A Pre-Registered, Cryptographically Timestamped Proving Protocol for Directional Market Forecasting, with Announced Echo State Network and Quantum Reservoir Computing Examinations

Directional forecasting claims about markets are, in common practice, unfalsifiable: the search behind a published strategy is unreported, the benchmark is chosen alongside the results, and failed attempts leave no record. We present a proving protocol that makes such claims checkable without trusting the authors. Every evaluative decision is committed to a cryptographically timestamped append-only record before it can be compared with outcomes. We report the protocol's first completed application: a baseline stage across five market target streams, in which four model paradigms -- from regularized linear models to a large language model -- all failed to exceed floors set by a pre-registered naive family operating on price alone, net of costs. The floors, the statistical machinery (a cumulative hypothesis family under Romano-Wolf stepdown control, validated on synthetic ground truth), and the enforcement patterns that make protocol violations machine-detectable or impossible by construction are published in full. On this record we announce, in registered-report form, the project's next examinations: an echo state network stage, and a quantum reservoir stage evaluated as the performance difference Δ, evaluated two-sided, against an exactly twinned classical arm. To our knowledge this is the first pre-registered forward test of quantum reservoir computing against a matched classical counterpart in financial forecasting. The design invariants are timestamped with this paper; the full designs undergo registered-report review before any run. Their outcomes, and the per-arm grid of the baseline stage, are published in a separate results report under the disclosure policy fixed here, whatever the sign of Δ. Version 1.1 — errata. Editorial corrections only; no change to methods, data, results, or conclusions. A substantive revision (v2) responding to external reviewer comments is in preparation. (1) Terminology: "symmetric difference" replaced by "performance difference Δ, evaluated two-sided" throughout (abstract, §1, §8.1, Fig. 1) [R-E1]. (2) Factual correction, §7.2: the smallest adjusted p-value (0.9388) was described as "three orders of magnitude" above the threshold; against the registry-level per-stream threshold of 0.01 the ratio is roughly 94, just under two orders of magnitude [R-E2]. (3) §7.2: the claim that numeric blinding makes corpus leakage "structurally inapplicable" is weakened to "substantially reduces the risk" [R-E3]. (4) References: four internal editorial notes removed and metadata completed [R-E4]. Concept DOI unchanged. SSRN mirror revised with the same PDF.