A retrodiction reconstructs an already known result from fixed, separate inputs. Knowing the target requires extra discipline against after-the-fact fitting. A successful reconstruction checks the application; additional explanatory power requires comparison with a reference model.
These records are mathematical validation examples. They are not new empirical successes or blind tests, since the targets are known.
Historical-revised ε≈1 for discharges is not an established fact here. A test must first specify costs, units, data and uncertainty; a chosen scale is model input.
The philosophical core alone is not yet an experimental model. If resistance can be adjusted after every outcome, nothing can count against the principle. An application must therefore state in advance which outcomes are less likely or impossible.
A counterexample can refute a mathematical theorem. A measurement can challenge a physical model. These are different kinds of testing; a valid algebraic derivation is not experimental confirmation of EEML.
A new commitment should contain at least:
The old site contained over thirty commitments concerning neutrinos, axions, gravitational waves and the absence of new physics, among other topics. That list is not being erased or rewritten as a success. It is archived in full, including its failure criteria, corrections and internal inconsistencies.
As of 16 September 2026 these commitments are not reaffirmed as consequences of the revised core. Their empirical outcomes should still be recorded against their original criteria; this editorial revision does not undo any miss. This page reports no new experimental results.
Complete historical prediction register →
A null result distinguishes models only when the model under study and a relevant competitor assign different expectations to it. The experiment must also be sensitive enough to exclude the effect. Absence of a signal is not the same as absence of a phenomenon.
“Our address then costs more bits” cannot be an unrestricted escape route. The measure and the probability of that outcome must be fixed in advance. A probabilistic prediction allowing rare exceptions needs statistical assessment, not a story selected afterwards.
The next useful checks are a reproducible comparison of cost functions, a normalisation test on a bounded research domain and checking new proof obligations. Negative results should be published too. Specific physical commitments can then rest on a checkable chain of reasoning.
Current status →Open Which partial explanations can jointly describe one consistent reality? Investigate what is fixed, free or incompatible. The route to α through K3 remains a focused subquestion.
Reverse engineering through glue →Reproduction and scope · Example check results
v0.25 · 27 September 2026