SEALED 256-PATH EVALUATION

Forecast evaluation report

Gold, current, T+1 rollout, and no-rollout checkpoints are compared on the same sealed 256-path post-cutoff corpus. The report separates recursive distribution quality from directional reliability calibration.

Recursive CRPS scores average symbols inside each temporal origin and then weight origins equally. Paired confidence intervals resample origins, not correlated stocks. Calibration maps were selected chronologically on earlier dates; the final calibration date remains diagnostic rather than pristine.

Paths per forecast256Exact production path count

Forecast horizons12Every 4h through 3 days

Temporal origins24 + 21Completed + developing

Symbols21Common cutoff Aug 7

RESEARCH DECISION

Keep rollout training; reject the no-rollout ablation

T+1 rollout has lower all-origin OHLC CRPS than no-rollout at every horizon in both domains, and every origin-clustered 95% interval excludes zero. T+1 remains research-only because its rolling calibrated Brier evidence still trails current at longer horizons.

Completed 4h CRPS delta+0.002419No-rollout minus rollout

Developing 4h CRPS delta+0.003889No-rollout minus rollout

Completed 3d CRPS delta+0.01110295% CI +0.009866 to +0.012335

Developing 3d CRPS delta+0.01026695% CI +0.009166 to +0.011433

ALL-ORIGIN PAIRED EVIDENCE

Recursive forecast distributions

Lower CRPS is better. Every model uses byte-identical truth and the same 256 Monte Carlo paths per symbol and origin. Later horizons contain fewer origins because truth is frozen through August 17 rather than extrapolated.

Completed 4h boundary

OHLC CRPS over 24 sealed completed origins; available origin count declines by horizon.

T+1 rolloutNo rolloutCurrentGold

Developing candle

OHLC CRPS over 21 randomized pseudo-live intrabar origins.

T+1 rolloutNo rolloutCurrentGold

Completed origin OHLC CRPS

OHLC is normalized by the causal origin close. Intervals compare no-rollout directly with T+1 rollout.

Horizon	Origins	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout [95% CI]	Lowest score
4h	24	0.007237	0.009656	0.018165	0.016222	+0.002419 [+0.001580, +0.003326]	T+1 rollout
8h	23	0.013865	0.017685	0.029029	0.024597	+0.003820 [+0.002790, +0.004927]	T+1 rollout
12h	22	0.018784	0.023833	0.038411	0.032645	+0.005049 [+0.004292, +0.005829]	T+1 rollout
1d	21	0.022612	0.028473	0.044870	0.038692	+0.005861 [+0.005241, +0.006508]	T+1 rollout
1d+4h	20	0.026002	0.032618	0.050018	0.043515	+0.006616 [+0.005866, +0.007367]	T+1 rollout
1d+8h	19	0.029345	0.036969	0.054965	0.047533	+0.007624 [+0.006768, +0.008480]	T+1 rollout
1d+12h	18	0.032421	0.040931	0.059332	0.051117	+0.008511 [+0.007716, +0.009387]	T+1 rollout
2d	17	0.035720	0.044796	0.062636	0.054653	+0.009076 [+0.008292, +0.009887]	T+1 rollout
2d+4h	16	0.039129	0.048956	0.065966	0.058322	+0.009827 [+0.008784, +0.010958]	T+1 rollout
2d+8h	15	0.042335	0.052644	0.069216	0.061474	+0.010309 [+0.009069, +0.011677]	T+1 rollout
2d+12h	14	0.045646	0.056338	0.071890	0.064226	+0.010693 [+0.009458, +0.012071]	T+1 rollout
3d	13	0.048255	0.059357	0.073272	0.064898	+0.011102 [+0.009866, +0.012335]	T+1 rollout

Developing origin OHLC CRPS

Randomized intrabar origins use the recorded learned-completion policy.

