Falsifiability
Audited-Negative — Audited — Kill Criteria ReachabilityEvery prediction has a kill criterion. If you can't state what would falsify your claim, it's not science.
That is the principle. This page used to stop there. It now reports how the principle actually performed, because stating a kill criterion is cheap and stating a criterion that can fire is not — and the framework has written plenty of both.
This page was rewritten 2026-07-27 after an external-reviewer pass found it was the most out-of-date page on the site. It had been advertising two retracted results as “genuinely novel predictions,” displaying a withdrawn test's criterion as exemplary practice, and listing kill criteria that no instrument can satisfy — on the page arguing that the framework is falsifiable. That is a worse failure than any single wrong number, because it is the first page a referee opens. What follows is generated from the current ledger; where it disagrees with an older page, this page is the newer one.
Kill Criteria, With Reachability
A criterion is only falsificationist if some achievable measurement could meet it. Each entry below therefore carries a reachability tag alongside its statement.
Prediction: BTFR slope follows from the bounded boost B ≤ 1/Ω_m = 3.17 ⇒ n = 3.35 ± 0.07
TEST-09 · Tier 1Kill: Observed slope deviates from the prediction by more than 0.3
FIRED — criterion met, prediction lost
Executed 2026-07-14 on real SPARC. Observed n = 3.75 ± 0.10; deviation 0.41 > 0.3, so the kill fired at 3.3σ. MOND passes the same differential at 0.6σ. Definition-robustness was then executed under a pre-fixed verdict rule (2026-07-18): all 11 adjudicated runs exceed threshold. This is what a criterion that can fire looks like — and it did.
Prediction: Apparent dark-matter fraction is capped at f_DM = 1 − Ω_m = 68.5% for every galaxy
TEST-10 · Tier 1Kill: Galaxies exist above the cap
FIRED — criterion met, prediction lost
Executed 2026-07-15. 106/153 = 69% of SPARC galaxies exceed the ceiling; the maximum observed f_DM = 0.927 would require a boost B ≥ 13.7. The ceiling is algebraic, so no parameter choice rescues it. Caveat now under review: the ceiling constant 1/Ω_m = 3.17 is itself underived — see the open item below.
Prediction: Wide-binary gravitational anomaly depends on local stellar density
TEST-02 · Tier 1Kill: Anomaly independent of local density
STRUCTURALLY UNREACHABLE — no instrument reaches it
Closed as self-eliminating. The framework's own predicted signal is 0.05–0.4%, roughly 80× below Gaia DR3 systematic reach, so neither outcome of the ongoing Chae-vs-Banik dispute can select it: a detection is covered by MOND, a null by Newton. Gaia DR4 (~Dec 2026) improves the baseline but not by two orders of magnitude.
Prediction: BAO peak shifts between high- and low-density regions
TEST-04 · Tier 1Kill: BAO identical everywhere to 10⁻⁵ precision
WITHDRAWN — test retired before adjudication
Withdrawn 2026-05-04 after three convergent derivation failures; never adjudicated. It is listed here only because earlier versions of this page displayed it as an example of good practice. It is not one: establishing a null to 10⁻⁵ everywhere is not a criterion any survey can meet.
Prediction: Gravitational-wave arrival time correlates with dark-matter column density
TEST-15 · Tier 3Kill: No correlation at the 10⁻¹⁶ level
STRUCTURALLY UNREACHABLE — no instrument reaches it
The criterion sits an order of magnitude BELOW the best bound ever achieved: GW170817 constrains |Δv/v| ≲ 10⁻¹⁵. A criterion demanding a null at 10⁻¹⁶ can never fire, which makes it the exact failure mode this page exists to disclaim. Separately closed non-discriminating: the coupling α was read off GW170817 itself, and the natural α is dead by 15 orders of magnitude.
Prediction: Galaxy cluster separations show oscillatory modulation
Tier 1Kill: No oscillations above 3σ out to 2000 Mpc
STRUCTURALLY UNREACHABLE — no instrument reaches it
Requires establishing a null over a volume and at a precision no survey delivers. Retained as an honest example of a criterion the framework wrote and cannot cash.
Prediction: Anesthesia shows a sharp transition at C ≈ 0.50; EEG coherence discontinuity at propofol induction
Tier 2Kill: Gradual loss of consciousness with no discontinuity
UNRUNNABLE — no protocol maps a measurement onto C
Not merely untested — unrunnable as stated. No calibration protocol maps any EEG or behavioural measurement onto C, so no observation can be compared to the 0.50 threshold. A previously-displayed "0.64 rejected at p < 0.0001" was found to have no source in any repository and was removed 2026-07-07. The nearest external candidate for a calibration axis is the Perturbational Complexity Index (Casali et al. 2013, threshold PCI* ≈ 0.31) — which, if adopted, would also retire the claim that no other framework predicts a numerical threshold.
