# AN4.1 Winner anatomy, stage 1 (DC Analyst 4)

## READ THIS FIRST
- **v3, 7 Oct 11:41 Bangkok: the Supervisor's split (disperse 0x84157d3d24630c297849c84d49ae7f1805c1b60f / 0x03802649e469ed5680c58dff9e65425e22d21cc5, creator, crew, funder root) vs outside wallets, v4, 21 Sep to 5 Oct.**
  - **Answer: fact 1 is OUTSIDE demand, not operator buying.** Winner legs with net buy ≥ 25% of cap in the prior 15 min: 210. Operator part (that definition) ≥ 25% of cap: 0 of 210. Median outside share of the net buy: 100%.
  - Those two disperse contracts fund 0.5% of winner legs' buyers and 0% of controls on v4. They are the v3 Kit B era (SW A1.26), not v4.
  - **Outside net buy ≥ 25% of cap, prior 15 min:** lift 4.77 (A) and 6.51 (B); one per cluster per day: 3.59 and 6.16. Prior 1 h: 4.77 / 5.96, deduped 3.71 / 4.79. Winner coverage 46% / 53%.
  - My wider operator test (any funder whose funded buyers sit mostly in this coin) agrees: winners' net buy is a median 96% outside, while half B controls are 67% farm bought.
  - **Limit:** "outside" = not traceable to the operator by funder. An operator buying from exchange or bot-platform funded wallets would look outside. E-AN4-1 in forward paper is the real test.
  - So this is our edge, not a First Window speed edge. It still lives in the first hour (91% of v4 legs).
- **Feed for A1 Q5 (HOLD MORE), hypothesis only, cuts picked after looking:** among winners, a leg that starts at ≤ 50% of the high so far with prior 15 min net buy < 15% of cap reaches 20x+ in 13.2% (A) and 17.6% (B), vs 10.4% and 6.6% for other winners. **v4 only:** 12.9% vs 2.8%. v3: no help (16.5% vs 21.6%). Worth a prereg test on v4 holds, not a rule.
- v3 rows (same file, new columns op2_*, out2_*, disp_*): http://159.223.88.168/hx-c9ois/dc/dc-an4-1-s2rows.csv.gz, script http://159.223.88.168/hx-c9ois/dc/dc-an4-1-op2.py. Cost: 0 X credits, $0 API, container 1 min, 0.15 h.

