We publish the strategies that failed

Ask what a verification company is incentivized to hide, and you have found the test of whether it can be trusted. Every gate that judges strategies produces failures; the only question is whether the issuer buries them. We do the opposite. A failing verdict at MIZAN is not an error to be quietly discarded. It is a real proof — minted, downloadable, and anchored — and arguably the most important kind of credential we issue.

01The first GPU mint was a failure, and we published it

On 18 July 2026 we ran our first credential on production GPU proving infrastructure. The strategy did not clear the gate. Its worst single bar breached the −12% floor. In most companies that result never leaves the building. Ours is in the public registry, with a real STARK behind it:

Update, 28 July 2026: the ledger grew its best entry yet — the first proof ever minted on our licensed survivorship-free US equity substrate is also a FAIL: our own trend engine, on the S&P 500, refused by our own gate (Sharpe 0.41 vs the 1.20 bar). The full receipt — downloadable bundle, hash, Bitcoin timestamp — has its own note.

The proof is exactly as rigorous as a passing one. The engine that judged it is the same locked engine — image id dafad185… — that judged each of these (v11 is now the current engine; v10 credentials remain checkable forever). The only difference is the verdict, and we do not treat the verdict as ours to curate.

Anyone can show you their winners. The number that tells you whether a verifier is honest is how readily it shows you the losers.

02Not every honest number is a failure — some are just "not significant"

Failure is the loud case; there is a quieter, equally important one. Our own flagship BTC trend book, run through the committed-trial Deflated Sharpe, comes back NOT significant (DSR 0.6779). We publish that too, unedited. A strategy that does not clear its own deflation is not dressed up as though it did; the honest number stands in the registry beside the strategies that genuinely survived their search.

03We hold our own research to the same standard

This discipline did not start with the product; it comes from how we run our own systematic research. When our internal engines produced results that later proved to rest on subtle errors, a phantom-fill assumption in a stop model, a same-month signal that quietly used information from the future, a universe that survived only because losers had been dropped from history — we did not paper over them. We retracted them, publicly, with the post-mortem attached.

These are not embarrassments we were forced to admit; they are, by our lights, the most valuable outputs of a research process, the errors caught before capital met them. A firm that cannot publish its own retractions has no standing to certify anyone else's honesty. The gate we enforce on others is the gate we failed, in public, first.

04Why the failures are load-bearing

A credential means something only if the same process could have produced a different answer. If MIZAN passed everything, a passing credential would carry no information, it would be a receipt for having shown up. The failures are what give the passes their value. Every published FAIL, every honest "not significant," is evidence that the gate is real, locked, symmetric, and beyond the issuer's power to bend. That evidence is not a byproduct of the standard. It is the standard's proof of life.