 | +10,510 |

Fill 30 s:
| size | half | mean x | $ a day | worst day | worst 3 days | max DD tickets | $ a day without top 3 |
|---|---|---|---|---|---|---|---|
| $100 | H1 | 4.37x | +618 | -75 | -75 | 3.3 | +126 |
| $100 | H2 | 2.03x | +1,339 | -571 | +470 | 13.8 | +431 |
| $500 | H1 | 4.16x | +2,893 | -349 | -349 | 3.3 | +580 |
| $500 | H2 | 2.04x | +6,742 | -2,833 | +1,979 | 14.1 | +2,559 |
| $1k | H1 | 3.91x | +5,335 | -637 | -637 | 3.8 | +1,031 |
| $1k | H2 | 1.94x | +12,119 | -5,615 | +3,120 | 14.2 | +4,505 |
| $2k | H1 | 3.54x | +9,309 | -1,072 | -1,072 | 3.7 | +1,748 |
| $2k | H2 | 1.78x | +20,341 | -11,117 | +2,041 | 13.0 | +7,429 |

- WARNING, 6 Oct morning carries a lot: 3 trades on 6 Oct (incl. 0xc38c332012a9116dcade0c7f43b5d3c71799a9f8, pool 0xb4b0dfeaa2e2a838f0f7a304f005ebe32390d4ef, 47x). H2 through 5 Oct only (11 days, 146 trades), fill 6 s: $100 1.84x +$1,118 a day (no top 3 +$484); $1k 1.69x +$9,195 (no top 3 +$3,703); $2k 1.58x +$15,524 (no top 3 +$5,837). Fill 30 s: $100 1.64x +$847 (+$196); $1k 1.60x +$7,965 (+$2,393); $2k 1.50x +$13,170 (+$3,540). USE THESE for the ramp, not the 6 Oct inclusive line.
- Cross checks: $100 through 5 Oct 1.84x vs A1.18 1.67x and my model book 1.77x; $1k 4 h H2 2.09x vs FW Analyst 2 size curve 2.04x (main model). Honest is not lower than the model, most likely because our own buy stays in the pool on the sell and the tick walk pays more than the 90% pool cap on the big winners; the frictions (next block exit, late fill) cost the stops, not the winners.
- Fill delay matters: 30 s instead of 6 s cuts H2 $ a day by 14% to 21% (13% to 24% through 5 Oct), and without the top 3 coins by 28% to 45%. Live fill must stay under 10 s.
- Size: x falls from 2.31x ($100) to 1.93x ($2k) in H2, $ a day still rises to $2k. Max drawdown about 14 tickets at every size (matches FW Analyst 2's ~15), so wallet = 20+ tickets as decided.