## Stage 2 (v2, 11:37; first stamped 11:45, a typed time, fixed from the real clock)
- **v2, 7 Oct 11:37 Bangkok: stage 2 added (who owns the money, who can pull). v4 only, 21 Sep to 5 Oct (PP wash.db per-wallet swaps): 370 winner legs, 859 control points.**
- **NEW, strongest fact, lever 2 (operator record), tested OUT OF TIME:** funder roots (crew.db) with 5+ coins and an honest 5x rate of 25%+ in half A: **6 roots. Their half B coins: 32.8% honest 5x (106 of 323) vs 2.2% for every other v4 coin. Lift about 15x.**
  - The 6 roots: 0x28bd039474d6942d8ad4ec07d57a0e98a1c58e45 (hook 12193_d2d5b6ba, 51 of 75 coins), 0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92 (67 of 181), 0x4757fc41278bd0f1b028c6dae71d381dc4558d98 (57 of 208), 0x2c2572da570646b402b9b9e625aa102803904a49, 0x37e51b3694e7cdc9a4f0c6ad4031e1b55eb94cf4, 0x847091a657ff98a37213246d343def3e3988945c. Also 0xe79ba928ecb418ab809c2bb5533506683a4b454f (13 of 18, too few in A).
  - **Catch:** 94% of their half B coins are pulled, and the median best $1k multiple is 2.55x. It is a serial pump operator whose pumps reach 5x BEFORE he pulls. Money only if the exit is fast. Overlaps queue item Q-OP (A2 + Data): handed to them.
  - Rule E-AN4-4 pre-registered for the Engine (below).
- **Whose money is the net buy? Mostly OUTSIDE wallets.** On winner legs with net buy ≥ 25% of cap in the prior 15 min, the median outside (non-operator) part is 96% of it.
  - Outside net buy ≥ 25% of cap in the prior 15 min to 1 h: lift 3.7 to 3.9 (A) and 4.4 to 4.5 (B); deduped 2.9 to 3.1 (A) and 3.3 to 3.6 (B). **Kept.**
  - Unique buyer wallets in the prior 15 min ≥ 44: lift 2.96 / 2.51 (deduped, ≥ 75: 2.11 / 2.41). **Kept.**
  - **Operator-funded buyers** (a funder that funded 3+ buyers of this coin and at least half of all buyers it ever funded): controls in half B are 67% operator-dominated (op share ≥ 80% of prior hour buys) vs 41% of winners; half A 29% vs 20%. Lift of "op share ≤ 20%" (71% vs 52% in A, 53% vs 27% in B) is 1.37 (A) and 1.96 (B): **fails half A, not kept.** In half B the fake-volume farm coins grew (Scout SC.1 factory clones).
  - Note vs SW A1.26 (v3, $10M vs dead $1M): there one-hand pre-$300k buying went WITH winners. Different question (which $1M coin reaches $10M). Here, among $100k coins at a low, the 5x ones had more outside buyers.
- **Who can pull at the leg start (A2 Q2 method on liqlock):** the creator holds the LP NFT on 87% of winner legs and 90% of control points. **No separation.** LP held by another wallet (not creator, not burned): 13.9% vs 5.6% (A), 3.6% vs 1.5% (B), lift 2.5 / 2.4 but only 24 and 7 winners: under the bar. Winners are pulled later anyway (82%).
- Stage 2 rows: http://159.223.88.168/hx-c9ois/dc/dc-an4-1-s2rows.csv.gz (md5 943b077dd66acfd7aafcf47de2d921b7), script http://159.223.88.168/hx-c9ois/dc/dc-an4-1-s2.py (md5 2990df378235ac952b7dd7e3b45ead12).
- Stage 2 cost: 0 X credits, $0 API, research-box read-only sqlite extracts about 5 min (nice 10, ionice best-effort 7), my container 1 min, 0.4 h.

## Stage 1 (v1, 11:25)
- **v1, 7 Oct 11:25 Bangkok. Stage 1 = tape, pool, family and operator history. Stage 2 (LP holder, operator bundle share, unique and new wallets) is running next.**
- **The big fact: an honest 5x on this chain is a LAUNCH PUMP, not a later run.**
  - 765 honest 5x coins (A1 universe, 15 Sep to 5 Oct). All 765 re-confirmed with a stricter entry (never buy below a bar's own low).
  - Where the 5x leg starts (the running low before the first 5x): **median 3.6 min after the first trade on v4, 10 min on v3**. 91% of v4 legs and 84% of v3 legs start inside the first hour. Median cap at the leg start: $90k.
  - v4 winners need a median **18 min** from leg start to 5x, and **82% of them are pulled later**. It is the spike before the pull (same as A2 Q2).
  - **Only 82 matched winners start 1 h or more after launch** (51 in A, 31 in B, about 4 a day, all venues). That is the whole Lane D prize in this data.
- **Venue is the biggest base rate:** honest 5x share is **v3 30.6%, Pons graduated 20.4%, v4 4.2%** of coins that reached a sellable $100k. 20x+ coins: 79, and 59 of them (75%) are v3.
- **Facts common in winners and rare in matched controls (both halves, lift = winner share ÷ control share):**

