I measured an 87% edge on my own trading. It was an artifact.
This page is the whole record: what I claimed, how I measured it, what was wrong with the measurement, what I did about it, and what I claim now. What I claim now is nothing.
What I claimed
In early August 2026 my closing-line-value engine reported that 87% of my positive-CLV trades beat the close, and a related cut of the same data put it at 90.5%. For a prediction-market trader those are absurd numbers. I believed them for about a day.
How it was measured
CLV compares the price you paid against the market's closing price. The engine took a snapshot of each market's book and, at settlement, compared my entry against that snapshot. The intent was to ask: did the market move toward me after I got in?
What was wrong
The closing snapshot was being captured after the market resolved. On Kalshi a resolved market sits at 0 or 1. So for a large share of rows, "closing price" was not a price at all, it was the answer. CLV computed against it did not predict the outcome, it restated the outcome. About three quarters of snapshots sat at ≥0.95 or ≤0.05, and the sign of CLV matched the result 89.3% of the time, which is the signature of a leak rather than an edge.
Nobody caught this because the instrument agreed with what I wanted to see. That is the actual lesson, and it is why this page exists.
What I did
Every affected row was quarantined rather than deleted, so the error stays inspectable. The method was rewritten to capture the closing line before the event begins, and only rows scored that way count toward anything.
| Observations | Count |
|---|---|
| Clean — closing line captured pre-event | 4,376 |
| Not yet scored (in flight) | 2,389 |
| pm_pre_event_v1 | 680 |
| No usable snapshot | 595 |
| Quarantined — closing line captured after resolution | 527 |
| Total observations | 8,567 |
What the clean data says
The honest denominator is settled trades scored by the fixed method, in the current strategy epoch. Both exclusions are named below rather than folded away, because reporting a figure computed on a superset of the clean set is the exact error above.
| Trades | Count |
|---|---|
| Clean settled, epoch v2.2 — prior epoch, excluded from headline | 8 |
| Clean settled, epoch v3.0 | 1,251 |
| Clean settled, epoch v3.0-inv — prior epoch, excluded from headline | 4 |
| Clean settled, all epochs | 1,263 |
| Excluded — retired method | 116 |
| Same epoch, all methods — NOT the headline | 1,362 |
Negative. On the metric this whole thing exists to measure, my own trading is behind. I would rather publish that with the method attached than publish a good number you have no way to check.
What I claim now
No edge. Not from this data, not at this sample size. When there is enough clean data to say something, it will say it here, with the method next to it.
If you trade Kalshi
How do you know if you got a good fill? Genuinely asking, and I would rather hear it in your words than guess.
Send me your Kalshi fills export and I will grade it against pre-event closing lines and write you back what it found. Free, and I am not going to pitch you anything.
First 5 exports, up to 250 fills each, so the offer stays real.
Export from Kalshi: Account → Settings → Export fills (CSV).
What happens to it: I score it, I write back, and I delete the file when I do. It is
not added to any dataset and it is not shared. Ask me to delete it sooner and I will.
Combined and multi-leg positions are excluded from grading. A combo is one instrument
with one price; grading its legs separately and adding them up produces a number that
is not that position's CLV.
pre_event_v2). Rows scored by
the retired method (leaked_v1), rows with no usable snapshot, and rows from
prior strategy epochs are excluded, and each is counted above.Sample size. Stated with every figure on this page. It is small.
Past performance is not indicative of future results.
Not an offer or solicitation. Nothing here is investment advice, and no fund, pool, or managed account is offered. Today Sayf trades its own capital only. The Edge plan described on the pricing page would change that. It is not built, nothing is for sale, and the terms and risk disclosures get written before anyone can buy.
Figures as of 2026-09-23 16:04 UTC. First 5 exports, up to 250 fills each.