DexIQ Signal Track Record -- Methodology

Skeptic-grade reproducibility. Every number on the landing page can be recomputed by a third party using only the published manifest.json, verify.py, and the rules below.

« back to landing

PnL exit rules (TRACK-02 / D-22-03, pnl_rules.VERSION = v1.0)

For each signal we evaluate the closing 4-hour bar at signal_time + horizon (defaulting to 4 hours) drawn from our local TimescaleDB candles_4h_closed view. Precedence:

1. take_profit_hit: long if high >= tp; short if low <= tp. Exit at tp.
2. stop_hit: long if low <= sl; short if high >= sl. Exit at sl.
3. time_exit: exit at bar close.
4. flat_skip: direction == 'flat'. realized_return = 0.0.

When both TP and SL are touched on the same bar (gap-through), TP wins. We cannot prove the intrabar order; this is a documented convention.

Return formulas

long  realized_return = (exit_price - entry) / entry
short realized_return = (entry - exit_price) / entry

Fees and slippage are not deducted -- this is the signal's performance record. Subscriber-side execution costs are out of scope here.

Deflated Sharpe (TRACK-04)

We follow Bailey & Lopez de Prado (2014) "The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting, and the Disposability of Trials" (Journal of Portfolio Management, 40(5), 94-107). The published value is a p-value in [0, 1] -- closer to 1 means the observed Sharpe is more likely real.

DSR = Phi( (observed_sharpe - E[max(SR)]) / std(SR) )

E[max(SR)] = sqrt(V) * ((1 - gamma) * Phi_inv(1 - 1/N) + gamma * Phi_inv(1 - 1/(N*e)))

We use variance_of_sharpe_estimates = 0.5 (a published-conservative default) and N = number of distinct engine_source values present + 1 as the multiple-testing trial count. Both choices are conservative for the skeptic.

OpenTimestamps verification (TRACK-01)

Every signal_log row is cryptographically timestamped against the Bitcoin blockchain via OpenTimestamps. To independently verify any signal:

1. Install the OpenTimestamps client: pip install opentimestamps-client
2. Download the proof: curl -O https://track.dexiq.io/proofs/<signal_id>.ots
3. Run: ots verify <signal_id>.ots

The proof attests the existence of the row's payload_sha256 at-or-before the upgrade time, which a skeptic can corroborate against any independent Bitcoin block explorer.

Signed manifest + verify.py (TRACK-04)

The dashboard publishes a deterministic manifest of every signal plus its outcome:

• manifest.json -- canonical JSON, sorted by signal_time ASC.
• manifest.json.sha256 -- the SHA-256 of the manifest bytes.
• verify.py -- a standalone Python script that reads manifest.json from a path and recomputes the 5 top-line metrics.

Use:

curl -O https://track.dexiq.io/manifest.json
curl -O https://track.dexiq.io/verify.py
python verify.py manifest.json

verify.py has no DexIQ imports -- only the Python stdlib + numpy + scipy.

Calibration drift monitor (SIG-10)

A weekly cron snapshots the latest 30 settled signals per engine_source and records Brier score, log-loss, ECE, and accuracy into calibration_runs. DRIFT verdict triggers a Telegram alert to operator (NOT customer) and a red banner on this dashboard. Thresholds:

• Brier > baseline 95th percentile + 2*std AND window_n >= 20 AND baseline_window_count >= 8
• ECE > 0.10 sustained over 2 consecutive weekly runs
• Accuracy < 0.50 sustained over 2 consecutive weekly runs

Engine status (D-22-08)

The Phase 20 placebo gate may report FAIL for a given engine variant. If so, that engine emits zero rows; the dashboard surfaces "0 signals" with the engine-paused banner. The infrastructure remains live; another engine variant (e.g. Phase 24 Path A) will start populating the same tables.