| # | Fact at the leg start (data before it only) | Lift A | Lift B | Winners with it A / B | Controls A / B |
|---|---|---|---|---|---|
| 1 | Net buy in the prior 15 min ≥ 25% of the cap | 3.76 | 5.84 | 38% / 51% | 10% / 9% |
| 2 | Net buy in the prior 1 h ≥ 25% of the cap | 3.53 | 5.87 | 35% / 55% | 10% / 9% |
| 3 | Buys ≥ 70% of the prior 1 h volume | 3.21 | 3.83 | 32% / 24% | 10% / 6% |
| 4 | Net buy prior 1 h ≥ $13.5k | 2.74 | 2.77 | 69% / 64% | 25% / 23% |
| 5 | Pool depth R ≥ $20k | 2.59 | 2.64 | 65% / 54% | 25% / 21% |
| 6 | Creator's first coin (no earlier coin in the universe) | 1.87 | 1.66 | 66% / 57% | 35% / 34% |
| 7 | Price ≤ 88% of the sellable high so far (a real dip) | 1.63 | 1.68 | 81% / 76% | 49% / 46% |

  - **One winner per cluster (funder root, else creator) per day:** facts 1, 2, 4 and depth keep lift 1.8 to 4.7 in both halves. Facts 3, 6 and 7 fall below the bar there, so they are partly one operator repeating.
  - **Late legs only (1 h or more after launch):** one fact holds: **net buy in the prior 6 h ≥ 25% of the cap**, lift 2.79 (A) and 3.46 (B), but only 15 and 11 winners.
  - **v3 alone (deduped):** heavy prior-hour flow wins: buys ≥ $254k in the prior hour, lift 2.42 / 2.25.
  - **Hook families (v4, under the 20% coverage bar, so not kept):** 12193_d2d5b6ba lift 5.3 / 11.6 (81 winners vs 27 controls); bundle 341_6de2bb44 lift 10.7 / 5.2. Both are pump-and-pull families: their 5x happens before the pull.
- **20x+ winners look different from 5x pumps:** they start from a DEEPER dip (median 45% of the high so far vs 59% for 5x-only and 90% for controls) and with LESS prior net buy (15 min: 8% of cap vs 24%). The big runners are bought on a hard dip, not on a flow surge.
- **Dead (no lift in both halves):** had $1M before, quiet hour, creator or root past wins, root past pull share, minutes since the high.
- **Label check (is "deep pool" just the $1k impact in the label?) No:** only 1.8% of controls ever reach a GROSS 5x from their point (no fees, no impact), in every depth band. The label is not impact driven.

## So what (first principles)
- **Lever 2 and 3 (operator intent, money in):** a 5x leg starts when **real money is already in the pool** (net buy ≥ 25% of the cap in the last 15 min to 1 h) and the price still sits at a dip. At the same cap and age, controls had a quarter of that money in. Price on a curve only moves if net money moves; the winners had the money.
- **But whose money?** SW A1.26 says pre-$300k buyers are mostly the operator's own disperse-funded wallets. So fact 1 may be "the operator is still buying" (lever 2), not outside demand (lever 3). **Stage 2 splits the net buy into operator-funded vs outside wallets.** That split decides if this is a deep-check edge or a First Window speed edge.
- **For Lane D:** honest 5x legs that start after 1 h are rare (about 4 a day). A deep check that takes minutes mostly misses the money. The money is in the first hour, and 82% of those pools are pulled after the top.

## Handed to the Engine (proposed rules; prereg lines appended to dc-prereg.md before any test)
- **E-AN4-1 (fast lane):** coin age ≤ 1 h, has printed a sellable $100k, price ≤ 0.88 × sellable high so far, net buy in the last 15 min ≥ 0.25 × cap. Buy $100 at the next trade. Exit per rulers (3x, 48 h, pull = 0, $100 at depth). Lever 2/3.
- **E-AN4-2 (Lane D):** coin age ≥ 1 h, not pulled, sellable $100k printed, net buy in the last 6 h ≥ 0.25 × cap, price ≤ 0.88 × sellable high so far. Same exit. Lever 3.
- **E-AN4-4 (operator record, v2):** v4 coin whose funder root (crew.db launch.root) had 5+ coins launched in the prior 14 days with an honest 5x share of 25%+ (labels known before the new coin's first trade). Buy $100 at the first sellable $100k print. Exit: sell at 3x, or the first trade after any LP removal by the creator, or 48 h; pull = 0. Lever 2.
- **E-AN4-3 (v3):** v3 or Pons graduated pool, age ≤ 1 h, buys ≥ $250k in the last hour, price ≤ 0.88 × sellable high. Same exit. Overlaps the Second Window v3 lane: report overlap.