Horizon	Origins	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout [95% CI]	Lowest score
4h	21	0.013844	0.017733	0.031474	0.019026	+0.003889 [+0.002945, +0.004842]	T+1 rollout
8h	20	0.018293	0.023242	0.039991	0.027948	+0.004948 [+0.004229, +0.005677]	T+1 rollout
12h	19	0.021699	0.027432	0.044785	0.034283	+0.005733 [+0.005068, +0.006391]	T+1 rollout
1d	18	0.025327	0.031815	0.049724	0.040733	+0.006488 [+0.005622, +0.007349]	T+1 rollout
1d+4h	17	0.028680	0.036059	0.054001	0.045666	+0.007379 [+0.006419, +0.008327]	T+1 rollout
1d+8h	16	0.031453	0.039835	0.058180	0.049194	+0.008382 [+0.007551, +0.009272]	T+1 rollout
1d+12h	15	0.034700	0.043748	0.061755	0.053238	+0.009049 [+0.008267, +0.009795]	T+1 rollout
2d	14	0.038138	0.047324	0.064741	0.057336	+0.009187 [+0.008203, +0.010226]	T+1 rollout
2d+4h	13	0.041477	0.051300	0.068195	0.060766	+0.009824 [+0.008697, +0.010985]	T+1 rollout
2d+8h	12	0.044671	0.055095	0.070874	0.063674	+0.010423 [+0.009343, +0.011690]	T+1 rollout
2d+12h	11	0.047359	0.057752	0.072375	0.065895	+0.010392 [+0.009426, +0.011491]	T+1 rollout
3d	10	0.050130	0.060397	0.073552	0.066980	+0.010266 [+0.009166, +0.011433]	T+1 rollout

Completed origin close CRPS

Close distributions are normalized by the same causal origin close.

Horizon	Origins	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout [95% CI]	Lowest score
4h	24	0.009383	0.011197	0.016392	0.014688	+0.001815 [+0.001196, +0.002491]	T+1 rollout
8h	23	0.014823	0.018015	0.027247	0.023485	+0.003193 [+0.002278, +0.004214]	T+1 rollout
12h	22	0.019512	0.024053	0.037164	0.032084	+0.004542 [+0.003752, +0.005363]	T+1 rollout
1d	21	0.023034	0.028467	0.043833	0.037648	+0.005433 [+0.004783, +0.006104]	T+1 rollout
1d+4h	20	0.026156	0.032715	0.049253	0.042264	+0.006559 [+0.005845, +0.007247]	T+1 rollout
1d+8h	19	0.029531	0.037034	0.053931	0.046358	+0.007503 [+0.006587, +0.008459]	T+1 rollout
1d+12h	18	0.032652	0.041016	0.058570	0.050387	+0.008364 [+0.007413, +0.009441]	T+1 rollout
2d	17	0.035998	0.044976	0.061764	0.053887	+0.008978 [+0.008095, +0.009899]	T+1 rollout
2d+4h	16	0.039321	0.049090	0.065076	0.057515	+0.009768 [+0.008657, +0.010951]	T+1 rollout
2d+8h	15	0.042491	0.052825	0.068237	0.060551	+0.010334 [+0.009012, +0.011763]	T+1 rollout
2d+12h	14	0.045662	0.056378	0.070844	0.062970	+0.010716 [+0.009333, +0.012248]	T+1 rollout
3d	13	0.048432	0.059595	0.072326	0.063861	+0.011163 [+0.009933, +0.012420]	T+1 rollout

Developing origin close CRPS

Close-only scoring confirms the joint OHLC result.

