When history helps prediction
In finite partially observed stochastic processes, when does recursively updated history recover predictive distinctions unavailable in current observation?
Cases and controls
E0: present sufficient · E1: history disambiguates · E2: history is future-irrelevant
Interpretation limit
Does not establish universal or fundamental memory, or full-history retention.
Execution status
Calibration, fresh variance selection and the separately authorized second confirmation completed. All 60 units and seven frozen decisions replay; the first incomplete attempt remains preserved. The scoped package is CLOSED under D-0004 after explicit internal review.
Runs, results and deviations
Package disposition - completed
D-0004 accepts the defined finite commitment after internal review. RP002A is CLOSED; Q-3 and Q-4 remain ACTIVE. D-0004 resolves the RP002A disposition obligation; D-0005 separately resolves RP003, D-0006 resolves RP004 and D-0007 resolves RP005A. The indeterminate B2-versus-B1 outcome remains unchanged.
Second confirmation completed and audited
60 units, 120 learned fits and 484 hashed outputs. Two benefit, four equivalent and one indeterminate decision under the frozen .01-nat rules and conditional normal-theory assumptions. No trace warnings, broad Claim or Cycle closure.
First confirmation attempt — incomplete
36/60 complete units and 73 saved fit traces are preserved and audited. A clock incident crossed the four-hour wall allowance; no aggregate decisions for that attempt. A separately authorized second attempt is documented independently.
First exploratory pilot completed
One bounded engineering pilot completed with verified artifacts. No scientific hypothesis decision or Cycle resolution.
Read pilot review and raw-artifact linksConvergence stage completed
Twenty independent units and 40 learned-model fits completed with verified artifacts. Many validation traces improved near the epoch cap; settled optimization is not established.
Read the exploratory stage reviewThe separately approved longer-budget stage also completed: 40 fits in about 6 minutes 24 seconds. Twelve near-cap warnings remain.
Read the frozen longer-budget protocolRead the longer-budget stage review
Review the remaining budget warnings
Read the frozen target-data-size feasibility protocol
Six target-size fits completed in 5 minutes 48 seconds with no predeclared warning. One replicate per world supplies no variance or confirmatory conclusion.
Read the feasibility review