quantdata_brooks_events

uk.quantdata/quantdata · Verify this server

Classical Brooks price-action events detected in the current trading window — the day's first range breakout, breakout follow-through, closes in the top or bottom third of an established range, long-lived-range breakouts, climactic spikes — each paired with the outcome rate measured for that exact definition in that exact window (pre-registered, ES 5-minute bars 2010-2026). Range breakouts also carry a calibrated per-event estimate of the probability the breakout closes back through its level within 10 bars (validated zero-shot on NQ 2023+: AUC 0.646, calibration error 5.1%, n=2,637). Answers 'what just happened structurally, and how did events like it resolve historically'. Works around the clock on 24-hour instruments, including the Asia-close-to-US-open gap window. Some measured rates contradict the classical claims; quote the measured numbers, not the folklore. Every read also carries two blocks that answer different questions and must not be mixed: `day_type` is the five-class day-type distribution, from the model trained for the current window. In the regular day session it is mode=day_session, with published accuracy: 66% top-1 over a complete session and 53.5% at the 90-minute mark, against a 37% majority-class baseline. In the Asian window it is mode=asia_session — trained natively on Asian windows, pre-registered, confirmed zero-shot on NQ — whose accuracy figures live in the response's own `accuracy` field and use their own label taxonomy: never quote the day-session figures for an asia read or vice versa. The gap window has no validated model and returns available=false there. `shape` is plain arithmetic over the bars already printed (where price closed inside its own range, how much of the travel was one-way, deepest pullback) — no model, no accuracy claim, available in every window including the gap. Report day_type as a distribution with its baseline; report shape as measurement, never as a forecast. **`day_type.analysis` is the block to lead with (day-session reads only; the asia_session read carries no analysis layer)** — it answers 'should today's read be trusted at all' before it answers 'what is today'. It carries: `confidence` (the model's own conviction crossed with how consistently the most similar past sessions resolved — both high has been right 95% of the time, both low only 41%, against a 66% baseline; say which bucket today falls in and stop treating a low-confidence read as a call); `live_now` (patterns mechanically confirmed on today's bars so far — e.g. 'broke yesterday's low then reclaimed it = bear trap confirmed' — this is what actually happened, trust it over any statistical checklist); `what_to_watch` (patterns over-represented on this day type, each with a status_today of yes/no/forming/unknown and a concrete way to verify it intraday; use the ordering, not the raw lift decimals); and `summary` / `summary_zh`, a one-line verdict written to be relayed verbatim — if you pass along only one sentence, pass that one. Descriptive statistics — not a recommendation. Check window_is_live before calling a read 'now'.