Horizon	Origins	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout [95% CI]	Lowest score
4h	21	0.014688	0.017907	0.029568	0.016764	+0.003219 [+0.002404, +0.004085]	T+1 rollout
8h	20	0.018657	0.023029	0.037953	0.026532	+0.004372 [+0.003580, +0.005172]	T+1 rollout
12h	19	0.022141	0.027614	0.043812	0.034059	+0.005473 [+0.004831, +0.006092]	T+1 rollout
1d	18	0.025568	0.031789	0.048493	0.039675	+0.006220 [+0.005415, +0.007023]	T+1 rollout
1d+4h	17	0.028548	0.035902	0.052857	0.044059	+0.007354 [+0.006312, +0.008436]	T+1 rollout
1d+8h	16	0.031881	0.040201	0.057428	0.048425	+0.008320 [+0.007341, +0.009354]	T+1 rollout
1d+12h	15	0.034957	0.043831	0.061295	0.052690	+0.008874 [+0.008109, +0.009656]	T+1 rollout
2d	14	0.038290	0.047261	0.063638	0.056495	+0.008970 [+0.007889, +0.010063]	T+1 rollout
2d+4h	13	0.041949	0.051645	0.067379	0.060270	+0.009696 [+0.008479, +0.010914]	T+1 rollout
2d+8h	12	0.044641	0.055019	0.069691	0.062612	+0.010379 [+0.009242, +0.011722]	T+1 rollout
2d+12h	11	0.047880	0.058125	0.071969	0.065018	+0.010245 [+0.009172, +0.011396]	T+1 rollout
3d	10	0.050512	0.060685	0.072472	0.065982	+0.010173 [+0.009058, +0.011359]	T+1 rollout

Completed origin volume log-CRPS

Volume is scored after log1p. Gold uses its documented fabricated slot-median policy.

Horizon	Origins	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout [95% CI]	Lowest score
4h	24	0.370054	0.382336	0.394820	0.411874	+0.012283 [+0.005315, +0.019870]	T+1 rollout
8h	23	0.385750	0.411632	0.423537	0.400251	+0.025882 [+0.015250, +0.035913]	T+1 rollout
12h	22	0.310645	0.352243	0.395815	0.411724	+0.041598 [+0.031365, +0.051813]	T+1 rollout
1d	21	0.425431	0.474957	0.510481	0.407361	+0.049525 [+0.036324, +0.062711]	Gold
1d+4h	20	0.434305	0.483474	0.516402	0.426952	+0.049169 [+0.032393, +0.065413]	Gold
1d+8h	19	0.440217	0.491671	0.513580	0.406741	+0.051454 [+0.038288, +0.064339]	Gold
1d+12h	18	0.365956	0.414483	0.435243	0.418026	+0.048527 [+0.032782, +0.064184]	T+1 rollout
2d	17	0.481836	0.526296	0.536519	0.411388	+0.044460 [+0.028131, +0.060378]	Gold
2d+4h	16	0.489070	0.531422	0.535698	0.427915	+0.042352 [+0.029321, +0.055159]	Gold
2d+8h	15	0.483970	0.532176	0.528082	0.415347	+0.048206 [+0.034910, +0.061049]	Gold
2d+12h	14	0.406874	0.450530	0.450937	0.437721	+0.043656 [+0.027274, +0.060034]	T+1 rollout
3d	13	0.532153	0.572542	0.564018	0.437997	+0.040389 [+0.023264, +0.056632]	Gold

Developing origin volume log-CRPS

The no-rollout comparison uses learned sampled volume in both checkpoints.

Horizon	Origins	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout [95% CI]	Lowest score
4h	21	0.371101	0.395938	0.414494	0.399864	+0.024836 [+0.015753, +0.033933]	T+1 rollout
8h	20	0.408997	0.446971	0.490957	0.414663	+0.037974 [+0.027100, +0.048103]	T+1 rollout
12h	19	0.337287	0.384050	0.431704	0.416602	+0.046763 [+0.035120, +0.058085]	T+1 rollout
1d	18	0.408432	0.460195	0.489187	0.397001	+0.051763 [+0.037198, +0.065592]	Gold
1d+4h	17	0.425510	0.477571	0.500217	0.411987	+0.052061 [+0.036043, +0.067479]	Gold
1d+8h	16	0.474888	0.519696	0.535145	0.406500	+0.044808 [+0.029597, +0.059969]	Gold
1d+12h	15	0.367814	0.406825	0.416735	0.406949	+0.039011 [+0.024253, +0.052947]	T+1 rollout
2d	14	0.435613	0.477096	0.477383	0.386061	+0.041483 [+0.026201, +0.055967]	Gold
2d+4h	13	0.450453	0.494911	0.492977	0.404932	+0.044458 [+0.030489, +0.058609]	Gold
2d+8h	12	0.522623	0.561335	0.557199	0.426451	+0.038712 [+0.023888, +0.053213]	Gold
2d+12h	11	0.419256	0.451399	0.447625	0.446097	+0.032142 [+0.017307, +0.048207]	T+1 rollout
3d	10	0.503880	0.543938	0.532353	0.451775	+0.040058 [+0.024468, +0.053376]	Gold

FINAL-DATE DIAGNOSTIC

Calibrated Brier curves

These curves compare all four checkpoints after calibration. The first 4h cells remain near neutral and should not be read as proven short-horizon alpha.

