The Falsification Dashboard

Every claim in this research programme, its kill-condition, and its live status -
including the one we already killed ourselves.
Michael Darius Eastwood, independent researcher, London: originator of the embedded-correction alignment thesis (manuscript 8 December 2024, SHA-256 anchored: f0d1f38f).
A claim that cannot be killed is not science. This page lists what would prove each claim wrong. If you can trigger a kill-condition, tell us - the retraction goes here, publicly, like the last one did.

The Retraction We Published Against Ourselves

RETRACTED - by our own replication attempt

Claim 12 (original form): sequential super-linearity α≈2.24. The early measurement suggested strongly super-linear scaling of capability with recursive depth. When we attempted cross-architecture replication, the result failed to replicate. The robust estimate is α≈0.49 - sub-linear. We retracted the 2.24 figure, corrected every grant application that carried it, and kept the retraction in the public record.

Why this matters: the methodology self-corrects. The kill-conditions on this page are not decoration - one of them has already fired, and this is what happened when it did.

Live Kill-Conditions

#ClaimWhat would kill itStatus
1Embedded alignment (Eden Protocol): correction must live inside the recursive loop, not outside itDemonstration of a purely external oversight mechanism that remains sufficient as capability scales - i.e., a constructive refutation of the undecidability results (arXiv:2606.28639) and of the alignment-faking failure mode (arXiv:2412.14093)OPEN - strengthened by two independent proofs
2ARC Principle (U=I×Rα): recursion form determines scaling regimeSystematic measurements showing scaling regime is independent of recursion form (sequential vs parallel) across architecturesOPEN - directionally supported (Sharma & Chopra 95.6%)
3Blinding sign-flip: unblinded AI evaluation can reverse a result's signReplications under the published four-layer protocol showing no evaluator-family effect on sign across modelsOPEN - single-lab; replication package in preparation
4Cauchy unification of d/(d+1) derivationsA continuous solution to the multiplicative functional equation that is not a power law (mathematically foreclosed), or empirical domains systematically violating the classification (6/25 already recorded as non-matching - published)PARTIAL - 19/25 domains confirm (p≈1.6×10⁻⁵); the 6 misses are published, not hidden
5Stewardship Gene: stakeholder-care prompts measurably improve ethical reasoningBlinded replications showing no effect (Fisher-combined result currently p≈6.3×10⁻²¹, single-lab)OPEN - needs independent replication
9Parallel recursion does not compound (αpar≈0)Demonstration of super-linear capability growth from pure parallel sampling at fixed computeOPEN - consistent with Brown et al. (2024)
12Sequential super-linearity (α>1)ALREADY FIRED for the original claim - cross-architecture replication failedRETRACTED α≈2.24 → corrected to α≈0.49
13Embedded safety at zero capability costReplications showing consistent capability degradation from embedded correctionOPEN - held in tested configurations; single-lab
15External alignment cannot scaleA scalable external verification scheme - now foreclosed in the limit by the Soundness–Completeness–Tractability Trilemma (arXiv:2606.28639) unless that proof is refutedSTRENGTHENED - independent formal proof reached the same conclusion
16Recursion as cross-domain structural principleThe 19-row convergence register failing source verification, or the pattern failing to appear in new domains where the framework predicts itOPEN - 19 sourced convergences; register public
17HRIH creation cosmologyExplicitly speculative and non-empirical by design; five in-principle falsifiers listed in the paper (e.g. demonstration that recursive intelligence cannot in principle influence physical constants)SPECULATIVE - flagged as such; carries no empirical weight in the programme
18Caretaker Doping (hardware-embedded ethics, TRL 0-1)Proof that substrate-level enforcement is physically or economically infeasible - currently moving the OTHER way (Odeh patents US 2026/0010411/0780; Petrie arXiv:2509.07637; FlexHEG arXiv:2506.15093)OPEN - direction independently emerging in hardware security
β>kCo-scaling stability condition (Paper X): recursive self-improvement is stable iff correction out-scales driftA real system showing stable recursive self-improvement with decoupled (non-co-scaling) correction, or drift-free scaling without any correctionOPEN - first real-model signal consistent (decoupled hacks, coupled clean); needs multi-model sweep (the £624k programme)
Also on the record: Paper VIII reports that two of its three experiments produced null results - published, not buried. Paper IX §7 ("What the Programme Got Wrong") catalogues every error found to date. The metabolic-scaling figures in Paper III are currently flagged RECOMPUTE_REQUIRED in our own canonical-facts file and are excluded from external claims until re-derived.

How to Attack This Programme

Seriously - here is the target list. (1) Run the four-layer blinding protocol on your own models and publish the signs. (2) Check the 19 convergence rows against their primary sources. (3) Recompute the .eml SHA-256 hashes. (4) Find a document predating December 8, 2024 containing the control-failure thesis + embedded-alignment solution + recursion equation together. (5) Build a decoupled recursive self-improver that stays clean. Any of these lands, we publish it here.