Evidence · Statistical Validation
Is the edge real? We test it every trading day.
After every close the live record is re-tested against RSP, the benchmark we committed to before trading began, and the reading is published favorable or not, never repainted.
- Trading days
- 331
- Benchmark
- RSP
- Gross at 95% (two-sided)
- significant
- Tested as of
- 20 Oct 2026
| Alpha Convergence total return | Gross | Net of 2% model fee* | RSP, total return |
|---|---|---|---|
| Since inceptionstrategy record, from 27 Jun 2025 | +56.8% | +52.8% | +22.3% |
| Last twelve monthsfrom 21 Oct 2025 | +36.3% | — | +15.5% |
Cumulative, as of 20 Oct 2026. Includes 27 Jun to 18 Aug 2025 on firm capital before client accounts began. Net deducts a 2% model advisory fee*, the top of the fee schedule, not the fees actually billed. The statistics that follow are computed on gross returns. Past performance is not indicative of future results.
* Net figures deduct a 2% model fee, the highest fee we charge. Our fees range from 1% to 2% a year, depending on the size of the account. The full schedule is in Form ADV Part 2A.
- Alpha t-statistic · gross · HAC2.38Newey-West HAC, q = 4
- Trading days on record331Strategy record since 27 Jun 2025 · includes 27 Jun to 18 Aug 2025 on firm capital
- Information ratio · gross1.91Against RSP · tracking error 10.1%
- p-value · gross · one-sided · HAC0.0086Not the probability of future success
FlightDeck · as of 20 Oct 2026. Active return is measured against RSP total return. Significance is measured gross, as the SEC marketing rule permits for a statistic of skill. Computed on gross returns; does not reflect the deduction of all fees and expenses. Net total returns for the same period are shown in the Alpha Convergence total-return table above. The primary test is one-sided (is active return above zero?); the 95% interval is two-sided, the more conservative choice. These are observations from a short record, not a forecast.
Significance over time
Gross · expanding window from day 60The test battery
Seven ways of asking the same question
Each test makes different assumptions. The question is the same: could a manager with no skill have produced this record by luck?
| Test | What it asks | Reading |
|---|---|---|
| Newey-West HACq = 4, primary, gross | Did the selection process beat RSP, allowing for serial correlation in daily returns? | t = 2.38 p = 0.009 one-sided |
| Size of the edgegross, a year · HAC | Mean active return a year before fees, with its 95% interval (Newey-West HAC, two-sided). | +19.3% 3.4% to 35.1% |
| Jensen alpharegression, HAC, gross | Controls for beta instead of assuming a beta of one | α = +20.8% t = 2.58 |
| IID t-testtextbook | The textbook version, kept for comparison | t = 2.19 |
| Block bootstrapgross · 20,000 draws | No normality assumption: resamples the record in blocks, preserving its day-to-day structure. Its own 95% interval for gross active return a year: 7.3% to 31.2% | p 0.0011 one-sided, gross |
| Alpha Opportunity daysselection, where it can register | Measured only on calm, differentiated days, when stock selection can be seen at all | 8.1 bp/day t = 2.19, p = 0.028 |
Read from FlightDeck’s edge-significance and ac-monitor feeds, never recomputed here. Active return is measured against RSP total return (price plus its dividend accrued evenly). The net readings deduct a 2% model fee*, not the fees actually billed. A p-value below 0.05 means a result at least this strong would arise by luck less than 5% of the time if there were no skill; it is not the probability that the strategy will keep outperforming.
* Net figures deduct a 2% model fee, the highest fee we charge. Our fees range from 1% to 2% a year, depending on the size of the account. The full schedule is in Form ADV Part 2A.
A monitor, not the claim
A statistic checked every day touches a fixed threshold more often than a single reading would, so a first crossing is never treated as a finding. FlightDeck draws the reading for every day since day 60 and never repaints a stretch that had not earned it.
Why three readings
The excess-return test, the Alpha Opportunity-day reading and the factor-adjusted readings answer different questions: did we beat the benchmark, did selection work when it could be seen, and is it more than factor exposure.
Early ratios run high
Information ratios this early in a record usually reflect favorable sequencing, and tend to compress toward 1.0–1.5 over full market cycles.
The pre-specified primary test
As of 15 Oct 2026: benchmark RSP total return; basis gross active return, with net of the 2% model fee* alongside; one-sided (is mean daily active return above zero?); Newey-West HAC standard errors, q = 4; from 27 Jun 2025. Factor-model and other tests are supplementary diagnostics on the same record. Every t, p and interval on this site states its basis, sidedness and method.
* Net figures deduct a 2% model fee, the highest fee we charge. Our fees range from 1% to 2% a year, depending on the size of the account. The full schedule is in Form ADV Part 2A.
Three things this does not mean
It does not predict future returns. It does not guarantee any level of outperformance. It is not third-party verification: it is our live record, tested in public.
Over time
The headline, as a live chart
FlightDeck’s Statistical Validation page draws the expanding-window t-statistic for every trading day since day 60, colored by what the number was on that day, never repainted by today’s reading, with the 95% and 99% thresholds marked.
Backstage tour
See it running live
A live screen-share of FlightDeck with Mark Tennenbaum, CIO: the command center driven in real time, how today’s market is classified and what the book does about it, and the record with the tests behind it.