Completed 4h boundary

Forecasts originate immediately after a completed canonical four-hour candle.

T+1 rolloutNo rolloutCurrentGold

Developing candle

Forecasts originate at deterministic randomized intrabar minutes, not only at four-hour boundaries.

T+1 rolloutNo rolloutCurrentGold

FOUR-MODEL CALIBRATION

Final-date comparison

All values are calibrated Brier scores on the same final diagnostic date. Bold cells are the lowest score for that horizon; no-rollout minus rollout is negative only when removing rollout training helps.

Completed 4h boundary

Direct four-checkpoint comparison at completed origins.

Horizon	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout	Lowest Brier
4h	0.251939	0.251203	0.253717	0.250786	-0.000736	Gold
8h	0.254757	0.266423	0.292083	0.273101	+0.011666	T+1 rollout
12h	0.251298	0.267536	0.278549	0.266172	+0.016238	T+1 rollout
1d	0.267300	0.281192	0.289545	0.298202	+0.013892	T+1 rollout
1d+4h	0.264405	0.270857	0.283354	0.292324	+0.006452	T+1 rollout
1d+8h	0.261562	0.262744	0.261069	0.271246	+0.001182	Current
1d+12h	0.235688	0.233643	0.233405	0.248142	-0.002045	Current
2d	0.231239	0.233955	0.231881	0.242471	+0.002716	T+1 rollout
2d+4h	0.221672	0.215357	0.205453	0.216717	-0.006315	Current
2d+8h	0.210894	0.196498	0.188396	0.191351	-0.014396	Current
2d+12h	0.215441	0.200500	0.191650	0.197955	-0.014941	Current
3d	0.214356	0.200990	0.194450	0.193318	-0.013366	Gold

Developing candle

Direct four-checkpoint comparison at randomized pseudo-live intrabar origins.

Horizon	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout	Lowest Brier
4h	0.247055	0.249449	0.259715	0.259811	+0.002394	T+1 rollout
8h	0.243924	0.253078	0.274249	0.259697	+0.009154	T+1 rollout
12h	0.251110	0.265278	0.283300	0.275366	+0.014168	T+1 rollout
1d	0.262002	0.272518	0.284005	0.296399	+0.010516	T+1 rollout
1d+4h	0.266335	0.277375	0.303543	0.327233	+0.011040	T+1 rollout
1d+8h	0.241568	0.240157	0.243823	0.273171	-0.001411	No rollout
1d+12h	0.226458	0.221528	0.216380	0.244567	-0.004930	Current
2d	0.215963	0.206291	0.203448	0.219705	-0.009672	Current
2d+4h	0.205480	0.187551	0.186230	0.197613	-0.017929	Current
2d+8h	0.209734	0.195524	0.187303	0.194922	-0.014210	Current
2d+12h	0.208444	0.197133	0.188069	0.191988	-0.011311	Current
3d	0.211906	0.204528	0.192292	0.188543	-0.007378	Gold

STRICT WALK-FORWARD EVIDENCE

Rolling chronological comparison

These calibrated scores aggregate strict walk-forward test folds before the final diagnostic date. This is the stronger promotion evidence: T+1 rollout trails current on the mean and most longer horizons despite its lower raw error.

Completed 4h boundary

Chronological fold scores using maps selected independently inside each fold.

