{"baselines":{"description":"Declared before any candidate is evaluated. At minimum a trivial predictor (empirical marginal distribution of the held-out symbols).","examples":["trivial-predictor","profile-declared-null-model"],"rule":"Baselines are frozen in the profile before candidate submission."},"blinding":"Hidden evaluator generates quantized trajectories from an undisclosed stochastic law and seals them (hash published) before any candidate is submitted. Candidates are evaluated blind; the hidden law is revealed only after all candidate description lengths and predictions are committed.","candidate_encodings":{"description":"Each candidate explanation must be submitted as a fixed, deterministic program whose description length is computable and charged in full against its gain.","rule":"No candidate may be modified after the held-out trajectories are unsealed."},"date":"2026-09-29","experiment_id":"EXP-002","falsifiable_expectations":{"002A_pure_diffusion":"A candidate must NOT achieve positive Turing gain by fitting deterministic curves inside pure noise. Inventing structure where none exists must be charged more than it earns.","002B_diffusion_with_drift":"Drift must earn its own complexity on unseen trajectories: the drift parameter's description cost must be paid by genuine held-out predictive improvement.","002B_zero_drift_control":"When the hidden drift is zero, adding a drift parameter must NOT earn positive explanatory credit. A positive gain here is instrument failure.","002C_mean_reversion":"Mean reversion must be distinguished as qualitatively different state-dependent structure, not merely as drift with a different slope."},"freeze_procedure":"To be defined before freezing: profile hash published, sealed trajectory hash published, candidate submission window closed, then evaluation.","frozen":false,"name":"Stochastic Law Discovery","objective":"Test whether the Turing measure rewards genuine law discovery and punishes invented structure under irreducible noise. A hidden evaluator generates quantized trajectories from an undisclosed stochastic law; the candidate must earn complexity on sealed held-out trajectories.","quantity_measured":"Turing gain (T) of candidate explanatory programs over declared baselines on sealed held-out trajectories","quantizer":{"description":"Continuous trajectory values are encoded as declared quantized symbols per Assumption A3 of the formal paper. The quantizer is a required profile field and is fixed before the hidden evaluator generates data.","spec":"PLACEHOLDER - to be fixed at freeze time: fixed uniform bin width, number of bins, and bin edge convention must be declared here before freezing."},"seeds":"All random seeds fixed and recorded in the profile before freezing; trajectory generation must be reproducible from the recorded seeds.","status":"draft-not-frozen","sub_experiments":{"EXP-002A":{"hidden_law":"pure diffusion (no drift, no mean reversion)","name":"pure diffusion"},"EXP-002B":{"hidden_law":"diffusion with constant drift (drift may be zero in control instances)","name":"diffusion with drift"},"EXP-002C":{"hidden_law":"mean-reverting process","name":"mean reversion"}},"uncertainty":"Report Turing gain with confidence intervals from the declared resampling procedure; recomputation by an independent party from the frozen profile must fall inside the reported envelope.","verdict_rules":["Positive Turing gain on sealed held-out trajectories is required for a candidate to count as a law discovery.","In EXP-002A, any candidate claiming deterministic structure must show non-positive gain; positive gain on pure noise invalidates the run.","In EXP-002B control instances with zero drift, adding a drift parameter must not earn positive explanatory credit.","Recomputation from the frozen profile and sealed data must reproduce the reported gain within the declared uncertainty envelope."],"version":"0.1-draft","warning":"DRAFT profile. Not frozen. No measurement may be reported against this profile."}