Dashboard › Three-Axis Probes

axis 3 · oracle · operational

🧬 Agnostic

Is this ONE well-defined algorithm — would two honest implementations agree? Terry msg 428, verbatim: "algorithm-invariant. Two correct implementations of the same feature should converge to the same answer."

Provenance correction. This is NOT asset-agnostic. The "works across multiple assets" reading was an operator-side drift; the cross-asset sweep belongs to the orthogonal axis. Sources: PROVENANCE.md.

Three tests

testquestiondemo result
A — no-dispatch (static)Does the code switch estimator on a data condition?FAIL on if len≥500: app else perm · PASS on clean feature
B — convergence oracleDo two independent implementations agree to ~1e-9?PASS hjorth (Δ 6.9e-18) · petrosian (Δ 0.0) · FAIL unpinned embedding (Δ 3.3%)
C — doc/code consistencyDoes the declared algorithm match the implemented one?FAIL declared DFA vs implemented R/S (the real lookback_hurst case)
The axes connect: the TEST B failure happens because the embedding order isn't pinned to the project-wide constant — an unpinned parameter is a parameterless FAIL and an agnostic FAIL. Terry's if n≥500 example fails both, by design.

Why PARTIAL

A dispatch/divergence FAIL is conclusive. But a full PASS needs a second independent reference implementation registered per candidate — until then the axis reports PENDING, never a faked pass. Registering reference impls is the open work item.

Deep dive: AXIS-3-AGNOSTIC.md

← ParameterlessThe Gate →