Horizon	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout	Lowest Brier
4h	0.250056	0.250373	0.251107	0.250225	+0.000317	T+1 rollout
8h	0.250145	0.250591	0.251421	0.251323	+0.000446	T+1 rollout
12h	0.245051	0.243599	0.246138	0.243858	-0.001452	No rollout
1d	0.246332	0.243937	0.237248	0.244909	-0.002395	Current
1d+4h	0.243532	0.241558	0.239667	0.249087	-0.001974	Current
1d+8h	0.243476	0.236766	0.229494	0.238049	-0.006710	Current
1d+12h	0.232833	0.227363	0.221573	0.229962	-0.005470	Current
2d	0.232121	0.224590	0.218105	0.225064	-0.007531	Current
2d+4h	0.223681	0.214063	0.204948	0.212360	-0.009618	Current
2d+8h	0.221552	0.209701	0.202573	0.202818	-0.011851	Current
2d+12h	0.222933	0.209815	0.200703	0.201184	-0.013118	Current
3d	0.220551	0.207506	0.197269	0.195362	-0.013045	Gold

Developing candle

Chronological fold scores for the recorded randomized developing-origin policy.

Horizon	T+1 rollout	No rollout	Current	Gold	No-rollout - rollout	Lowest Brier
4h	0.248601	0.247994	0.240075	0.250349	-0.000607	Current
8h	0.245809	0.243659	0.236207	0.244304	-0.002150	Current
12h	0.239822	0.237148	0.237156	0.247684	-0.002674	No rollout
1d	0.242546	0.239932	0.234739	0.250428	-0.002614	Current
1d+4h	0.244076	0.243023	0.244696	0.262203	-0.001053	No rollout
1d+8h	0.235852	0.231288	0.227834	0.244634	-0.004564	Current
1d+12h	0.230018	0.223494	0.217400	0.232603	-0.006524	Current
2d	0.219580	0.210114	0.204841	0.217368	-0.009466	Current
2d+4h	0.218540	0.203376	0.207933	0.215226	-0.015164	No rollout
2d+8h	0.218304	0.204461	0.199940	0.204136	-0.013843	Current
2d+12h	0.218496	0.204278	0.195143	0.198752	-0.014218	Current
3d	0.211906	0.204528	0.192292	0.188543	-0.007378	Gold

RAW AND CALIBRATED DETAIL

Final diagnostic tables

Positive skill means calibrated Brier is below the neutral 0.25 reference. The highlighted row is the lowest calibrated Brier within that model and origin domain.

FINAL-DATE DIAGNOSTIC

T+1 rollout model, completed origin

Step-3000 T+1 rollout candidate at a completed four-hour boundary.

Strongest calibrated horizon2d+8hBrier 0.210894

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.285314	0.251939	0.250000	-0.001939
8h	0.271709	0.254757	0.250000	-0.004757
12h	0.286363	0.251298	0.250000	-0.001298
1d	0.303551	0.267300	0.250000	-0.017300
1d+4h	0.303498	0.264405	0.250000	-0.014405
1d+8h	0.293327	0.261562	0.250000	-0.011562
1d+12h	0.229566	0.235688	0.250000	+0.014312
2d	0.228352	0.231239	0.250000	+0.018761
2d+4h	0.212385	0.221672	0.250000	+0.028328
2d+8h	0.188939	0.210894	0.250000	+0.039106
2d+12h	0.204926	0.215441	0.250000	+0.034559
3d	0.211219	0.214356	0.250000	+0.035644

FINAL-DATE DIAGNOSTIC

T+1 rollout model, developing origin

Step-3000 T+1 rollout candidate at randomized pseudo-live intrabar cutoffs.

Strongest calibrated horizon2d+4hBrier 0.205480

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.266807	0.247055	0.250000	+0.002945
8h	0.259402	0.243924	0.250000	+0.006076
12h	0.275481	0.251110	0.250000	-0.001110
1d	0.300246	0.262002	0.250000	-0.012002
1d+4h	0.314233	0.266335	0.250000	-0.016335
1d+8h	0.255073	0.241568	0.250000	+0.008432
1d+12h	0.219762	0.226458	0.250000	+0.023542
2d	0.194147	0.215963	0.250000	+0.034037
2d+4h	0.164410	0.205480	0.250000	+0.044520
2d+8h	0.176192	0.209734	0.250000	+0.040266
2d+12h	0.187887	0.208444	0.250000	+0.041556
3d	0.213950	0.211906	0.250000	+0.038094

FINAL-DATE DIAGNOSTIC

No-rollout model, completed origin

Step-5000 ablation trained without the optional T+1 sampling rollout objective.

