ATOEVO

RPT Evidence Lab: Exact Experiments and Verifiable Results

RΛIN / ATO · Evidence Lab 01

When the relationship
does the work.

Two bits. Three possible repairs. A pair can remove dependence that every single repair leaves behind. Change the budget, inspect every candidate and verify the result yourself.

1 · Configure2 · Enumerate3 · Record4 · Verify

The declared model

History H is a pair of independent fair bits, a and b. The retained summary Z is constant. A mediator M can retain a subset of the three features below.

afirst bit · cost 1
bsecond bit · cost 1
a ⊕ bparity · cost 3/5
Every history has probability 1/4.
HabParity
00000
01011
10101
11110

A repair passes when H and the future F are conditionally independent given M, for every selected intervention. Acquisition cost and information entropy are different quantities.

Between 0 and 3. Fractions work: try 3/2, then 8/5.

The calculation runs on this device. A pending run can resume after you reload this page.

Ready. Run all eight candidates to find the lowest-cost repair.

Candidate decisions within the selected budget
Retained featuresCostWorst residual (bits)Exact decision
No run yet.

Inspect the evidence record and failure witnesses
Run the search to create a record.
Local activity

Only these experiment actions are counted on this page. No analytics service receives them.

    Formal verification / a separate research branch

    A checked algebraic identity

    The pointwise mixing identity underlying T8-02 has been checked with Lean 4.35.0 over a general commutative ring:

    KV((1−e)v + ea) + KA((1−e)a + ev) − (KVv + KAa) = e(KV−KA)(a−v)

    The proof also covers equal-kernel cancellation, equal-density cancellation and preservation of v+a. This certifies the algebraic statements; an integral formulation and physical interpretation require additional assumptions and evidence.

    Evidence, scope and interpretation

    SAT: a completed search found an allowed repair. UNSAT: a completed search found none in this grammar within the budget. UNKNOWN: the evaluation limit stopped the search before completion.

    Decisions use integer cross-products of exact probability weights. The displayed mutual information uses floating-point logarithms and is descriptive. Hashes detect changes to records; they are not signatures or proof of who created a record.

    The model’s intervention laws are supplied by construction. This is not identification of a causal effect from observational data, a physical validation or a novelty claim.

    Explore context repair · Read the research boundaries · Research Atlas