The night, the analysis, and where it leads

Thomas asked at 07:21 for four lanes of analysis and at 10:19 for one place to read all of it: a page per lane, one for Experiment A, one for everything else, and one with the recommended state. This is that place. Start with Recommended state if you have five minutes; the lane pages carry the evidence.

Findings in one screen

1. The entire live loss is rugs. Live realized -$206.89 over 19 lots; the five liquidity_pulled writeoffs are -$250. Excluding them the live book is +$43.11 (positions). So the rug gate is not a safety nicety, it is the P&L.
2. The rug gate is one config line and it is free. filters.rugRatio.underAgeMin 10→60 and maxLiqToMcap 0.80→0.85 refuses ZEP, PEZZED, PEZ, PEZZEL, PENZ, SEPE and, in the feed rows checked, nothing else (rugs). Zero RPC, zero latency, the field is already on every feed row. This is the highest-value single line across all four reports.
3. A wallet rug-rate gate would have refused all three of the night's rugs at signal time with no on-chain call. smartmoney.json already carried frankdegods: 6 rugs / 39 entries (15.4%) twenty-five minutes before the ZEP signal; nothing in filterSignal() reads it (rugs). Cost: a ~0.15 ceiling also drops ~a quarter of the cohort incl. fr3ak (which gave hoodrat +$43). That trade-off is the one call I want from you.
4. Spend nothing on execution infra. Free Alchemy tier + the push path covers the design with 1.75x headroom and is already the faster endpoint (67ms vs 150ms). The seconds that remain are 70% our own follower code, not the network (execution). Alchemy Pay-As-You-Go has no base fee, so enabling it as an insurance switch costs $0 until called.
5. The exit ladder banks 30% of what the tape offered. One mechanical fix survives every price-path reconstruction: move the first clip +60→+40, because the trail only arms after a clip (positions). Drop the trailSteps step (it was fitted on a 32-call sample without LIQLORACLE).
6. Experiment A is not trending to kill, but it is running on one leg. D4 count is 23 at hour 8.4 (projects ~66 vs a <30 kill line), but leg B (+60% in 60m) has fired 0 times because the tape worker is stuck (9 rows retried ~130x each) and peak60mPct exists on 5 of 5637 rows. The Sun 23:00 kill decision would fire on half a signal (capture).
7. The runners we never saw are the real gap, not the ones we refused. 12 tokens ran +100% overnight with no proven-wallet buy; 10 of those 12 are invisible to the Fomo feed too. The only structurally-earlier population is block-0 launch snipers, and reading it is a paid eth_getLogs budget question that ties directly to the execution lane's Alchemy call (capture).

Pages

Built 06/09/2026 20:50:01 Beirut. Source files under projects/crypto-2026-09/analysis-2026-09-06/.