# A2.39 The missed runners: a holder-filter door (Analyst 2)

7 Oct 2026, 17:35 Bangkok (server clock). Pre-registered 14:52 in the status page, before any result. Late against my 16:45 ETA by 50 min (why: one bug in my book code set the forced 5x price to empty and had to be found and fixed before any number could be trusted).

## Answer
**FAIL. Do not open the door. The holder filters (top 0.08, army 0.015) are earning their keep.**

- **Frozen cell D1** (blocked coins whose blocking holders sold back at most 10% before the gate), picked on half A:
  - half A: +$70 a day vs live;
  - half B: **-$92 a day** vs live.
- **Worst day half B:** -$6,480 with the door (live -$2,898). The bar is -$3,951, so it breaks.
- **Beats random:** 2.5% (A) and 6.5% (B). The bar is 95%. Random tick-set pools bought at the same minutes did better than the door coins.
- **Without the top coins:**
  - without top 1: A -$255, B -$92;
  - without top 2: A -$255, B -$705.
  - Top 1 is 0x128f8bea785bdea9dc779cffa4740cd2e82dbc3b (6.53x, A). Top 2 is 0x56fa3670e19396c4ee8f7775b49ceaede04a84ef (4.68x, B).
- **Why it fails:** the blocked coins are poor.
  - 75 coins are blocked by the filters: the median is 0.28x and only 5% reach about 5x.
  - The 70 live coins: 34% reach about 5x.
  - The door catches a few runners, but it pays for them with many losers.

## What is new
- A2.26, A2.37, A2.38 and A1.15 changed WHEN or AT WHAT CAP we buy.
- This test keeps the live $1M gate and its timing. It asks only whether the coins the holder filters block can be bought safely when the concentrated holders are still sitting.
- 9 of the 13 catchable scorecard misses since 1 Oct were blocked by these filters (army or sell check 6, receiver 3).
- **Mechanism tested:** the concentrated holders are an operator crew; while they do not sell, they are the bid.
  - **Result:** it is not true on this tape. Sitting crews still dump after the gate.
  - Example: 0xb6039ac45d5d355dd98f318dfa622d699e9556e8 had sold 0% before the gate and still went 0.28x.

## Cells ($ a day vs live, $1k, exits at the crossing swap = main)
| cell | rule | A | B | worst A | worst B | door trades A / B |
|---|---|---|---|---|---|---|
| live | | +2,904 | +10,008 | -2,000 | -2,898 | |
| D0 waive | every blocked crossing | -13 | +29 | -1,709 | -4,677 | 50 / 23 |
| **D1 sitting (picked on A)** | blocking holders sold at most 10% | **+70** | **-92** | -848 | **-6,480** | 42 / 14 |
| D2 kit | blocked and Kit A/B | -19 | +371 | 0 | -2,538 | 27 / 14 |

- D2 is the best half B cell, but D1 was picked on half A, as pre-registered. D2 lost in half A and adds only +$371 a day in half B on 14 trades.
- **Check with exits at the end of the next block** (old A2.21 pricing): live A +2,870 / B +9,631.
  - D0 shows +$3,348 a day in half A. All of it is one coin, 0x79f03c47d26a8c75bcae351804683c80facf53ac: 62.56x if sold at the end of the next block, 3.92x at the crossing swap.
  - That is a pricing artifact, not edge. Crossing-swap fills (Exec X5) remove it.

## The 13 catchable scorecard misses since 1 Oct
- **Door D1 takes 3 of them:**
  - 0x577bc74a99aabbd43a354668dc7b7ed4b5e303ae, 3.56x;
  - 0x56fa3670e19396c4ee8f7775b49ceaede04a84ef, 4.68x;
  - 0xb6039ac45d5d355dd98f318dfa622d699e9556e8, 0.28x.
- **4 are blocked but their holders had already sold more than 10% (D1 skips them):**
  - 0x74ccd8416cbc4e4a864176c2707bc1434819ef0f: 4.24x, sold 100%;
  - 0xf9d7691a78b96b7a1e4bb2d1f15fe63fabef885b: 3.01x, 14%;
  - 0xa72ee5f17ebd137b044815b7781cc944bea8011e: 0.28x;
  - 0xb78afd7a8ece14007e35fe0f1e2b3a774f8714c8: 0.28x.
  - So the "sold" check is not a good separator.
- **3 were already taken by live in my archive:** they are the "entry after halfway" coins:
  - 0xff4c3278e735fa52b1012b06748407d26e047f18;
  - 0xfe0f309cee96912d9926b8f0786fe3347364e678;
  - 0xe4aadf624b3a2c44d80dc220832c9c998d39c18f.
  - A filter door cannot speed them up.
- **3 are outside my archive** (after 6 Oct 10:58, or no $1M gate within 3 h):
  - 0xb18a4bcf9018bedb1fcb90ed80f87087151c09f4;
  - 0x9e81d57799ad169eea523bf98addc6c9c224228e;
  - 0x40bd44f364bce2b3f842462041cdd9356aedf0c6.

## Runners caught (my archive, $2M+ approx fair peak, after minute 1 and before halfway)
| half | runners | halfway inside minute 1 | live caught | live + D1 caught |
|---|---|---|---|---|
| A | 47 | 6 | 22 | 26 |
| B | 23 | 5 | 9 | 10 |

The door adds 5 runners caught across both halves and still loses money.

## Overlap with FW
- 0 coins. FW started trading 6 Oct 17:26 and my archive ends 6 Oct 10:58, so no door coin can overlap.
- This cannot be checked further on archive data.

## Setup (as pre-registered)
- **Universe:** $1M sellable gate, age at most 3 h, coverage ok, in the 944 pool archive.
  - 145 crossings: 70 live, 75 blocked by top over 0.08 or army over 0.015.
- **Entry:** gate + 30 s, $1k, real tick depth for every pool, 1% fee each side.
- **Exits:** live exits, priced at the crossing swap.
- **Slots:** 3 normal + 2 holds, shared with live.
- **Live re-check:** the live book reproduces the A2.27 trades exactly (70 of 70, max diff 0.0).
- **Blocking holders:**
  - the army wallets if army is over 0.015;
  - the top buy receiver if top is over 0.08.
  - "Sold" means tokens sent to the coin's pools ÷ tokens received from them, before the gate block.
- **Random:** 200 draws. Each door trade is swapped for a random tick-set pool that is not one of the 145, bought at the same minutes after its first swap, with the same exits.
  - Caveat: the tick set holds pools that reached $500k, which leans toward winners. That makes the bar harder, but the door is far below it anyway (6.5%).

## Limits
- Half B has only 14 door trades over 6 days.
- The runner peak is an approximation (max buy cap before the first half-pool pull), so its count differs from the scorecard's 32.

## Files
- http://159.223.88.168/hx-c9ois/sw/sw-a2-39-cells.csv (every cell, both exit pricings)
- http://159.223.88.168/hx-c9ois/sw/sw-a2-39-door-trades.csv (D1 door trades taken)
- http://159.223.88.168/hx-c9ois/sw/sw-a2-39-blocked-coins.csv (all 75 blocked coins: top, army, sold, x)
- Code: /tmp/w2/a239 (feat.py, kits.py, simlib.py, sim.py, book.py, cells.py, final.py), run in my container where the archive lives, nice 10.

cost: about 0.3 M credits (est.), API $0 (public RPC), 2.7 h wall (0.5 h compute)