Every measurement, in full
This engine writes 21 read-only artifacts and publishes all of them. Several are corrections against our own earlier numbers and several are nulls — results that closed a line of enquiry rather than opening one. Those are the entries worth reading first, because a record that only contains its wins is not a record.
Each page below is generated from the artifact itself. No figure on any of them is typed by hand, so a number that moves in the engine moves here, and a number that exists nowhere in an artifact cannot appear at all.
The measurements
Admission dry run / Result
Admissibility MECHANICS on already-registered identities. Registers no hypothesis identity, proposes no sleeve, and makes no claim about any candidate's edge.
Atlas reachability screen
Screens obtainability only. Opens no return data, registers no hypothesis identity, runs no backtest, and claims nothing about any family's edge, sign, correlation…
Book without alphavintage
A measurement of two allocations on one common window, both reported. Opens no new data, registers no hypothesis identity, changes no allocation and recommends…
Cost model realism
Reads the live trading databases read-only. Registers no hypothesis identity, opens no return data, and changes no cost parameter. 0 trials.
Drawdown live estimator
A simulation through a vectorized twin of the PRODUCTION covariance estimator, asserted equal to ewma_cov on real states before any result was reported.
Execution gap power
Reads trading databases and walk-forward curves read-only. Registers no hypothesis identity and opens no return data. 0 trials.
Feasibility gate reachability
Asks only whether each near-miss gate is reachable by repairing extraction.
Identity redesign / Form universe
Counts filings by form type and enumerates the action types held in the corporate-action lake.
Ledoit wolf effective sample
Measures an approximation's error on real price data. Changes nothing, registers no hypothesis identity, and opens no return data on any untested identity. 0 trials.
Live covariance memory
A property of the ESTIMATOR, derived from its exact weighting and verified against the production function by unit impulse.
Orthogonality prior
A PRIOR, stated before measurement and labelled as such in every row. No candidate correlation is measured, no hypothesis identity is registered, no return data is…
Reachability harness
Reads frozen feasibility artifacts read-only. Registers no hypothesis identity, opens no return data, and authorises no candidate. 0 trials.
Record continuity
Reads the per-sleeve trading databases read-only. Opens no market data, registers no hypothesis identity and changes nothing. 0 trials.
Sleeve admission contract / Contract
Passing this contract makes a result technically eligible for an explicit admission decision.
Sleeve atlas / Atlas
Sleeve atlas / Atlas: a read-only artifact regenerated from the engine run that produced it, published so the figure and its limits can be checked together.
Sleeve atlas / Audit
Passing this audit is not return evidence and does not create a sleeve. It only permits a separately sealed key-free feasibility review.
Sleeve atlas / Lineage audit
This audit proves registry consistency against the enumerated internal ledgers. It does not prove global economic novelty, returns, or sleeve admission.
Sleeve discovery
Sleeve discovery: a read-only artifact regenerated from the engine run that produced it, published so the figure and its limits can be checked together.
Sleeve quality decomposition
A FIRST-ORDER decomposition from published artifacts, read-only. Each dollar component is converted to an annualised return on starting equity and then to Sharpe…
Spinoff prorata gate
A measurement of document LANGUAGE on 98 frozen Form 10 filings, every one hash-verified against the feasibility artifact.
Trial accounting
This ledger proves what was measured and how selection N is counted. It does not prove that any identity is profitable, independent, or admissible.
Where these come from
Each is regenerated by the engine that produced the result and copied verbatim to the glass box, where a downloadable verifier re-checks every content hash and signature against the published bundle. The written-up versions live in the research library.