Reproduce
Completed · STOP documented
The maximum-likelihood benchmarks are reproduced. The baseline evidence test ended at its preregistered STOP.
Study and status →Research focus · Lane–Seifert
Same data.
Different evidence.
From reproducing the published Pantheon+ analyses to a planned prediction test on DES-Dovekie: our four-stage programme asks how strongly supernova data support timescape and how much that depends on the analysis.
Research: Harald Zierhut & Thomas Schwarz
Scientific responsibility: Harald Zierhut · Technical execution: Thomas Schwarz
Timescape × ΛCDMPantheon+ → DES-DovekiePreregistered comparison
One programme · Four connected stages
Reproduce. Compare.
Understand. Test.
The four-stage framework was designed from the outset: reproduce, compare, understand, then test on further data. Each detailed study builds on the completed preceding results. The aim is to establish how strongly supernova data support timescape and how much that conclusion depends on the analysis, whatever the outcome.
Compare
Completed · Results published
With identical P+1690 data and common priors, changing the implemented likelihood raises ln B by 1.89 nat.
Study and status →Understand
Preregistered · 31 October 2026
Strand 3 tests the assumptions behind the shift, including a ΛCDM false-positive control. No results yet.
Study and status →Test
Plan published · 1 November 2026
Strand 4 is planned to test a prediction learned from Strand 3 on DES-Dovekie. The programme outline is public; no results yet.
Study and status →Published finding: a controlled comparison of our implementations on P+1690 at zCMB ≥ 0.075. Strand 4 is planned, not yet preregistered.
Finding, limits and sources →The research programme behind it
Prephysics: Which claim does the evidence warrant?
A calculation can be correct and a measurement precise. Prephysics asks whether the conclusion drawn is warranted by its assumptions, model and available evidence.
Understand the method →In 30 seconds
Three questions create the structure
What must be settled first?
Is the target clear, is the theory connected to observation, and are the assumptions that matter made explicit?
Scientific foundation: P1What warrants the inferential move?
When an argument moves among preconditions, theory, and observation, is that transition itself justified?
Scientific foundation: P2Does the evidence state warrant this exact claim?
A concrete, dated claim is tested for any remaining applicability-relevant deficit.
Scientific foundation: P3What a result means
Pass, open — or a genuine boundary
Each result belongs to one concrete claim at one evidence state. It is not a blanket verdict on a theory.
Pass PASS
No applicability-relevant deficit remains for the tested claim under the fixed conditions.
Not the same as “true”.Open OPEN
Something material remains unresolved, while impossibility has not been established.
Not the same as “false”.Loss of Applicability LoA
The strongest result requires positive evidence that a necessary applicability boundary has been reached.
An unresolved problem is not enough.Six public-level runs
What the method has shown so far
The public-level balance is three OPEN, three PASS, and zero LoA. A seventh frozen run, GW-P (OPEN), is documented on the GW150914 case page as expert-level depth and is not included in this public-level balance. These cases demonstrate how the method works; they are not yet a statistical performance validation.
Featured test case · GW150914 · LIGO
One famous event, two different claims
The detection claim is PASS; the stronger categorical source attribution is OPEN at the same 11 February 2016 evidence state. The pair shows why detecting a signal and identifying its source are different inferential tasks.
02
→
03
→
04
→
2014 open · 2015 pass
BICEP2
Two different claims at two evidence states show why later evidence creates a new record rather than rewriting the old one.
Open
Quantum contextuality
The experiment was strong, but full control of measurement disturbance for the exclusion claim remained incomplete.
Pass
Muon g−2 2021
The 4.2σ tension holds — but only relative to the explicitly fixed theory benchmark.
Additional documentary case · outside the public-level 3/3/0 balance
Galactic Centre Excess
A communication case: how a class- and assumption-relative result can appear stronger in the citable surface.
Scientific foundation
Three papers, one connected task
Conditions
What requirements must a physical claim satisfy in the first place?
Paper ↗Transitions
What additional warrant is owed when an argument moves among preconditions, theory, and observation?
Paper ↗Concrete test
Are those requirements met for a specific, versioned claim at a specific time?
Paper ↗P9 · Functional Incompleteness Argument (FIA). It identifies three kinds of information that must be recoverable and cannot substitute for one another in a complete, auditable applicability diagnosis: requirements, transition warrants, and token satisfaction. P1 → P2 → P3 is one concrete realisation of this Applicability-Diagnostic Function. P9 remains the separate necessity and pattern anchor, not a fourth verdict layer.