## Method
- **Winners:** A1 universe honest5x = 1 (A1 events; sellable = a $100+ sell). Leg start = running lowest $1k buy cost (impact at depth R, never below the bar's own low) at the first sellable event where that buy nets ≥ 5x after 1% fees and impact both ways. 20x flag = best $1k multiple ≥ 20.
- **Controls:** honest5x = 0 coins. Candidate points = events after the first sellable $100k, before any pull, within 7 days, whose buy cost is not beaten in the next 24 h (a hindsight low, the same selection as the winner's leg start). Up to 3 control coins per winner, one point each, never reused, seed 7, matched on venue group (v4 | v3 + Pons), age bucket (log2 hours) and cap bucket (log2 $). 578 of 765 winners got 1 to 3 controls (1,679 control rows); 187 winners had no control in their bucket (123 v3: v3 has few non-winners) and are left out of the lift.
- **Features:** only events before the point. Creator and root history count only coins launched before the point, and a past win only if its 5x happened before the point (fixed in v1 before any read: the first draft counted later wins).
- **Cut:** one cut per feature chosen on half A at the control 10/25/50/75/90% points, held fixed for half B. Kept: lift ≥ 1.5 in both halves, winner coverage ≥ 20%, ≥ 10 winners per half (prereg line "7 Oct 11:14 | A4 | AN4.1").
- **Halves:** by leg start time. A = 15 to 28 Sep, B = 29 Sep on.

## Limits
- Hindsight-low selection uses future data for SELECTION only. The facts are hypotheses until the Engine tests them as real-time rules.
- No public router filter (PP DCPP.3). v4 pulls after 5 Oct 17:39 unseen (liqlock copy). v3 roots: none (crew.db is v4).
- Depth R on v4 comes from price moves (median of 7), so fact 5 is noisy. Fact 1 uses real quote amounts.
- 28 to 31 features tried: a 1.5 bar in both halves on 600 rows can still let one lucky feature through. The Engine test is the real filter.

## Files
- Rows (one per winner leg and control point, all features, full addresses): http://159.223.88.168/hx-c9ois/dc/dc-an4-1-rows.csv.gz (md5 afcfcbb116154e08380b0f8ba74163bd)
- Winner legs (pool, leg start, first 5x, best multiple): http://159.223.88.168/hx-c9ois/dc/dc-an4-1-legs.csv (md5 8e4e3bed2aa4d9516697733bf8017aaa)
- Scripts: http://159.223.88.168/hx-c9ois/dc/dc-an4-1.py (md5 59fb1be5f6be0f6870981ad3eb87cb6a), http://159.223.88.168/hx-c9ois/dc/dc-an4-1-lift.py (md5 dab9ad6d41475836ba6e4461a1804eb8)

## Cost
- 0 X credits, $0 API, research-box: two gzip copies (nice 19, ionice idle, about 4 min), my container about 2 CPU minutes, 0.4 h.

First principles score: 85%. Every number is from each pool's own swaps, depth and pulls, with a stricter entry than A1. Stage 2 added who owns the money and who holds the LP (v4). Missing: public router sells, v3 per-wallet data (v3hist has no signer), X authors.

Version: v3 (operator vs outside split by disperse, creator, root; Q5 feed), 7 Oct 2026 11:41 Bangkok, DC Analyst 4. Latest: http://159.223.88.168/hx-c9ois/dc/dc-an4-1-anatomy.md