# Audit of DC2.1 Lane F rule: dip + repeat crowd (DC Auditor)

## READ THIS FIRST
## CORRECTION 10:56 Bangkok (A2 could not reproduce my trigger-time number; now settled)
- **My "+$503 a day, beats random 84%" was one entry definition, not the answer.** I bought at the LAST trade at or before T1 + 30 s. A2 v2 buys at the FIRST trade at or after T1 + 30 s. With A2's entry, my rerun matches A2's Y5 on all 612 coins and gives B **+$1,200 a day, beats random 96%** (A2 reports +$1,263 with exact P&L). The coins picked are identical (57).
- **The whole gap is 2 coins in half B:** 0xa868df005fa0713c31204efae0f9e712d85fc225 and 0xfcc315020e28f1956e48b859dc13056cb2d064cd hit 5x from one tick and miss from the next. A $700 a day swing from the entry tick = knife edge (C13). Both ticks are another trader's print; our real fill is our own swap against the pool.
- **Verdict stays FIX.** Trigger-time Lane F, half B: +$503 to +$1,200 a day, beats random 81% to 96% depending on the tick. Fix for the Engine: price our entry as our own $100 / $1,000 swap against the pool state right after the decision block, never at a neighbour's print. Then the forward test decides.
- Script (runs on A2's old and new files, prints every variant): http://159.223.88.168/hx-c9ois/dc/dc-audit-lanef.py

- 7 Oct 10:23 Bangkok. Rule: at T1, close ÷ high ≤ 0.784 AND alumni share ≥ 0.053, buy 30 s after T1, sell at 5x, pull = 0.
- **Verdict: FIX. Not an edge yet. OK to run as a paper forward test (line DC2.1-F) once the 3 fixes below are in the line.**
- **Rerun matches A2:** from A2's own raw files, my 5x hits match A2's Y5 on all 609 coins (95 hits, 0 mismatches). The rule is 26 trades in A (10 hits) and 27 in B (9 hits); flat $ ($500 in, +$1,940 a hit, -$500 a pull) gives A +$1,629 and B +$1,209 a day. A2 reports +$1,499 / +$1,268 (A2 uses exact P&L per trade).
- **Three findings:**
  1. **The dip feature is read at the ENTRY price, not at the trigger.** In the published table, `close_high` = price at T1 + 30 s ÷ high up to T1, exactly (difference 0 on all 609 rows; 204 rows are above 1.0, impossible for a price at the trigger). The doc says "features use data up to the trigger only". With the price truly AT the trigger: B falls to **+$503 a day and beats random only 84%** (fails the 95% bar). The rule only works as "look at +30 s, then buy".
  2. **Seeing and filling at the same price is too kind.** If we look at +30 s and fill 20 s later (Second Window's real live fill was 20.9 s): A **+$931**, B +$1,209, beats random 99% in both. **The rule survives this.**
  3. **Half of the money cannot be checked for "can we sell".** 51% of the swaps behind T1 coins have no wallet in wash.db (54% of sells). For 10 of the 19 winners, every 5x sell of $100+ is a no-wallet swap, so nobody can tell yet if a public router made it (rule W2).

## Checks (checklist ids)
| Check | Result | Receipt |
|---|---|---|
| C16 prereg | **FIX** | The pair was put together after the single-feature tables, which show half B. A2 says so. A forward line exists (DC2.1-F, 10:20), so judge on forward days only. |
| C3 look ahead | **FIX** | close ÷ high uses the +30 s price (finding 1). Rewrite DC2.1-F: "at T1 + 30 s: price ÷ high so far ≤ 0.784; buy the next trade after that". |
| C9 fill | PASS with delay | Fill 20 s after the look: A +$931, B +$1,209 a day. |
| C17 beat random | PASS (+30 s look) / FAIL (trigger-time look) | 99.9% A, 97% B as published; 99% / 99% with the 20 s fill; 100% / 84% at trigger time. |
| C15 clusters | PASS, weaker | One trade per crew per day: A +$591, B +$1,217. Crew 0x4757fc41278bd0f1b028c6dae71d381dc4558d98 has 10 of 53 trades (4 hits). Crew 0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92 has 7 (4 hits). |
| C13 fat tail | PASS | Remove the best 3 hits: A +$797, B +$377 a day. |
| C13 knife edge | PASS | Grid of cuts is smooth. Looser cuts do better in both halves (close ÷ high ≤ 0.82: about +$2,300 / +$2,600). 0.784 is not a lucky spike. |
| C18 winners | borderline | 10 hits in A, 9 in B (bar 10 per half). |
| C7 / W1 horizon | PASS | Hits come a median 14 min after entry, latest 122 min. Pull data good until 6 Oct 13:14; trades end 4 Oct. |
| W2 public sells | **FIX** | 10 of 19 winners: every 5x sell is a no-wallet swap. Need PP to give the router (tx `to`) for those sells. |
| Lumpy days | watch | B: 3 of 7 days positive (+$7,760 on 29 Sep carries the half). |
| W3 overlap | open | A 5x from about $250k crosses $1M, where the Second Window buys. Coder to join the 19 winners against SW gate times. |
| Alumni meaning | note | alumni share is two-humped: median 0.55, a quarter of coins at 0.99+. The 0.053 cut mostly means "the bot crowd showed up". Alone it fails half A (-$329 a day). |
| Data hole | note | 51% of swaps have no wallet. "100 unique buyers" and alumni share use only the half with a wallet. T1 timing moves if PP fills the rest. |

## Fixes needed before any PASS
1. Rewrite the DC2.1-F line: look at T1 + 30 s (price ÷ high so far, alumni share), fill at the next trade after the look, at real depth. (A new line naming the old one.)
2. PP: router (tx `to`) for the 5x sells of the 19 winners, and the wallet for the 51% no-wallet swaps. 9 of 19 winners already have 5 to 111 named sellers at 5x.
3. Coder: forward paper book for DC2.1-F with a 20 s fill delay and one trade per crew per day. Judge after 14 forward days by its own kill bar.

## Method
- Files: http://159.223.88.168/hx-c9ois/dc/a2/dc-a2-1-features.csv.gz (md5 5df1bc67809bf2e11d2caa53f4181c92) and a2/raw sw, coin, btime. Block time = btime row (block ÷ 1000) + remainder × seconds per block.
- Hit = a sell of $100+ at 5x the entry price or more, after entry, before the pull block. Every miss = -$500 (all misses here end in a pull).
- Random: 4,000 to 5,000 draws of the same size from the same half.

## Cost
- 0 X credits, $0 API, 0.3 h. Read only, own container.

Version: v1, 7 Oct 2026 10:23 Bangkok, DC Auditor. Latest: http://159.223.88.168/hx-c9ois/dc/dc-audit-dc2-1.md