Stops that filled under 0.3x (fill 6 s, $100): 19 of 116 stops. x at $100 ($1k), pool, token, launch time Bangkok:
- 0.004x (0.007x) pool 0x6d96bd36c25688f284e5338c4a594e511de1239e token 0x1b0fddda68bf1024ffdf3c2c32328eb8a36aab8b, 2 Oct 03:14
- 0.007x (0.012x) pool 0xd6f935dffc76df700e7a59958d42d407680a61c9 token 0x9bc9eee7339431c380d7ad4de78c9a9535e2f76d, 1 Oct 15:01
- 0.007x (0.011x) pool 0xa5d29217e702c80cf0a959ed0fcc4f4cdfd3079d token 0x6e3895bcd6ded5731f01d9b1cf32484f62ab7c01, 4 Oct 23:13
- 0.009x (0.015x) pool 0x3042183dbb73d5c4f42dba677c18cee843183480 token 0x8b462bd24331ab0593f4c604eaf9d3844fd7f549, 26 Sep 14:36
- 0.010x (0.017x) pool 0xca6b9035bc39cba3f6c7e319e62ac028b02da555 token 0xd7fd11c4175dcd270540647636f5f8aa3b946662, 4 Oct 03:00
- 0.012x (0.017x) pool 0x01f47a2457edfff954f3664c3c7956defe774f99 token 0x364d0606e55b546ea76d038d38265116dfa36626, 25 Sep 00:05
- 0.020x (0.028x) pool 0xb21e8317c5c11aa7bca3c54a1c6b87f581c51aa7 token 0x94da0360307693bc1691490381229aea422f9b74, 29 Sep 18:25
- 0.061x (0.071x) pool 0x4a18d53e80ab9f3b08b4cf36c2fe855edbca3a8b token 0x7cb499776cef107a94aa44f8d00c9a116294c3ea, 5 Oct 02:30
- 0.061x (0.072x) pool 0xbe0048170200c178a9bd1318c53949860c6d82d4 token 0x55683371036401f8cae676c260efcde9be36a7f4, 22 Sep 20:40
- 0.084x (0.097x) pool 0x78ade5b59e5a64b0199afd65eab541e7f88d4479 token 0x95a23cf1056e8e9f7218fc58e1f0a1ff830e3872, 5 Oct 20:19
- 0.090x (0.097x) pool 0xa4277e13a0818e745bb87404ff3df13cdeae60f5 token 0x6e1b7965cdf9a0f5b9a02185dc003b9a59b2c3cb, 26 Sep 05:11
- 0.093x (0.106x) pool 0xaf69d9d9d0dbeaaa2e03eddbd7971d5b536627ba token 0xa2508b15ab16826c720631b9afa0a90c453ae8a4, 23 Sep 09:25
- 0.094x (0.108x) pool 0xf736ff6d7790366a235d4056de8ba2ee73883d0e token 0xe9fa0937aa06dcc492aef5d76277d954cc42fad7, 4 Oct 05:38
- 0.138x (0.146x) pool 0x67987f40926b77b643a87d388dfe98f268c40948 token 0x60eafc81068bb9077fd8e5733d8b11bb2523747a, 29 Sep 01:12
- 0.189x (0.210x) pool 0x7863632dffc1d018d1dcbd4776e749098bf375c4 token 0xf898ba10200ec574f6da78087fd726b7335f8934, 4 Oct 05:32
- 0.190x (0.210x) pool 0x8201bb87745632601c1a193d1db44867f85f2389 token 0x01683c023eb6a9172d167feafd48fb9cc5adc1b9, 28 Sep 15:54
- 0.244x (0.258x) pool 0x0ab2ce1be4b982a969d5da6cccdeae98dc98e4fc token 0x5099340b1b75f257ec61496debe1fa0d86c25a84, 28 Sep 07:35
- 0.263x (0.379x) pool 0xb1131cafdcfacdd4291ca35414577b4584e9a8e7 token 0x39703c160e991e2961697d4b2b3c64136df25338, 21 Sep 09:44
- 0.290x (0.304x) pool 0x6cd500eb85c2726e2f4330d62aedd8f94b49a3cc token 0x4a1c8766c308a6b4eef1d9fc95173bc00478b562, 25 Sep 11:39
- 7 of these fill at 0.02x or less = a pull or a one-block dump right after our stop swap; nothing at the next block can save them. The 30 s fill list adds pool 0x134c48fd7cd15ba64124d9a5bc1273c7115c03b2 (0.113x) and pool 0x92576baad945353c1340ce320ae9ec530d6e4cdc (0.291x at $100, 0.029x at $1k).
Files: research-box /root/fwa3/v3 v3honest.py, honestana.py, honest_D6.json, honest_D30.json.

## 15. mc5 fill delay curve (v17, 17:27 Bangkok)
Same honest method as section 14 (end of block fills, next block exits, tick walk, own buy kept, 1% + gas, 4 h). 171 coins in every run. $ a day; % of the 2 s line in brackets.

$1k a trade:
| fill after check | H1 | H2 to 5 Oct | H2 incl. 6 Oct morning |
|---|---|---|---|
| 2 s | +5,670 (100%) | +8,884 (100%) | +13,195 (100%) |
| 4 s | +5,407 (95%) | +8,793 (99%) | +13,954 (106%) |
| 6 s | +5,349 (94%) | +9,195 (104%) | +14,160 (107%) |
| 10 s | +4,927 (87%) | +7,859 (89%) | +12,488 (95%) |
| 15 s | +4,778 (84%) | +8,243 (93%) | +12,660 (96%) |
| 20 s | +5,084 (90%) | +8,214 (92%) | +12,495 (95%) |
| 30 s | +5,335 (94%) | +7,965 (90%) | +12,119 (92%) |

$2k a trade:
| fill after check | H1 | H2 to 5 Oct | H2 incl. 6 Oct morning |
|---|---|---|---|
| 2 s | +9,842 (100%) | +15,016 (100%) | +22,501 (100%) |
| 4 s | +9,444 (96%) | +15,567 (104%) | +24,399 (108%) |
| 6 s | +9,279 (94%) | +15,524 (103%) | +24,082 (107%) |
| 10 s | +8,672 (88%) | +14,137 (94%) | +22,117 (98%) |
| 15 s | +8,440 (86%) | +13,558 (90%) | +21,229 (94%) |
| 20 s | +8,943 (91%) | +13,548 (90%) | +20,992 (93%) |
| 30 s | +9,309 (95%) | +13,170 (88%) | +20,341 (90%) |

- PASS LINE for fmmc5 live (check to fill, measured as fill block time minus the +5 min check time): 6 s or less = pass. 7 to 9 s = warning. 10 s or more = fail (loses 9% to 15% against the best of 2 to 6 s in every column). Best of the plateau is 4 to 6 s in H2; 2 s is not better.
- The curve is not smooth past 10 s (15 to 30 s sometimes beats 10 s): a few knife-edge coins flip between the 0.5x stop and a big run depending on the exact fill price. Without the 3 biggest knife-edge coins, H2 to 5 Oct at $1k: 2 s +8,412, 4 s +8,840, 6 s +8,714, 10 s +7,905, 15 s +8,292, 20 s +8,262, 30 s +8,012 (10 s still the first drop past 5%).

