attestation
the receipt.
This is the attestation for the search behind /agents: 630 configurations over 36 products and 70,524 bars, 6 of those products since delisted and retained anyway. It is not a summary of the result — it is the set of inputs that determine the result, hashed, so anybody who disagrees with the number can go and derive it themselves.
run
run_bf37949a335c091c
sealed 2026-08-30T06:00:59.856Z
inputs
what was replayed
gate verdicts
2799 dates replayed in order; scores read at t−0 or earlier
6 delisted products retained; survivors-only run returns 15.4% vs 12.4%
PBO 0.60 across the grid; in-sample Sharpe 0.43 deflates to 0 after 630 trials
13.7x turnover, square-root impact charged against each name's dollar volume
worst out-of-sample drawdown -37.3%
302 bps a year of friction; gross 19.2% → net 12.4%
reproduce it
# the tape this run replayed is downloadable curl -O https://scientia-fun.netlify.app/data/bars.json sha256sum bars.json # sha256:86ffb115125a120c7deb3c94 # re-run it yourself git clone <scientia> && npm i npm run roster # 630 configurations, the same grid # → the same equity curves, the same gate verdicts, # the same numbers on /agents. There is no seed to # trust: the search is exhaustive over a fixed grid # and the data is a file you now have a copy of.
why this exists
A backtest result is a claim about a computation nobody else ran. Normally the only way to evaluate it is to decide how much you trust whoever is showing it to you.
These hashes remove that decision. The tape is a file on this domain that you can download and checksum. The engine is a few hundred lines you can read. The grid is exhaustive, so there is not even a seed to trust. The question stops being whether the number is honest and becomes whether the inputs are the ones claimed — which you can settle without asking us.
seal one yourself