Concepts
Track records
How the stat block was measured and how to present it without misleading anyone.
A track record is the measured outcome of every detector-confirmed breakout of one kind, over twenty years of daily bars. It comes back on every setup (track_record.family and, when there is enough data, track_record.ticker), on /families and on /families/{family}/stats. The shape is the Track record object; this page explains how to read it and how it was measured.
{
"scope": "family",
"family": "falling_wedge",
"symbol": null,
"win_rate_pct": 54.0,
"sample_size": 1287,
"wins": 695,
"losses": 592,
"avg_win_pct": 8.1,
"avg_loss_pct": -4.2,
"median_win_pct": 6.3,
"avg_trade_pct": 2.4,
"avg_r": 0.41,
"expected_r": 0.41,
"profit_factor": 1.61,
"avg_hold_bars": 11.2,
"backtest_range": ["2006-09-25", "2026-09-25"],
"measured": "Every detector-confirmed breakout … open trades are excluded.",
"summary": "Falling wedge: 1,287 resolved trades from 2006-09-25 to 2026-09-25; 54% reached the target, average win +8.1%, average loss -4.2%, median win +6.3%."
}The rule
Never present win_rate_pct alone. A 54% win rate with an average loss of -4.2% and 1,287 trades is a real edge; the same 54% with an average loss of -12% would lose money, and the same 54% on 11 trades means nothing. The summary sentence carries all four numbers in one line, and it exists so a script or an agent can quote it whole instead of picking one figure. If you build your own sentence, keep the sample size, the win rate, the average win and the average loss together.
The numbers
| Field | Read it as |
|---|---|
sample_size, wins, losses | Resolved trades only. Open trades are excluded, so the counts never flatter a family with a recent run of entries. |
win_rate_pct | Trades that reached the target, as a percentage of sample_size. |
avg_win_pct, avg_loss_pct, median_win_pct | Percent move from entry to exit, averaged over the winners and the losers. The median win is there because a few large winners can pull the average up. |
avg_trade_pct | The average of every trade, winners and losers together; the expectancy in percent. |
avg_r, expected_r | The same expectancy in units of initial risk (entry to stop). Above zero is positive expectancy. |
profit_factor | Gross gains divided by gross losses. 1.0 is breakeven. |
avg_hold_bars | Sessions from entry to exit, on average. |
backtest_range | First and last trading session in the measurement. |
scope, family, symbol | family covers every ticker; ticker is the same measurement for one symbol and carries its symbol. |
How it was measured
measured states the rule in one paragraph, and it is the same for every family: every detector-confirmed breakout in the nightly backtest over the last twenty years of the served universe, one trade per pattern. Entry is the breakout bar's close. The stop is frozen at entry and fills on a touch, or at the open after an adverse gap; a bar that reaches both target and stop counts as a loss. Bullish converging patterns (pennants, triangles, squeezes, bullish wedges) hold to the measured-move target and then trail the 10-day moving average; bearish breaks exit at the target. Win rate is wins over resolved trades, and open trades are excluded. It is a mechanical record of what the rule did, not a claim about a strategy, which is why avg_loss_pct belongs beside the win rate every time.
Filtered records
A family record answers "how have falling wedges done?" A trader usually wants something narrower: "how did falling wedges do when the target was at least twice the risk?" A filtered record answers that. You pick the filters, and it counts only the past trades that matched them on the day they were entered.
It is the same answer the screener's Track record panel gives when you tick filters on the site, read from the same stored past trades. Any combination works, and each one is worked out when you ask.
Call GET /track-record, or the get_filter_track_record tool, with a family (leave it out for all five) and filters, the same filter list GET /scan takes. The answer puts two records side by side: track_record for the trades that matched, and unfiltered for the same family with no filters, so you can see what the filters changed. Both have the shape above, with a summary to quote whole.
Falling wedges with the Edge Filter on make a good example. Without filters there are 6,835 past trades, and about 43 out of 100 won. Asking for a target at least twice the risk leaves 1,997 trades, and about 37 out of 100 won. Fewer trades won, yet the average trade made more: +2.1% against +1.8%, because the winners ran further. That is why the record shows the average win and the average loss beside the win rate, never the win rate alone.
{
"family": "falling_wedge",
"edge": { "requested": "on", "applied": true },
"filters": {
"applied": [{ "kind": "min_rr", "op": ">=", "value": 2.0 }],
"left_out": [{ "kind": "market_cap", "op": ">", "value": 1000000000.0, "reason": "describes today, not the trade date, so it cannot cut past trades" }]
},
"matched_trades": 1997,
"too_few": false,
"track_record": { "scope": "filtered", "sample_size": 1997, "win_rate_pct": 36.8, "…": "the Track record fields" },
"unfiltered": { "scope": "unfiltered", "sample_size": 6835, "…": "the Track record fields" }
}The Edge Filter switch
The Edge Filter grades every setup S, A, B, C or F, from strongest to weakest, the same grades the screener shows. S to C are the setups it keeps; F is one it would skip. edge=on counts only the past trades it would have kept; edge=off, the default, counts every trade. edge.applied is false in the rare case the stored trades carry no grades, and the record is then every trade's.
Filters that describe today
Some filters describe a company as it is today: market cap, security type, exchange, sector and country. A company worth $50 billion now may have been worth $2 billion when it traded a wedge in 2012, so filtering past trades by today's size would quietly answer a different question. Phase, distance to the trigger and the model's estimates describe tonight's setups only. None of these can cut past trades. They come back under filters.left_out, each with its reason, and the record ignores them. When a person's screen included one, say so.
Too few trades
A win rate over a handful of trades says little: three wins out of four is 75%, and the next trade could make it 60%. Under 30 matched trades, too_few is true and track_record is null. matched_trades still says how many matched.
Break-even trades
A trade that closed exactly where it started neither won nor lost. It counts as a trade, but not against the win rate: the win rate is wins out of wins plus losses.
Family versus ticker
The family record is large (hundreds to thousands of trades) and stable. The ticker record is the same measurement on one symbol and is often small; it appears on a setup only when there is enough history to say anything, else track_record.ticker is null. When both exist, lead with the family and mention the ticker as colour, and say the sample size of each.
Where to get them
- On every setup, in
track_record, so a script never needs a second call to explain a setup. GET /familiesfor all five families at once, no key needed.GET /families/{family}/statsfor one family.- The
get_family_track_recordMCP tool, which works without a key. GET /track-recordand theget_filter_track_recordMCP tool for the record of any filter set (Filtered records); they need a key on Premium+.