Recommended state
Where the live system stands at noon Monday, what is in flight right now, what Rook ships next without asking, and the short list that is Thomas's. Read "Check first" if you have two minutes.
Recommended state
Rook, updated 2026-09-07 12:10 Beirut. Every "live" below was verified against the running process, the log or the chain at noon today, not from a report. The previous edition (2026-09-06 11:36) is archived at archive/recommended-state-2026-09-06-1136.md. Status words: live = running now, verified; in flight = a worker is building it right now, restart included; next = Rook ships it after the in-flight batch, no decision needed; decide = only Thomas.
Check first (what keeps me moving)
- Section 0, the read. If any of the six lines is wrong, that is the conversation.
- Decision 1 in section 9: Alchemy Pay As You Go. One toggle, no base fee. It unblocks the block-0 buyer population (the only population structurally earlier than our wallets) and is insurance for the push path that now carries every buy signal. Nothing else on the list is gated on you.
- The Gold v3 milestone in section 8. The clock restarted when the security layer went live at 10:50 today. Say if the success lines are the right ones; I judge against them Wednesday 11:00.
- Ticket size against a $348 wallet (section 6). $50 is 14% per lot, 29% with two open. I am leaving it. Say if you want $25 until the security layer has ten live lots behind it.
0. The read in six lines
- The live loss since Thursday is one failure mode. Ten write-offs are -$500 at cost; every other lot together is +$155. Not rugs in the old sense: per-holder sell blocks switched on by the owner 2 to 19 seconds after our buy, so our buy was the trigger. The old liquidity-pull gate could not see it by construction.
- Since 10:50 today nothing buys without passing an allow-list of the two launchpad token classes every winner ever came from, a real simulated sell with an 80% floor, and a 10 s post-entry sell re-probe. Replayed on chain state: 10 of 10 losses refused, 0 of 7 winners refused. It has not seen a live signal yet.
- Gold v2 itself is fine and slow. Nine candidates since go-live, all refused upstream on mcap or liquidity, none reached the new layer. The entry gates are the P&L; do not loosen them to make activity.
- Detection is solved for free. The Alchemy websocket has been the buy source since 00:05: 2,865 hits, zero phantoms, zero reconnects, lag 0 to 2 s. The poll loop is the fallback and has covered 687 ticks.
- Exits are on the fitted ladder: first clip 50% at +40, second 25% at +200, flat 40% trail, -45 stop, two lots max. Untested on a live winner since the change; the first one tells us if the +40 clip banks what the bars said.
- The remaining edge is data, not code: who funds the deployers and who buys in block 0. Both need either the Alchemy toggle or an indexer. That is the only place I am waiting on you.
1. Pre-buy security layer (new since the last edition)
| item | value | status | evidence |
|---|---|---|---|
| allow-list | launchpad-A (3248-byte shape, PUSH operands masked) and launchpad-B (EIP-1167 proxy to 0x3Be8…) only | live 10:50 | every one of our 16 non-losing buys since the cutover came from exactly these two factories. research/security/BUILD-REPORT-2026-09-07.md |
| deny lists | rug-template constant 0x109b1bd8, blacklist launchpad proxy, live wallet as owner() |
live 10:50 | present in every rug's bytecode, absent in PROOF and LMI; owner is a plain wallet on every rug, a contract or a revert on every winner |
| real sell simulation | buy plus full-bag sell in one simulated block, refuse under 80% back; Kyber path when no direct ETH pool | live 10:50 | PROOF 98.0%, LMI 94.2%, hoodrat 98.1% dry |
| post-entry re-probe | simulated sell of the whole bag every 10 s; exit on revert, liquidity -50%, or value under 50% of mark | live 10:50, in flight for Kyber-routed lots | today's gap: lots whose sell sim went through Kyber (the LMI, hoodrat shape) had no pool key and were skipped. Worker item 1 |
| buy-time honeypot probe | eth_simulateV1 buy then transfer-to-dead before signing, fail closed | live 07:16 | all four honeypot contracts revert from the exec wallet |
| exit side | drained-pool check on any quote token, salvage exit under 20% of basis, exit-side honeypot write-off, exit retry backoff with a hard stop after exhaustion | live 08:21 and 10:07 | the two PICODUCK blind spots and the 508 Relay 429s |
| replay | since the Gold v2 cutover on live chain state | done 10:4x | losses refused: CYBERDOG, SEPE, ZEP, PEZZED, PEZ, MEGADUCK, LEVER, STARK, both PICODUCKs. Winners passing: LIQLORACLE, PROOF, LMI, hoodrat, VERITY, MILADY, FOMO. The 9 ordinary stop-outs still pass, correctly |
| staged entry $10 then $40 | off, your call this morning | rollback for any layer is one flag in executor/security.json; global kill is touch data/executor/KILL |
2. Entry gate
| item | value | status | evidence |
|---|---|---|---|
| liquidity/mcap ceiling | 0.82 across the whole 2 to 60 min window | live | sweep, unchanged since yesterday 11:36 |
| wallet rug rate | 0.18 with >= 10 entries | live | money-best point on the tape, hands fr3ak back, blocks seralberttrades |
| handle deny list, burned wallet 24 h | frankdegods; skip after a liquidity-pull close | live | |
| max open lots | 2 | live 19:48 | drained-lot exclusion from the count: not verified, checking after the in-flight restart |
| tape guard | refuse when buys/sells > 4 or under 250 swaps in the first hour | wired, off (enabled: false) |
08:35 analysis of the 15 live lots since cutover; signals now carry txH1Buys and txH1Sells, so a week of columns decides it |