Strongest calibrated horizon2d+8hBrier 0.196498

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.302640	0.251203	0.250000	-0.001203
8h	0.311519	0.266423	0.250000	-0.016423
12h	0.324318	0.267536	0.250000	-0.017536
1d	0.330655	0.281192	0.250000	-0.031192
1d+4h	0.319569	0.270857	0.250000	-0.020857
1d+8h	0.295038	0.262744	0.250000	-0.012744
1d+12h	0.239383	0.233643	0.250000	+0.016357
2d	0.253626	0.233955	0.250000	+0.016045
2d+4h	0.223986	0.215357	0.250000	+0.034643
2d+8h	0.187058	0.196498	0.250000	+0.053502
2d+12h	0.202665	0.200500	0.250000	+0.049500
3d	0.212886	0.200990	0.250000	+0.049010

FINAL-DATE DIAGNOSTIC

No-rollout model, developing origin

Step-5000 ablation at randomized pseudo-live intrabar cutoffs.

Strongest calibrated horizon2d+4hBrier 0.187551

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.285626	0.249449	0.250000	+0.000551
8h	0.299231	0.253078	0.250000	-0.003078
12h	0.303567	0.265278	0.250000	-0.015278
1d	0.320265	0.272518	0.250000	-0.022518
1d+4h	0.335595	0.277375	0.250000	-0.027375
1d+8h	0.260332	0.240157	0.250000	+0.009843
1d+12h	0.227872	0.221528	0.250000	+0.028472
2d	0.199556	0.206291	0.250000	+0.043709
2d+4h	0.156972	0.187551	0.250000	+0.062449
2d+8h	0.176484	0.195524	0.250000	+0.054476
2d+12h	0.196087	0.197133	0.250000	+0.052867
3d	0.222983	0.204528	0.250000	+0.045472

FINAL-DATE DIAGNOSTIC

Current model, completed origin

Step-5000 checkpoint, cached four-hour boundary stream.

Strongest calibrated horizon2d+8hBrier 0.188396

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.307002	0.253717	0.250000	-0.003717
8h	0.385851	0.292083	0.250000	-0.042083
12h	0.371667	0.278549	0.250000	-0.028549
1d	0.370865	0.289545	0.250000	-0.039545
1d+4h	0.378351	0.283354	0.250000	-0.033354
1d+8h	0.324795	0.261069	0.250000	-0.011069
1d+12h	0.276246	0.233405	0.250000	+0.016595
2d	0.288620	0.231881	0.250000	+0.018119
2d+4h	0.243058	0.205453	0.250000	+0.044547
2d+8h	0.201004	0.188396	0.250000	+0.061604
2d+12h	0.213970	0.191650	0.250000	+0.058350
3d	0.229154	0.194450	0.250000	+0.055550

FINAL-DATE DIAGNOSTIC

Current model, developing origin

Step-5000 checkpoint with learned partial-candle completion.

Strongest calibrated horizon2d+4hBrier 0.186230

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.329963	0.259715	0.250000	-0.009715
8h	0.351610	0.274249	0.250000	-0.024249
12h	0.362084	0.283300	0.250000	-0.033300
1d	0.364272	0.284005	0.250000	-0.034005
1d+4h	0.412162	0.303543	0.250000	-0.053543
1d+8h	0.314398	0.243823	0.250000	+0.006177
1d+12h	0.264051	0.216380	0.250000	+0.033620
2d	0.229374	0.203448	0.250000	+0.046552
2d+4h	0.169753	0.186230	0.250000	+0.063770
2d+8h	0.186344	0.187303	0.250000	+0.062697
2d+12h	0.212625	0.188069	0.250000	+0.061931
3d	0.230698	0.192292	0.250000	+0.057708

FINAL-DATE DIAGNOSTIC

Gold model, completed origin

Preserved step-3750 checkpoint at a completed four-hour boundary.

