7. Methodology

In plain terms: before any rule was allowed to accuse a token of anything, it was run against the real chain to see who it would accuse — and twice, the first version accused the innocent. Those two corrections are documented below in full, because the corrections ARE the method: a rule that fires on healthy tokens is worse than no rule.

The Bundle Rule

FIRST ATTEMPT — TIMING ONLY: "N WALLETS ENTERING WITHIN THE LAUNCH WINDOW."
MEASURED ON 2,432 FULLY-SWEPT TOKENS: 27% HAD >= 5 DISTINCT BUYERS WITHIN
100 BLOCKS (~10 SECONDS). INDEPENDENT SNIPER BOTS WATCH TokenCreated AND
PILE IN ORGANICALLY. TIMING ALONE WOULD HAVE ACCUSED A QUARTER OF THE CHAIN.
REJECTED.

SHIPPED RULE — FUNDING CORRELATION: ORGANIC SNIPERS FUND THEMSELVES
INDEPENDENTLY; A BUNDLE IS ONE OPERATOR FUNDING N WALLETS. THE RULE REQUIRES
>= 3 SAME-BURST WALLETS (<= 20 BLOCKS FROM LAUNCH) SHARING A FIRST FUNDER.

SECOND CORRECTION — THE INFRASTRUCTURE FUNDER: ONE UNLABELED EOA TURNED OUT
TO BE THE FIRST FUNDER OF 11.5% OF ALL TRACED SNIPER WALLETS ACROSS
UNRELATED TOKENS — ON-RAMP-SCALE BEHAVIOR, NOT AN OPERATOR. FUNDERS WITH
>= 30 TRACED FUNDED WALLETS ARE DISQUALIFIED FROM SUPPORTING A BUNDLE CLAIM.
THIS IS DELIBERATELY CONSERVATIVE: IT CAN ALSO EXCLUDE A VERY-HIGH-VOLUME
BUNDLE OPERATOR. A MISS IS ACCEPTED OVER A FALSE ACCUSATION.

RESULT: THE SHIPPED RULE FIRES ON ~0.5% OF TOKENS. ONE CLUSTER WAS VERIFIED
BY HAND TO THE FUNDING-TRANSACTION LEVEL BEFORE THE SEVERITY SHIPPED:
OPERATOR FUNDS WALLET A; WALLET A FUNDS WALLETS B AND C; ALL THREE ENTER THE
SAME TOKEN WITHIN TWO SECONDS OF LAUNCH; ALL THREE LATER EXIT TO ZERO.

The LP Rule

FIRST ATTEMPT: "BURN ADDRESSES HOLD < 99.9% OF LP SUPPLY" — FLAGGED 3 PAIRS
CRITICAL. INSPECTION OF THE EVIDENCE SHOWED ALL 3 HAD THE FULL GRADUATION
MINT BURNED, PLUS LIQUIDITY ADDED LATER BY THIRD PARTIES WHO HELD THEIR OWN
LP. THE SENTENCE WAS TRUE; THE SEVERITY WAS A FALSE ALARM.

SHIPPED RULE: THE GRADUATION MINT IS DERIVED FROM CHAIN-EMITTED AMOUNTS
(isqrt(LaunchedToDEX.amount x LaunchedToDEX.eth)) AND COMPARED AGAINST THE
BURN-ADDRESS BALANCE. SHORTFALL AGAINST THE GRADUATION MINT => CRITICAL
(CURRENTLY ZERO ON THE ENTIRE CHAIN). POST-GRADUATION ADDITIONS => NOTE.

THE STANDARD, STATED ONCE: CRYING WOLF ON HEALTHY PAIRS WOULD HAVE BEEN THE
PRODUCT'S FIRST LIE.

Holder Balances

BALANCES ARE EVENT-SOURCED FROM EACH TOKEN'S OWN Transfer LOGS — A GLOBAL
WINDOWED SWEEP WITH AN ADDRESS FILTER OF EVERY INDEXED TOKEN. NO THIRD-PARTY
HOLDER API IS USED. CORRECTNESS CHECK: TOP-HOLDER BALANCES WERE COMPARED
AGAINST DIRECT balanceOf READS PINNED AT THE SWEEP BLOCK AND MATCHED
EXACTLY.

What Regenerates When

EVENT INDEX: EVERY MINUTE (CRON), CURSOR-GUARDED, EXACTLY-ONCE.
STATE SNAPSHOTS: NEW AND RECENTLY-TRADED TOKENS EACH RUN; EVERY VALUE
  CARRIES ITS READ BLOCK.
HOLDER SWEEP: ADVANCES TOWARD THE EVENT CURSOR EACH RUN.
LAUNCH-COHORT (BUNDLE) ANALYSIS: ONCE PER TOKEN, AFTER ITS LAUNCH WINDOW IS
  FULLY SWEPT.
FINDINGS: REGENERATED FROM CURRENT DATA; A FINDING WHOSE EVIDENCE NO LONGER
  HOLDS IS DELETED, NOT KEPT FOR HISTORY.
GRADES: NEVER STORED — COMPUTED AT RENDER TIME FROM THE ROWS ON THE PAGE.