Signals
A signal is one Radar condition, from the day it opened to the day it resolved. The detectors, thresholds and lifecycle rules are on Radar: daily discoveries.
Status on a day.
/v1/signals?date=… returns the signals active on that day, with the status they had then,
from lifecycle events up to that day only. A later re-entry never rewrites an earlier day. Without date, the list
is the signals first detected in a date range, with their status in the latest run.
/v1/signals/{id} adds insight, the signal’s full opening evidence (headline, how rare, reference window,
definition, quality flags, method), and lifecycle, its events: NEW, ESCALATED, ONGOING, RESOLVED.
Point in time
A signal’s evidence, and the earlier cases it is compared with, use only data dated on or before its first day. They are computed once and never revised. Onlyrealised (what followed this signal) fills in as sessions pass.
Outcomes: what followed earlier cases
/v1/signals/{id}/outcomes compares the signal with earlier cases:
- the same market, metric, bucket, kind, horizon and direction, strictly before the signal’s day, on days that passed the data-quality checks;
- at most one case per 20 trading days, so overlapping signals count once;
- for each horizon, only cases whose outcome day was already past on the signal’s day (
as_known_on).
volpts values are vol as a decimal: -0.0275 is −2.75 vol points. logret is a log return: 0.0102 is about +1.0%.
For each horizon of 1, 5, 10 and 20 sessions you get n (cases), median, mean, p25, p75, share_up (the
share of cases in which it rose), min, max and evidence. Evidence labels: under 5 cases insufficient
evidence, 5–9 limited evidence, 10–19 moderate sample, 20+ larger sample. Counts can differ by horizon:
a case counts only once enough sessions had passed. Add ?cases=true for each case’s values.
realised is what this signal did over the same horizon, from the same arithmetic. It is null until that many
sessions have passed.
Structures
/v1/signals/{id}/structures shows how option structures did after the earlier cases of a signal, as the
investigation page’s section D does at its defaults:
- Which signals: notable discoveries in single markets (not market pairs). Others answer 404 with the reason.
- Which trade: each structure entered on each earlier case’s day, on the nearest listed expiry with at least
7 days left that is still alive at the exit (
m: 1), held 20 sessions, with costs of 0.5 vol pt per leg each way. The most recent 20 earlier cases whose trade had closed before the signal’s day. - Four results per structure: long and short, unhedged and delta-hedged. Each has
n(trades),medianandworstP&L as a fraction of the futures price, andshare_profitable. suggested: the side the Radar lists first for this kind of move. A fixed rule of thumb, not fitted to the results, and not advice.
The grid and series
/v1/snapshot?date=… returns the Radar grid: one cell per market, metric and bucket, with the signals active in it.
A date with no data returns the latest day before it; the response’s
date says which.
/v1/series?market=…&metric=… returns a metric’s daily values, the series the detectors read. Buckets are 30D
(default), 90D, 180D and M1. Vols and spreads are decimals (0.604 is 60.4%), futures_price is the price.

