HFT Algo Scanner: Clean-Room Replica of a Locked Option-Flow Indicator

A client paid for an invite-only "HFT Algo Scanner" and wanted to understand it. The source is locked, so the method was to treat the original as an oracle: read its plotted outputs, labels and table text over the DevTools Protocol, vary its inputs programmatically (ATM offset across thirteen values, thresholds, lookback) and observe what changed, then write an open Pine v6 script that reproduces every output.

The decoded logic: ten calls above and ten puts below the at-the-money strike on 1-minute data; flow is volume times mid-price, counted only when the option moves against the index candle; flow is banked per side and released when a Supertrend with specific parameters on the index agrees; chart bars sum the 1-minute releases; strength is a z-score of put writing minus call writing with a window that depends on timeframe. The verification table shows 2,093 of 2,093 one-minute bars and all labels over six days, and 120 of 120 three-minute bars on the latest day, with older days at 87 to 95 percent attributed to thin strikes. Dynamic ATM, further expiries and support lines were explicitly not replicated.

The reusable lesson is a discipline: accept a rule only at zero misses, because every near miss (93 to 97 percent) turned out to be a wrong detail such as a factor of 0.60 versus 0.61 or a mean not subtracted.

How it works

1. Dump the original's plots, labels and table over CDP
2. Vary inputs on the original and record output deltas
3. Hypothesise a rule and implement it in open Pine
4. Diff replica vs original bar by bar; iterate until zero misses
5. Publish the replica with a verification table and known gaps

Screens