What it refuses to tell you
The refusals built into EdgeDepth Research: what the query grammar makes impossible to ask, why absence is reported instead of coerced, and why a verified zero is a finding.
Research computes what happened, with denominators, and refuses to go further. The refusals are enforced by the query grammar, not by policy.
- No win rates. A win rate needs an entry, an exit, a size and an account. A market record contains none of those, so any win rate computed from one is a modeling choice dressed as a measurement. You get counts over denominators: N occurrences across M eligible symbol-buckets scanned.
- No predictions. Outcome distributions measure what followed past matches; they do not estimate what will happen next.
- No outcome filters. Each occurrence shows what the market did after its anchor, computed forward and labeled as such, but you cannot filter or sort on it. Ask for “times vpin spiked and price then dropped 2%” and the engine returns an error code, OUTCOME_IN_PREDICATE: filtering on the future is survivorship bias (the lookahead bias every backtesting guide warns about), so the grammar makes it impossible instead of warning about it.
- No silent shrinkage. When a feature is not yet warm at some minute, that bucket leaves the denominator and the exclusion is reported. Absent is absent; it is never coerced to zero.
- No advice. A verified zero (“this condition did not occur in the scanned record”) and a verified thousand get identical treatment: counted, evidenced, replayable. What either means for your trading is yours to decide, on your own account.
FAQ
Why won’t it show win rates? Because a win rate smuggles in an entry rule, an exit rule and a holding period that the record does not contain. Whatever numbers we picked would be editorial. Counts and denominators don’t pretend.
What happens when a scan finds nothing? You get a verified zero: 0 occurrences across the same reported denominator, under the same reproducibility key. The condition did not occur in the scanned record. That is a finding, and it is publishable.
Can I filter by what happened after an occurrence? No. Outcome columns are computed forward from the anchor and displayed, never filterable and never sortable. That rule has an error code, OUTCOME_IN_PREDICATE, because selecting on the future is survivorship, not research.
Occurrences describe only the recorded markets, dates and eligible buckets reported by the scan. They are observed examples, not a guarantee. Nothing here is trading advice.