Does gamma state predict the size of tomorrow's move?
Why it was rejectedThe common claim: above the Gamma Flip, dealers absorb volatility, so tomorrow's move is smaller. We tested it before building a screen on it: end-of-day state against the next day's move, 404 pairs across 34 tickers and 12 days. The result: the absolute move was 1.99% below the flip versus 2.26% above it. So the signal is reversed, and it is not statistically significant anyway (t = −0.78).
What it cost usWe don't sell a promise that the state predicts the size of tomorrow's move. We show the flip as a verifiable level, not a prophecy.
Predicting direction from net exposure
Why it was rejectedSame sample: does positive or negative exposure predict where price goes tomorrow? The difference in directional return is t = +0.34· nothing. In the wall test itself, the wall held to the next session close in 83.4% of 22,113 cases, and a random line at the same distance would have held in 83.1%. The difference is negligible: distance explains the hold, not the wall. Over only 12 days the difference seemed to favor the wall, then it faded as the sample grew.
What it cost usWe show the wall as a map of where dealers hedge now, not a barrier that stops price, and we show no direction. The breakdown by distance is in the methodology.
Sweep/Floor order classification
Why we rejected itIt needs to know the executing exchange for each trade. The feed we read carries only price, size and time.
What it cost usCompeting platforms offer this feature, and dropping it weakens our feature list.
Splitting executions at the bid or ask
Why it was rejectedIt needs the quote at the exact moment of the trade, and that is not stored historically. We tried it and 100% of trades were tagged "above the ask" · a meaningless number.
What it cost usInstead we show "an estimate from price direction between trades (Tick Test)", call it an approximation, and keep an "undetermined" category visible without spreading it out.
Volatility rank before the window matures
Why it was rejectedA rank over ten sessions reads as "lowest of the year" when it is the lowest in two weeks. Example: SPY ranked 0 over 10 points · a trader reads it as a yearly low and buys contracts on a false picture.
What it cost usWe say "in progress: 13 of 60" instead of a number that looks finished.
"Winner/loser" for every contract
Why we rejected itWe don't know whether the contract was bought or sold. Whoever sold it gains when it falls, so calling the change a "gain" assigns the reader a position we don't know.
What it cost usWe named it "contract move since the trade": less exciting and more honest.
Placeholder prices when the feed is missing
Why it was rejectedThe ticker bar held fixed values shown when the live price did not arrive. The result: nine made-up prices shown to every visitor, one of them off the market by 40%.
What it cost usThe ticker is shorter now and carries only what we have, and an automated check blocks any hard-coded price from coming back.
A cherry-picked track record
Why it was rejectedProfit factor 1.24 and expectancy 3.04% over 215 closed trades. A record that gets filtered loses its meaning, and anyone reading it knows that.
What it cost usWe published it as is, with the full formula, on a public page.
3D volatility surface
Why we rejected itIt is read by rotating and flipping, not at a glance, and it contradicts what the platform is built on: a verdict in ten seconds.
What it cost usWe built a skew curve across seven expirations instead.
A filter called "high-probability setup"
Why we rejected itThe name promises three conditions, and only one is available: the trendline is not stored at all, and the signal candle is not output for each row.
What it cost usWe didn't build it. A name is a contract with the reader.
