THREE DEFECTS. THREE REPRODUCTIONS.
Make the failure
observable.
Change the input. Compare the behavior.
Inspect what the correction actually proves.
OBSERVED DEFECT / OUR RECORDER
Truthy is not
a wire format.
A Borsh boolean has two valid bytes: 00 and 01. Coercing every nonzero byte to true can turn corrupt input into a plausible BUY event.
The remaining bytes are a synthetic zero-filled 113-byte prefix. No token or account is queried.
Read the reproduction and fix ↗bool(payload[56])value in (0, 1)Malformed input must be rejected before it becomes an observation.
TRUST BOUNDARY / TRANSACTION LOGS
A mention
is not an emitter.
A transaction can invoke multiple programs. A matching event prefix says how bytes decode; the active invocation frame says which program emitted them.
PUMP and OTHER are symbolic labels. TRADE:1 stands for a valid event prefix. The experiment sends no transactions.
Read the attribution investigation ↗MODEL VALIDITY / EXECUTION ORDER
A stop price
is not a fill.
A price gap can cross a standing stop before the next bar begins. Filling the simulated exit at the stop erases a loss the observed opening price already implies.
Synthetic long, zero spread. If the open is above 95, this fixture assumes a later stop touch. R is return divided by initial risk.
Read the execution investigation ↗REPRODUCE IT YOURSELF
The source is
part of the proof.
One Python file. Standard library only. The same published fixture values drive the browser experiments and the independently runnable checks.
python reproduce.py6 TEST METHODS / 22 SHARED FIXTURE VECTORS