#60 — Indicator-vs-optimal-direction screen

2026-06-17

Verdict: negative · ⛔ negative — no indicator predicts optimal-trade direction

Screening ~70 library indicators against the oracle's optimal-trade direction found nothing above ~53% agreement (coin flip = 50%); the only cross-pair survivor is a weak contrarian DeltaPrice at r≈−0.06.

Hypothesis — what it tests

Whether any indicator in the library, momentum or contrarian, carries usable agreement with the direction of the trades a perfect-foresight oracle would take — i.e. whether the entry-direction problem is solvable at all with these features.

Description

With the oracle's optimal trades in hand (entry #59), the natural next question was whether ANY indicator in the project's library agrees with the direction a perfect trader would take at the entry bar. All ~70 indicators — 45 from the causal Kalman10 parquets plus 24 'encyclopedia gap' classes computed bar-at-a-time — were aligned to each oracle optimal-trade entry with a closed-M5 lookahead guard, then scored by point-biserial correlation and best-sign directional agreement (sign flips allowed, so a contrarian read counts). Nothing predicts direction: the best agreement was ~53% against a 50% coin flip, and the only signal surviving on both test pairs was DeltaPrice = tanh(ΔClose/ATR) at r = −0.06 — contrarian, like every other above-noise reading. RSI at M5/M15 never helps (only D1 RSI2, weak and contrarian). This is one plank of the session-wide conclusion that direction is unpredictable net of spread while magnitude and exhaustion are not.

Key result

Best directional agreement ~53% vs 50% coin flip; only both-pair survivor DeltaPrice=tanh(ΔClose/ATR) r=−0.06 (contrarian); every above-noise signal contrarian, never momentum; RSI at M5/M15 never helps

Indicators

~70 library indicators incl. RSI(14)Stochastic %K/%DCCIAroon oscillatorDonchian positionrange positionEMA8/EMA21 ratiosATR ratioBollinger band widthDeltaPrice (tanh(ΔClose/ATR))

Algorithms

point-biserial correlation screendirectional-agreement scoring with best-sign selectionperfect-foresight oracle labeling

Code

research/experiments/oracle_traits/indicator_vs_oracle.py research/experiments/oracle_traits/screen_gap_indicators.py

Interactive version (search, filters, figures): the experiment explorer.

← #59 Oracle optimal-trade ceiling (GBP_JPY S5)#61 Continuation vs exhaustion inside resolvable moves →

This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.