The promise
We grade stored dealer levels against price in the next session with available scoring data. We keep the measurements alongside the grades and publish the rate with its sample size. The record below distinguishes covered close snapshots from historical backtests and from delivery of individual public cards.
What gets published
Twice a market day, for the most options-active names:
- The call wall. Where dealer hedging tends to defend a move up.
- The put wall. Where it tends to defend a move down.
- The gamma flip. Where hedging switches from pinning a range to chasing the trend.
- Heat. A one-number read on how active a name is right now.
- Plays at the walls. The covered-call and cash-secured-put strikes worth selling, sized to the levels.
How accuracy gets measured
The morning after each close, an automated pass reads yesterday's walls and asks one question of each: did price stay near it, move away from it, blow through it, or never test it? Those are the four grades, pinned, rejected, broke, and untested. Each is a fixed-threshold math question, applied the same way to every name. No human judgment touches the call.
What we publish about ourselves
The homepage rate covers classified call and put walls from close snapshots across the covered universe, from the first scored session through the displayed date. It includes names beyond the featured cards. Historical backfill snapshots and gamma flips are excluded. Held means pinned or rejected; the denominator is held plus broke. The homepage rate appears at n=30 tested walls, with a Wilson 95% interval. This describes wall outcomes, not trading returns.
We publish a rate, not a dollar P&L. The rate is our honest claim; the trade sized to it is yours. The denominator is tested walls only (held + broke), so untested walls never pad it. Per-ticker reliability shows on each name's page once it has enough scored walls to mean something, and the Friday card posts the week's covered-universe grades to Discord. Each surface states its own period; a ticker's rolling rate and the cumulative universe rate need not be the same.
Each scored level also carries a hold-rate: how often a wall there has held when tested, measured across the past year of sessions. It's the base rate behind the level, a deeper read than the running tally above.
Recovered closes and publication
A missed close snapshot can be rebuilt later from end-of-day data. Those recovered closes are included in the close-snapshot record. Inclusion does not establish that a level was publicly visible before its outcome. The archived featured-selection flag is reconstructed during scoring and is not a verified record of delivered cards. We do not describe its subset rate as the accuracy of all posts customers received.
Historical archive and model evaluation
The larger scored-wall archive also includes historical backtests. Its count is separate from the close-snapshot count. The breach-model evaluation shown on the Proof page is an out-of-sample historical backtest; it is not a record of predictions saved before their outcomes. Historical base rates, model probabilities, and observed close-snapshot rates answer different questions and should be read with their stated horizons.
The fine print
A few conventions are baked into the numbers. None is a mistake; they are the standard lane, worth knowing if you cross-check against another source.
- Dealer positioning is a model, not a measurement. The gamma figures assume the standard convention that dealers are long the call side and short the put side. Real books carry both and are not observable from outside a firm. Read the walls and the flip as a probability map, never a claim about who is literally long or short.
- Horizons. Unusual and Premium rows carry a horizon badge by days-to-expiry: weekly (1 to 7), monthly (8 to 45, the wheel range), and LEAPS (46+). 0DTE is filtered out. The badge helps you self-filter; it never changes the ranking.
- Confluence. When a wall sits within 0.5% of a standard Fibonacci retracement of the trailing ~60-session swing, the row says so. It is a credibility multiplier on the wall, not a level of its own.
- ATR is a simple average. We use the arithmetic mean of the trailing 14 true ranges, not Wilder's smoothing. More reproducible and easier to audit; expect a few cents of difference from a charting platform.
Why this matters
A rate is useful only alongside its population, sample size, period, and scoring rule. Walls from the same session or ticker can be correlated, so a large row count alone does not establish independent evidence. We publish these distinctions so you can assess what the record supports.