Strongest calibrated horizon2d+8hBrier 0.191351

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.308201	0.250786	0.250000	-0.000786
8h	0.346834	0.273101	0.250000	-0.023101
12h	0.333505	0.266172	0.250000	-0.016172
1d	0.374365	0.298202	0.250000	-0.048202
1d+4h	0.393002	0.292324	0.250000	-0.042324
1d+8h	0.333303	0.271246	0.250000	-0.021246
1d+12h	0.284393	0.248142	0.250000	+0.001858
2d	0.284203	0.242471	0.250000	+0.007529
2d+4h	0.238819	0.216717	0.250000	+0.033283
2d+8h	0.190387	0.191351	0.250000	+0.058649
2d+12h	0.210265	0.197955	0.250000	+0.052045
3d	0.207936	0.193318	0.250000	+0.056682

FINAL-DATE DIAGNOSTIC

Gold model, developing origin

Preserved legacy checkpoint under its explicit developing-origin policy.

Strongest calibrated horizon3dBrier 0.188543

Horizon	Raw	Calibrated	Neutral	Skill vs neutral
4h	0.344668	0.259811	0.250000	-0.009811
8h	0.317947	0.259697	0.250000	-0.009697
12h	0.350520	0.275366	0.250000	-0.025366
1d	0.375237	0.296399	0.250000	-0.046399
1d+4h	0.443403	0.327233	0.250000	-0.077233
1d+8h	0.349947	0.273171	0.250000	-0.023171
1d+12h	0.285703	0.244567	0.250000	+0.005433
2d	0.230923	0.219705	0.250000	+0.030295
2d+4h	0.165722	0.197613	0.250000	+0.052387
2d+8h	0.186030	0.194922	0.250000	+0.055078
2d+12h	0.204394	0.191988	0.250000	+0.058012
3d	0.212924	0.188543	0.250000	+0.061457

DEVELOPING ORIGINS

Random intrabar coverage

Each symbol and candidate origin receives an independent deterministic cutoff from minute 1 through 239 of its active four-hour slot. This avoids calibrating the developing head only at slot boundaries.

Assigned cutoffs483

Distinct minutes210 / 239

Observed range1 - 239

Accepted origins21 / 23

Two candidate origins failed the fixed 90% symbol-coverage floor and were excluded in full. Missing one-minute observations are never synthesized or imputed.

REPRODUCIBILITY

Immutable provenance

Every accepted source file is inventoried and hashed before fitting. Production publications carry the fitted artifact hash and fail closed on an incompatible artifact.

Current/gold calibration v2: e0dfe2f435db647b4dbac55606972811bdda69cc6c0e67281d343c82e29e8d24
T+1 rollout checkpoint: 5c0af6e4ed1991584655909c4fb22818cfc78feefce2d4d802e6351060e942e7
T+1 rollout calibration: e31f10cb1140c2107ae054a493ee20e5adc8ab12af93c9c56b46a0f87522f453
No-rollout checkpoint: 0c36e1a59c84c21830a267852e492e5f64403200c8152ae6194982fa29b2f7c9
No-rollout calibration: d6e3a603c3a0ff61136b4c68d20af5a4a7797edc5ebaf6c1c60d381f0920c856
No-rollout completed corpus: 9b13fdcac4538978df9112a4bb72676c486703ab875b65807006beca9d02a82f
No-rollout developing corpus: 960924107d8a3a9b21bf8c9ffd724ff429741ff20822c1c48febef46a864bd88
All-origin four-model report: 4ebc90b66ae019ecbc35710d54cc52914157cf90c33900cdceedd7cf617bd1a6

Statistical limit: 256 paths reduce Monte Carlo sampling noise for each model distribution. They do not create additional independent market dates, symbols, or regimes. Recursive intervals resample temporal origins with symbols kept inside each origin; calibration uses cluster-balanced scoring by target timestamp and sector.