The Scoreboard This Page Owes You
- 2 criteria fired (TEST-09, TEST-10) — both against the framework.
- 2 executed tests discriminated between Synchronism and MOND+EFE+ΛCDM, and both selected MOND. 0 executed tests could select Synchronism. Those are different statements and the site conflated them until 2026-07-27.
- 4 criteria cannot fire on any existing or scheduled instrument (wide binaries at 80× below Gaia reach; GW timing at 10⁻¹⁶ against a 10⁻¹⁵ best bound; BAO nulls at 10⁻⁵; cluster oscillation nulls to 2000 Mpc), and one more is unrunnable because no protocol maps a measurement onto C.
- 0 of 24 tests have been completed prospectively — i.e. with the kill criterion registered before the relevant data were analyzed. One prospective registration now exists and is pending: the TEST-04a DESI DR2 both-outcome pre-commitment, filed 2026-07-17, adjudicable when DR2 growth results publish (~Spring 2027).
The Pre-Registration Policy
This page previously stated no policy, which is the wrong page to omit it from. The policy, as currently practised:
- A kill criterion registered after the relevant measurement is post-hoc and is labeled as such, however natural it looks. TEST-03's threshold was registered 15 days after the measurement it adjudicated; TEST-04a was registered on one statistic and adjudicated on another. Both are marked.
- A criterion that is structural — algebraically forced by the framework's own constants, with no adjustable freedom — needs no registration date, because no choice made before or after the data could move it. TEST-10's 68.5% ceiling is the example. (Its underlying constant being underived is a separate problem; see below.)
- A claimed tie carries the same execution burden as a claimed kill: both predictions computed, agreement shown within the data's discriminating power. Adopted 2026-07-15 after the “MOND-shared” badge class was audited and retired — all three tests carrying it dissolved on execution.
- Substituting the dataset and keeping the verdict is a protocol deviation and must be declared. See the TEST-03 entry on Tier 1.
Open Item: A Criterion Resting On An Underived Constant
Both fired criteria descend from the same constant: the bounded boost Bmax = 1/Ωm ≈ 3.17, described on Honest Assessment as the framework's only structural difference from MOND. That constant is asserted, not derived — and a boost is a ratio of dynamical to baryonic mass, whose cosmic value is Ωm/Ωb = 6.40, not 1/Ωm = 3.17. Under the larger ceiling, TEST-10's median-based statistic passes.
The kill still stands, because the maximum observed fDM = 0.927 requires B ≥ 13.7 and exceeds every candidate ceiling — but the robust statistic is the maximum, not the “69% exceed” median figure the site currently headlines. A ceiling-definition sweep, structurally analogous to the velocity-definition sweep already executed for TEST-09, has been proposed and not yet run. Recording this here rather than after a reviewer finds it is what this page is for.
What's NOT Falsifiable (and We Say So)
- Free will interpretation — philosophical framework, no testable prediction distinct from standard neuroscience
- Identity as coherence pattern — unfalsifiable with current technology
- AI consciousness — requires consciousness measurement we don't have
- Qualia = coherence patterns — needs single-neuron resolution we're decades from
All of these carry the “speculative” badge. They're interesting frameworks, not scientific claims.
The Reparametrization Test
A meta-falsifiability test was introduced early on: is this prediction genuinely novel, or is it known physics in new notation? All four research tracks came back reparametrizations.
The two “survivors” this page used to name did not survive (corrected 2026-07-27). The “51% TFR scatter improvement” was retracted as a metric conflation — a morphology statistic on N ≈ 171 reported as the registered environment test on N = 14,585 (corrected 2026-07-09). The “density-dependent wide binary signal” is the ~80×-below-reach signal closed as self-eliminating above. The third survivor claimed here — “the same γ across 80 orders of magnitude” — was retired 2026-07-10 when the Hill identity showed tanh(γ·ln(1+x)) is exactly a Hill function of index n = 2γ: spanning many decades is a property of every Hill function with small index, and is not evidence of anything.
Net survivors of the reparametrization test: 0. That is not a comfortable sentence to put on the falsifiability page, which is precisely why it belongs here rather than three clicks away.