Videos already examined, so finished work is not re-requested. Depth is the point: read means line by line, extracted means it went through the mechanism extractor and the quality gate, swept means it was counted in a batch — touched, not examined. Asking for a proper read of a swept video is still reasonable.
Killed strategies are conditional rejections, not permanent ones — they were judged against what was known at the time. A retest needs named new information, the same reason cannot be used twice, and it registers a new strategy id so its trials add to the family count and raise the bar for everything else. Retesting is not free, and it is not supposed to be.
Every feed this project is connected to, when it last answered, and what it has actually produced. A source with no yield is not a failure — several are new — but it has to be visible as such rather than blending in with the ones doing work.
Every signal committed to the append-only prediction logs, newest first — across all strategies, not just today. Each row carries the rule that produced it and that rule’s current registry standing. Withdrawn means the strategy was later invalidated: the breakout-retest family selects its events on the future, and rebuilt walk-forward it has no edge. Those rows are kept because their paper fills still measure execution cost, which is the open question on momentum — they are not recommendations. The macro and cross-sectional research rigs log predictions here too; they have no registered trial, so they are hidden by default rather than shown as if they were trade intent.
Click any row for what fired, the price it was referenced to, and when it exits. Ranked on is the value each rule sorts by. It means different things per rule and is labelled per row — momentum shows a 12-month return, the retest rules a count of zone touches — so the column is not comparable across strategies and sorting it across all of them is meaningless.
| For bar | Ticker | Strategy | Standing | Ranked on | Reference px | Exits after |
|---|
The breakout-retest scan for today, kept for continuity. This family is withdrawn; the rows below are not signals to act on.
Positions inside the 63-bar hold. “Entered at” is the close that confirmed the hold; “Now” is the latest close. Change is raw between them, not benchmark-adjusted — the strategy's claim is excess over SPY, so read a green number against what the index did over the same span.
| Ticker | Entered | Held | Left | Touches | Zone | Entered at | Now | Change |
|---|
| Bought | Ticker | Rule | Shares | Entry | Now | Cost | Value | P&L | % | vs SPY | Exits | Bars left | Source |
|---|
These are BACKTEST signals, not a trading record. Every trigger the rule would have fired across the scan window, with the outcome where the 63-bar hold has completed. The live forward record began 2026-08-27 and is a few dozen rows — it is on the Triggers and Open book tabs. Reading this table as a track record is the mistake it is labelled to prevent.
| Triggered | Ticker | Rule | Touches | Zone | Entry | Pullback | Status | Return | Exited |
|---|
You are looking at a subset. A win rate computed after filtering is not evidence about the rule — picking the slice with the best number is the selection this project spent its trials rejecting, and it is how the breakout-retest family produced 4.5 points of edge that did not exist. Use the filters to understand what the rule did, not to find a version of it that worked.
Every methodology this project has catalogued, and what happened to it. A strategy nobody wrote down as untried is indistinguishable from one that failed — so untested is a recorded state, not an absence.
Formally registered trials. Each had its pass bar written down before the data was fetched, and the verdict is computed against that bar rather than chosen afterwards.