Field clips — Michael Kaminski
One finding, ten seconds. Each clip compresses a single result from a system he runs; the full arithmetic is in the write-up underneath it.
The gate that stops nothing
A statistical gate on an agent's instruction file, and the power calculation that shows the 5% threshold written next to it never binds. At this variance a 14-day window catches a true 5% lift 44% of the time — a coin flip on your own best edits.
The gate is Welch's t-test on the agent's instruction file: p < 0.10, a 5% minimum lift, 14-day windows. It looked like a control until the power calculation was run on the gate itself.
At the observed variance a 14-day window can only detect an 8.3% effect. The 5% threshold sitting next to it in the config is inside the blind spot the whole time — it never binds, and it never did any work. Run it forward and an edit that truly delivers 5% clears the gate 44% of the time.
The barrier in the clip is closed the entire time and everything drifts straight through it. That is the finding as a single image.
Transcript
I gated my agent's instructions behind a statistical test. Then I ran the math on my own gate. It's a coin flip.
Zero trades, and the level still moved
A dealer-gamma pipeline walked the QQQ gamma flip 3.9 points and dropped the put wall ten dollars across eight consecutive writes with spot pinned at 717.12. Nothing traded. An overwritten file has no history to argue with — one append-only CSV is what caught it.
The pipeline recomputes call wall, put wall, and gamma flip off the live option chain every fifteen minutes, 33 runs a day across seven underlyings, and writes a thinkScript study the chart reloads. The first version only did that one write: compute, overwrite, reload.
It looked correct for weeks, because an overwritten file has no history to argue with. The fix was not a fix — it was one more write, a per-symbol append-only CSV next to the study. Eight consecutive rows later: spot constant at 717.12, flip 718.3 → 719.5 → 720.0 → 722.2, put wall 710 → 700, net GEX 598 → 626 → 611. Market closed. The chain re-fit and the levels walked on their own.
That matters because the study fires an alert on price crossing the flip. When the flip is a re-estimate and price is static, the alert is firing on its own arithmetic — and for weeks the only artifact was a file that overwrote its own evidence, so none of it was reconstructable after the fact.
Transcript
QQQ hadn’t traded a share. Walked the gamma flip 3.9 points. Overwrite each run and the file has no history to argue with — so I appended.