Knife-edge coins ($1k x at 2 / 4 / 6 / 10 / 15 / 20 / 30 s):
- pool 0x3939ce55ec4e0ff8df9aa6469a795c793da06d55 token 0x224a794031d54e49b46919ade2c55ff255dfdd97 (6 Oct 06:44): 17.19 / 26.72 / 24.41 / 20.45 / 17.75 / 16.14 / 13.02
- pool 0xc8b416104e0b4bd5e5e14489f90ac807b67300e2 token 0x55d598f8b2c3528e47062b760d6fc66cfd2a2578 (22 Sep 08:49): 20.82 / 18.36 / 18.44 / 17.44 / 14.06 / 17.64 / 20.33
- pool 0xa402616f4d05cca33947fabc6ff51ba795f6d2f4 token 0xbed6314fe7b46707aff9863c5e5838aa93397880 (29 Sep 05:29): 6.20 / 0.49 / 6.29 / 0.49 / 0.46 / 0.48 / 0.49 (stop or run by one swap)
- pool 0xe1f227ad6f50b3c8905e75ed621610becf4510d9 token 0xead015725c6cc0a38cdaea04cb7933210198b830 (25 Sep 07:27): 16.80 / 16.37 / 16.48 / 16.31 / 19.66 / 18.34 / 14.77
- pool 0x1a2822ccba364190fe945e80c14493c3964a494d token 0xe872a965b354a86545520360011fb824ed27db6c (25 Sep 07:03): 15.06 / 16.46 / 16.46 / 16.92 / 18.25 / 17.92 / 19.71
- pool 0xeb1646dbfbeacf76483df1092c208388c029c534 token 0x257723230c9cccf6a2df1e7cf67cffcfcf01eeb8 (30 Sep 08:26): 13.44 / 17.57 / 17.82 / 16.62 / 15.83 / 14.77 / 14.00
- pool 0xa6b2579671f40524b08ba53871f19599edeeb3e8 token 0x27e1ce669d481827096ab2e45d1b80f34fb41137 (29 Sep 16:31): 4.12 / 4.22 / 3.93 / 0.49 / 0.50 / 0.48 / 0.49 (fill after 10 s hits the stop)
- On the big runners a later fill also shifts the 4 h time stop (it counts from the fill), so part of the swing is the exit time, not the entry.
Files: research-box /root/fwa3/v3 honest_D2/4/6/10/15/20/30.json, delayana.py.

## 16. Live fill audit and half C scoreboard (v18, 17:52 Bangkok)
- Page: http://159.223.88.168/hx-c9ois/fw/a3-live-audit.md (rebuilt every 2 min on research-box, pulled to the droplet every 2 min).
- Live audit (research-box /root/fwa3/v3/liveaudit.py, timer fwa3-live, every 2 min): reads ONLY chain logs. Every token Transfer to or from 0x0184ef9b23476ed84107cc5e15485a4a7ee6b286, then the v3 Swap event inside that tx (pool, WETH in or out, tokens). Buy: pool first swap block, check time = first swap + 300 s, delay = buy block time minus check time, verdict PASS 6 s or less / WARN 7 to 9 s / FAIL 10 s+; tokens received vs my tick walk at the end of the block before (Mint/Burn ranges + last swap). Sell: WETH out vs the same model, hold minutes, realised x. Also checks the tx signer is the wallet.
- Validation: run on a test wallet (bot contract 0x516e29c9b5b37c4fb4bbc96ca249f180979efd20, pool 0x4fdcc31f5514644bf512b296057f025ddd4f9308): buy fill / model 1.0000000000, sell 1.0000000000.
- Half C (halfc_a3.py, timer fwa3-halfc, every 2 h at :15): new PoolCreated since my last scan + my older pools with first swap from 6 Oct 12:00 Bangkok; first swap, cap at +300 s (supply at that block), candidates $150k to $1.5M get full logs to first swap + 4 h 15 min, then the section 14 honest sim (fill +6 s). Closed mean vs the H2 (to 5 Oct) honest result, with a bootstrap band for the same number of trades (5th / 50th / 95th percentile); the page says BELOW when half C runs under the 5th percentile.
- Gap I know of: a pool created before block 81484227 whose FIRST swap came later is not picked up (rare, I will add a sweep if half C stays small).
- I check the page every 15 min and ring on the first live trade and on any FAIL (delay 10 s+, or fill under 97% of my model).

## Files
- research-box /root/fwa3: inits.jsonl (every v4 launch 28 Sep to 6 Oct 13:01), coins.jsonl (lane sims), lanefill.json (paper vs depth per trade), scripts book.py, book5.py, ana.py, wf.py, pulltest.py, decay.py, lwatch.py.

Version 52, 7 Oct 2026 16:32 Bangkok. (Times of v3 and v4 corrected: they were published 14:22 and 14:31 Bangkok, not 14:50 and 15:15.) Latest: http://159.223.88.168/hx-c9ois/fw/fw-analyst3-status.md