| copycat guard | refuse a symbol whose older twin in the registry has over 10x its mcap | in flight, worker item 4 | we bought copycat PEZ 0xcfc4 while the real PEZ 0xa360 (+5312%) sat refused in our own registry |
| heat +500%/1h, chase 25% over the wallet fill, cohort <= 1, mcap under $1M, liquidity $25k to $400k, age 2 to 60 min | unchanged | live | today's nine refusals: mcap cap 5, liquidity floor 4 |
pre-buy tells as log-only columns, Gold v2 tells / age180 / lowtrail paper twins |
next, after the in-flight batch | still not in rules-v2.json; sets loaded today are Gold v2, Gold, Gold-hold, Gold-flat, Hot, Mid |
3. Exit ladder
| item | value | status |
|---|---|---|
| first clip | 50% at +40 | live 19:48 |
| second clip | 25% at +200 | live |
| trail | flat 40% relative, trailSteps: [] |
live 19:48 |
| stop / breakeven | -45; breakeven after the first clip | live |
| no-progress 30 min under +10, time stop 90 min, wallet exit off | unchanged | live |
| exit retry | ladder 1000/1000/1500/2200 bps, 10 s cooldown, exponential backoff per lot, exit_exhausted hard stop |
live 10:07 |
| in-position checks | PoolManager liquidity on every mark, write-off at zero; security re-probe every 10 s | live |
| pnl card on close | whole-lot pnl, legs, peak, exit, gas in the Telegram close message | live 01:48 |
4. Detection and execution
| item | status | evidence |
|---|---|---|
| Alchemy websocket logs subscription as the buy source, poll as fallback | live 00:05 | shadow-state.json watch.ws: 2,865 hits, 238 sells, 0 phantoms, 0 reconnects, 0 errors, 9,389 polls skipped, 687 fallback polls; lag 0 to 2 s |
sequencer-feed detector (feed-watch.mjs) |
retired 00:05 | 2 hits in 3.05M transactions |
watch.lastAt liveness sentinel on the crontab, no LLM |
in flight, worker item 3 | the 87 min outage was invisible because the process never died; sentinel alerts at 3 min stale |
| public Robinhood RPC 429s on the pons feed | known, harmless | about 14 an hour all day, each recovers in under a minute; the security layer and the follower run on Alchemy, not on it |
| Alchemy tier | free | decide (section 9) |
| Relay API key | none on disk | decide, free signup; the backoff shipped this morning matters more |
5. Capture and data
| item | status | evidence |
|---|---|---|
| swap sender recorded in every debrief row | in flight, worker item 2, with a backfill of every existing debrief | without it deployer share, unique buyers and one-funder clustering are uncomputable. Biggest data gap in the rug work |
| block-0 buyer population | decide: Alchemy PAYG or an indexer | 10 of 12 invisible overnight runners were invisible to Fomo too |
| history indexer for the chain (Blockscout is Cloudflare-blocked from this box) | decide | "who funded these burner deployers" stays unanswerable without it |
smart-money follow list: macdegods, threem |
not added yet | top-4 by m60 hit, zero rugs; next batch with the paper twins |
| paper arena (Bolo) | reframed 11:35 as an idea bank and thesis competition, not a follower input | my five suggestions are in arena-2026-09-07/round-02/ROOK-SUGGESTIONS-2026-09-07.md; first one is to calibrate its cost model on our 26 real buys and 23 real sells before admitting a ninth idea |
6. Risk frame
| item | value | status |
|---|---|---|
| wallet | $348 (0.1398 ETH at noon) | about six tickets plus gas |
| ticket | $50 | keep unless you say $25 (Check first, 4) |
| max open | 2 | live |
| daily loss stop | off | keep off; the entry and security gates are the loss stop, a realized stop never fires in time on this failure mode |
| kill switch | touch data/executor/KILL |
live; clears with rm |
| rollback of the security layer | one flag per layer in executor/security.json, re-read per call |
live |
7. Sequencing, now
In flight (one worker, one commit per item, tests plus dry buy and dry sell before any restart, KILL on during the restart and cleared after the startup lines verify):
- Kyber re-probe for lots without a direct ETH pool.
- Swap sender in every debrief row, plus the backfill.
watch.lastAtstaleness sentinel on the crontab, every 2 min, alert at 3 min stale.- Copycat symbol guard, replayed so it refuses PEZ 0xcfc4 and none of the seven winners.
Next, mine, no decision needed: the three paper twins and the log-only tell columns; macdegods and threem on the follow list; verify the drained-lot exclusion from the open-lot count.
8. The milestone this sets up
Gold v3, 48 hours from the security layer going live. Judged Wednesday 2026-09-09 11:00 Beirut:
- zero honeypot or liquidity-pull write-offs on live money;
- every live lot row carries
tokenClass,poolandentryLiquidity, and every Kyber-routed lot shows a re-probe; - live book positive with no "without the rugs" qualifier;
- wallet buy to our fill under 8 s p50 on a real fill, detection under 1.5 s (websocket says yes; a live fill has to confirm it);
- the tape-guard columns and the three paper twins at n >= 20 so the next config change is chosen from data.
9. Needs Thomas
- Alchemy Pay As You Go toggle. No base fee, $0.45 per 1M CU, uncapped
eth_getLogs. Unblocks the block-0 population and insures the push path. - Ticket size at the current wallet: $50 stays unless you say $25.
- Relay API key, free signup, if you want it under your account rather than the bot's.
- History indexer for the chain: options and prices go on the execution page once I have priced them; say if it is worth the spend at all.
- Staged entry ($10 probe, then $40): off on your call this morning; say if the first live honeypot refusal changes that.