How we research

Pre-registered falsification. Sealed before it runs. One attempt.

Every candidate strategy — from published papers, trading books, championship records, practitioner claims — faces a test designed to kill it, written down and cryptographically sealed before it runs. Eighty-two sealed tests so far. Five clean passes. Everything else was refuted, and the refutations stay on display — that is why the survivors mean something.

82

sealed hypothesis tests

5

clean Charter passes

34

retired & on display

0/2/5

bps cost tiers, frozen

Counts, not performance — this page publishes no return numbers. Results live on the record-only surfaces linked below, where they are read from the sealed ledgers at runtime.

The process

Five stages. The third one kills most ideas.

01

SOURCE

claims

Candidate edges come from published research, trading books, championship strategies, and practitioner claims. Each one is reduced to a falsifiable statement — exact entry, exit, universe, and timeframe — before any code runs. If a claim can't be pinned down precisely enough to kill, it doesn't enter the queue.

Inside this stage

Falsifiable construction (exact rules, universe, timeframe)Element audit — every layer of the source strategy accounted forNovelty check against everything already tested
Hands offa testable hypothesis

02

PRE-REGISTER

the test

Before a single bar is evaluated, the complete test — thresholds, cost tiers, verdict criteria, and the SHA-256 of the exact dataset — is written down, hashed, and timestamped. Cost tiers are frozen at 0 / 2 / 5 bps so a marginal result can't be rescued by quietly moving the goalposts. The single-attempt rule is binding: a sealed verdict is consumed whether we like it or not.

Inside this stage

SHA-256 pre-registration + OpenTimestamps sealDataset content-hash bound into the seal (data-integrity gate)Cost tiers frozen at 0 / 2 / 5 bps — no goalpost creepSingle-attempt rule
Hands offa sealed pre-registration

03

TEST

to destruction

The battery is designed to produce refutations. Walk-forward and combinatorial purged cross-validation ask whether the edge is consistent; bar-, action-, and placement-permutation nulls ask whether it beats chance; a matched-random control asks whether the entries carry information beyond their exposure pattern; the beta gate asks whether the 'alpha' is index exposure in disguise; and de-Prado deflation discounts any result that came from a best-of-N sweep. A look-ahead audit re-runs the feature builder to prove future bars can't leak in.

Inside this stage

Walk-forward + CPCV consistencyBar / action / placement permutation nullsMatched-random entry controlBeta-neutrality gate (alpha, not disguised index exposure)Deflated Sharpe / PBO for any best-of-N selectionPoint-in-time look-ahead audit
Hands offa verdict that binds

04

PUBLISH

the verdict

Every sealed test ends in a public verdict. Refuted strategies are retired to the Hall of Retired with their cause of death quoted from the sealed record — they are the majority, and they stay up. Passes receive a descriptive Charter grade card and a capacity estimate. When we find a defect in our own published work, the correction is published too: the v3-ORB DRL line was retired in the open when a clean retrain showed its edge was a training-data look-ahead.

Inside this stage

82 sealed hypothesis tests to date — 5 clean passes34 retired strategies on display in the Hall of RetiredPublic corrections when the defect is ours (v3-ORB retirement record)
Hands offgrade card or retirement plaque

05

TRACK

forward, record-only

A pass is where scrutiny intensifies, not where it ends. Every survivor is tracked forward on data that did not exist when it was sealed — record-only, no capital, no orders — and the board shows the decay when it happens: some certified strategies have already gone negative forward, and that is displayed rather than removed. The forward record, not the backtest, is the number we ask you to judge us on.

Inside this stage

Record-only forward shadows on unseen dataWarts-and-all display — negative forward marks stay upNightly refresh from the sealed ledgers
Hands off/forward-shadows

Enforced by artifacts

The integrity layer

The discipline is enforced by artifacts, not promises: sealed pre-registrations you can hash yourself, a build-time data-integrity gate that refuses corrupt bars, and an open-source verifier that re-runs the Charter rules on the same inputs in minutes.

SHA-256 + OpenTimestamps sealsbuild-time OHLC / look-ahead data gatecharter-spec (open-source verifier)public correction records

When the defect is ours, the correction is public: the v3-ORB DRL line was retired when a leak-corrected retrain collapsed its edge — the retirement record stays up. An entire earlier era of 1-minute results was retracted when a data-integrity audit found mislabeled OHLC columns; the data gate that catches that class of defect now runs before any test.

The output

Where the results live.

Numbers on this site come from the sealed ledgers and are rendered at runtime — never typed into marketing copy.

Forward Shadows

The survivors, tracked record-only on unseen data — no capital, no orders, warts-and-all. Some are already negative; they stay up.

Track Record

The current verified book at a glance: what passed, its descriptive grade, and honest capacity / minimum-account estimates.

Hall of Retired

Thirty-four strategies tested once, sealed, and refused — each displayed with its cause of retirement, quoted from the sealed record.

Alpha Leagues

Fantasy portfolios on the verified cells — every team sealed with a SHA-256 creation event and tracked forward in public.

Don't take our word for it

The verifier is open source.

The Charter rules are published as charter-spec (MIT-licensed), and /charter/verify walks you through re-running any published score in about five minutes. Running your own strategy? The Charter audit intake takes ten minutes and your code never leaves your machine.

The discipline produces few survivors, on purpose. See what it's for.

Back to Lifestyle Alpha