Live signals fire on entry (status=open). Closed trades fall back to the most recent for context. Polling every 5s during 09:30-16:00 ET.
| Time (NY) | Side | Level | Rule | Entry | Stop | Target | Status | Exit | Exit Reason | R | MAE (ticks) | MFE (ticks) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||||||
| Rule | N | Win % | Exp R | PF |
|---|
| Session | N | Win % | Exp R | PF |
|---|
| DOW | N | Win % | Exp R | PF |
|---|
| In News | N | Win % | Exp R | PF |
|---|
Edge tab — Phase 2
Available after n≥50 live closed trades across all entry rules.
Each of the 14 hypotheses from the deep-research synthesis runs against current data and reports whether the evidence yet supports, refutes, or is inconclusive about that finding. The framework's job is to tell you when each becomes answerable — not to fabricate effects from noise.
| ID | Hypothesis | Lens | n / min n | Status | Recommendation |
|---|---|---|---|---|---|
| Loading… | |||||
Decide on a PENDING signal within 5 minutes to keep it marked as in-window. Older decisions are flagged as hindsight-tainted and weighted down in analysis. This is the bias-correction layer the deep-research synthesis flagged as the biggest hidden lever.
| Signal Time (NY) | Obs ID | Side | Decision | Within Window | Δ sec | Actions |
|---|---|---|---|---|---|---|
| Loading… | ||||||
| Constant TF | Side | Variable Set | n w/outcome |
Avg Score | Win % (Beta-shrunk) [90% CI] |
Confidence | Avg MFE (pts) |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||
This dashboard refuses to claim more than the data supports. The conventions below are why you'll see insufficient_data on every validation hypothesis today and why most confluence cells render at 42% opacity.
Every confluence observation is forward-evaluated against three barriers — upper TP (1R / 2R / 3R), lower SL, and a time-out (30 / 60 / 240 / 1440 min). Whichever fires first becomes the label. This is Lopez de Prado, AFML Ch. 3 — replaces the naïve win/loss with multi-class outcomes and a fixed evaluation horizon.
outcome_evaluations.first_barrier_hit ∈ {TP_1R, TP_2R, TP_3R, SL, TIME, NO_FEED}
Win-rate at small N is mostly noise. We use a Beta(2, 2) prior — weak, neutral, equivalent to 4 pseudo-trades at 50%. The posterior pulls every cell toward 50% inversely with N. A cell with 1/1 doesn't render at 100% — it renders at 60% (= 3/5 with prior).
90% credible interval shown as Wilson interval — more reliable than normal approximation at N < 30.
These are calibrations, not promises. Sample-size = number of observations with attributed outcomes, not raw observation count.
Every signal fired by the indicator opens a decision slot. The trader records
TAKEN, SKIP_quality, SKIP_session,
SKIP_risk, or SKIP_other within 5 minutes.
Decisions recorded after 5 min are flagged hindsight and weighted down in
analysis. Without this layer, the matrix measures only "outcomes conditional on having
been taken" — selection-biased.
Cron auto-expires PENDING signals past their 30-min window to EXPIRED.
14 hypotheses derived from the deep-research synthesis. Each runs against current data and returns one of:
A "trustworthy" chip at N=120 still means the 90% CI on win-rate is roughly ±7%. An "institutional" chip means roughly ±5%. Move position-size and rule-hardening decisions only when both the chip and the CI agree with you.
The contrarian-lens persona explicitly warned: "the infrastructure will output numbers at a precision vastly exceeding the statistical precision of the sample." The chip-and-CI design is the guardrail.
Source: docs/research/2026-05-28-ob-confluence-d2-e2/00-SYNTHESIS.md
· 8-persona deep-research squad · synthesized 2026-05-28