Inside the lab

How we validate a signal before it ships.

SENTINEL is a public crypto-perp research desk. Most ideas don’t survive contact with the data — and we show that work. This page is the validation pipeline, the route simulations, and what’s currently soaking.

Live figures load on each visit from the same endpoint our admin dashboard uses. Everything labeled “research” or “reconstructed” is exactly that — not a live track record.

ELLIE tracks validation health and helps spot model/alert drift before a rollout.

ELLIE waiting
Live route window

Current CORE window

Live route-simulation evidence for subscriber-shipped CORE tiers, loaded from /api/scanner/tier-performance. Bars below 1.0 stay visible; thin samples stay flat. Forward receipts outrank this chart.

live endpoint14d window
Loading live CORE route window...
Research stress test

$1,000 through the reconstructed backtest

The Skeptic’s Console

$1,000 starting · Mar 16 – Jun 2, 2026 · 78.3-day window · exit = time_up

RECONSTRUCTED · not a live ledger
Realistic (30s latency) Idealized 0-latency ceiling BTC buy & hold (last 48d)
Realistic return
+11.4%
Profit factor
1.26
Win rate
44%
Max drawdown
5.7%

Idealized 0-latency return +35.9% · realistic 30s-latency cuts it to +11.4%. The gap is the honest cost of execution.

Read this before you trust the curve. Reconstructed via the in-repo parity backtester on snapshot data — NOT a live track record. Survivorship-limited, single 78-day window, 0.22% round-trip cost, 10% position sizing.

  • Backdrop: 76% flat / 16% down / 8% up — a chop-dominated tape, not a bull run. BTC fell -8.2% over the covered window.
  • Block-bootstrap / walk-forward put the probability of profit-factor < 1 near 71%, with fold fragility (train PF 2.66 → test 1.12). Treat the curve as a hypothesis, not a promise.
  • Recent reality check: this is one favorable 78-day window. Over the last ~2 weeks the live tape has been materially tougher — see the live CORE window further down this page for the current numbers. We’d rather you compare them than only see the flattering one.
  • When our live ledger fills, it replaces this reconstruction. Until then, treat every number here as a hypothesis we are testing in the open — not a track record.
The bar

How candidates earn live delivery

step 1

Discovery

Mine candidate signals; reconstruct outcomes path-aware from history.

step 2

Shadow soak

Run live but invisible for ≥12 weeks, ≥20 entries, ≥10 resolved exits.

step 3

Latency stress

Re-test under realistic execution lag — idealized edges that collapse are cut.

step 4

Live ledger

Accrue real forward-traded outcomes (no reconstruction).

step 5

Graduation

Promote only after clearing predeclared profit-factor + sample gates.

Methodology receipt

Why we test under real-world latency

An exit policy that looks great at zero latency can fall apart once a real alert-to-fill delay is applied. We ship the one that holds up, not the one with the prettiest backtest.

profit-factor retained at ~30s execution laghigher = more robust

time_up is the live policy (accent). Idealized trailing policies look good at zero latency but collapse under real alert-to-fill delay. Reconstructed from the parity backtester.

Validation health

Live validation posture

Research posture

ELLIE tracks what still needs evidence before anything is promoted.

The research lane keeps the same live route telemetry visible while marking where lab evidence is preliminary, thin, or paused for sample risk.

Core route source
...
loading
Top deployed bucket
loading
Backtest confidence
awaiting backtest
Backtest source
unavailable
1h IC / n unavailable

ELLIE telemetry is still loading in this preview.

ELLIE waiting
Model operations

Model and alert diagnostics

Model tuning diagnostics

Public view of ML edge evidence and threshold-risk posture from live bucket telemetry.

Loading optimization telemetry…

Loading model telemetry...

Currently soaking

In the lab right now

HTF Supertrend

soaking
soak 3.5/12 weeks

Higher-timeframe (weekly) trend-rider. Reconstructed research estimate ~PF 2.0–2.5 (corrected down from an inflated 4.4) — NOT a live track record, and bear-market behaviour is untested. Live-shadow soaking now.

Exit policy · time_up

latency-robust

Retains 87% of profit factor under a 30s execution lag, versus 35–42% for trailing-stop policies. Shipped live behind an instant env kill-switch.

HTF-trend overlay

tested · no edge

Tested whether a trend filter lifts the live tiers. It did not survive a cluster bootstrap (confidence interval spanned zero). Not shipped — a deliberate negative result.

House rules

The honesty contract

  • We don’t market returns we can’t prove on a live ledger.
  • Reconstructed and shadow numbers are labeled as such — never as a track record.
  • Negative results ship too: an idea that fails validation is a finding, not a failure.
  • Small samples make us more cautious, not more excited.
Next proof step

Verify, then decide.

The lab page explains how ideas graduate. Verify the live ledger, inspect losses, then choose CORE or Research Lab from the desk.