Validation-as-a-Service
Submit a strategy spec or a backtest. We run the same 26-method validation battery, purged combinatorial cross-validation, and corpus-contamination tests that our internal research uses. You get back a per-method pass/fail report — the honest version.
Who this is for
You built a strategy. The backtest looks great. You want a leak-clean, execution-realistic, fee-stressed outside opinion before deploying live capital.
You have an internal strategy library and want an independent stamp on a candidate before allocation. We sign an NDA, ingest your spec, return a per-method report.
You're writing a paper or a course on quant validation. We provide a white-labellable per-method scoresheet on your candidate strategy plus our methodology in attached citation form.
What we run
All 26 methods, in five families. We do not skip families to pad the pass rate.
- · Purged combinatorial CV with embargo (López de Prado)
- · Deflated Sharpe ratio
- · White's reality check / SPA
- · Probability of backtest overfitting (PBO)
- · Synthetic-null markets
- · Post-resolution price scrubbing (L1)
- · Calibration leak (L2)
- · Full-sample fit detector (L3)
- · Snapshot misalignment (L4)
- · Truncated-tape leakage (L5)
- · Time-shuffle test (B6)
- · Orderbook-depth fill simulator with market-impact
- · Adverse-selection model (maker quotes)
- · Fee stress (+25/50/100%)
- · Slippage stress
- · Capacity scan (2×/5×/10×/25×)
- · Bull / bear / chop partition
- · Volatility quartile partition
- · Liquidity quartile partition
- · Time-of-day / day-of-week buckets
- · Cross-universe OOS transfer
- · Pre-registered hypothesis (Sharpe / win-rate / per-trade)
- · 14-day SPAN gate against the live tape
- · Realistic-fill validation (live book vs mid)
- · Champion-challenger A/B vs null persona
- · In-production walk-forward Sharpe monitor
Tiers
All tiers include the full per-method scoresheet. The difference is depth, turnaround, and inclusions.
Corpus & leak audit
Family B only. Five leak tests + time-shuffle, against the corpus you supply. 5 business days.
- · L1–L5 + time-shuffle
- · Per-test pass/fail + magnitude
- · Corpus-contamination rebuild recommendation if any test fails
- · 30-min results call
Full battery
Families A–D, 21 methods. Family E (forward-paper) optional add-on. 7 business days.
- · All 21 backtest-only methods
- · Capacity curve
- · Fee/slippage stress contours
- · Regime-stratified P&L heatmap
- · 60-min results call + written report
Battery + forward-paper
All 26 methods including 14-day forward-paper SPAN gate. 21 calendar days end-to-end.
- · Everything in Full battery
- · 14-day live-tape paper run
- · Pre-registered hypothesis log
- · Champion-challenger vs null
- · Daily P&L vs backtest delta report
Enterprise (≥10 strategies, white-label, recurring) — pricing on request.
How a slot runs
- 1. Intake. You submit a strategy spec — code, paper, or natural-language description — plus the corpus (or pointer to a public corpus we already have). NDA optional, signed before code lands.
- 2. Mapping. We map your strategy to the 24-class edge taxonomy (E01–E24) and identify the most likely failure modes a priori, before running the battery.
- 3. Battery. We run the chosen tier's methods on our F1106 engine. Each method's output is logged with raw numbers, not just pass/fail.
- 4. Report. Per-method scoresheet + an executive summary of the failure modes (if any) and the cheapest fixes.
- 5. Call. 30/60 minutes (tier-dependent) to walk through the report. Your code stays yours — we keep an anonymized hash of the strategy classification only.
Honest disclosure
Most strategies we'll run through this battery will fail something. Of the 43 internal strategies we ourselves pushed through it after passing purged-CV, zero survived all 26 methods — five are battery-clean on A–D and currently in the 14-day SPAN window. We don't pad pass rates. We will tell you the specific gates your strategy fails, the magnitude, and the cheapest engineering fix. That is the value. If your strategy needs a "looks good" stamp instead of an audit, this is the wrong service.
Past validation does not predict future P&L. None of this is investment advice. Slot availability is limited because each engagement consumes real compute on the same engine that runs our internal research.