Research

measured, not predicted

Locard has not published a prediction yet. This page shows what the desk has measured so far: the graduations it observes as they happen, and a backtest that replays past graduations through the same six traces and settles them against what really happened next. None of it is a Locard prediction, and none of it is in the ledger.

01 · on the desk

Graduations observed

Observed since the desk started

430

first one 2026-10-15 01:33 UTC

In the last 24 hours

430

pump.fun tokens that reached PumpSwap

Token	Graduated (UTC)
$fg FU4sR…wpump	10-15 14:22
$Stank! 578wi…opump	10-15 14:22
$LEEKDUCK A3PmW…qpump	10-15 14:21
$Dacity GDbNW…npump	10-15 14:18
$YouTube H72Cc…ypump	10-15 14:16
$SATOSHI HuC1W…Kpump	10-15 14:15
$wire Cic4c…Lpump	10-15 14:13
$H2O 6M846…Apump	10-15 14:11
$ch HXNPk…Lpump	10-15 14:09
$AUTCRAFT 8XVZU…6pump	10-15 14:08
$BLUBBER 9sfAt…Fpump	10-15 14:06
$CRS 7cNGy…bpump	10-15 14:00
$KnightDog ocjMS…5pump	10-15 13:59
$XMART HdTAR…3pump	10-15 13:55
$FEEMAIL ApzQz…Upump	10-15 13:49
$SOLCAT 2QNiZ…Fpump	10-15 13:49
$CHUCKIE G JAQMh…dpump	10-15 13:48
$LEGO s8ZZy…Tpump	10-15 13:44
$SOS oVzja…mpump	10-15 13:43
$TESLA SI AnX5p…Spump	10-15 13:42

02 · backtest

What happened to past graduates

603 past graduations from 2026-09-28 11:58 UTC to 2026-10-13 18:49 UTC. Each was examined as of 10 minutes after graduation, seeing nothing later, and then settled from the real next 24h. These are research rows: they carry no probability and are not Locard's record.

Fell 70% or more within 24h

65%

391 of 603 tokens

Does the trace score tell them apart? (AUC)

0.51

0.50 means no better than a coin flip

Trace score	Tokens	Fell 70%+	Share
0.0 – 0.2	180	127	71%
0.2 – 0.3	149	98	66%
0.3 – 0.4	204	114	56%
0.4 – 1.0	70	52	74%

How to read it: so far the six-trace score does not separate the tokens that fell from the ones that held. 65% of all these tokens fell whatever their score, so a probability built on this score would add little. Locard will not seal a protocol on it until that changes.

The wallet-cluster claim could only be checked on 101 of these tokens (10 showed a cluster selling half its position); the rest had no cluster to track. With samples this size the rates carry wide error bars. How calibration works.

03 · the typical graduate

What a graduate usually does in its first day

The price of every backtested token over the 24 hours after the cutoff, as a multiple of its price at the cutoff. The line is the median token; the shaded bands hold the middle 50% and the middle 80%. It is a pattern in past graduates, not a prediction for any one token.