Ozone Archive evidence audit: 11 November 2026
Version 1.0.0. Evidence review and research prioritization, not a new experiment.
The live archive was fetched directly on 2026-11-11 at about 15:51 UTC.
Finding
At the audit snapshot, the site had 109 research records. It contains useful computational work and
explicit negative findings. It does not provide scientific proof that O/0 is
a physical or metaphysical phenomenon. Its own scope documents exclude that
claim. A publishable next step is to test the strongest remaining operational
result against conventional alternatives and independent observations.
Coverage and limits of this audit
Inventoried the complete live catalog and compared its 109 IDs with the local
corpus. Read the scope/claim registry, program charter, contact-and-revelation
program, recent CRP-034 through CRP-038 records, T1, T2 and T2-R2a methods,
and the relevant executable training and metric code. This is a targeted
methodological audit, not an independent reproduction of all 109 records or
a systematic review of the external literature. The preserved live catalog
and audit manifest identify the snapshot. Search-engine copies were stale
(63/90 records); the direct live response was used instead.
What has been published
- The catalog counts records, not independent experiments: 28 SIMULATION,
14 SIMULATION STUDY, 13 REPLICATION, 14 FORMAL MODEL, 11 CHARTER, six
SCIENTIFIC PAPER, six EXPERIMENT, five SYNTHESIS, four HYPOTHESIS, three
ANALYSIS, two CRITIQUE, and one each CLAIM AUDIT, NULL RESULT and SYNTHESIS
PAPER. Some paper records wrap existing studies and do not add observations.
- Boundary formation, synchronization, information measures, observer models,
matched-source discrimination and compression are established as research
topics. Positive computational verdicts concern the particular models and
decision rules, rather than the physical truth of the motivating framework.
- Negative and corrective results are already present. STUDY-002 and CRP-021/
CRP-029 do not support the claimed prediction benefit of functional separation.
CRP-032 corrects the earlier chaos-washout interpretation under better
measurement. CRP-035 preserves correction of a context-dependence bug.
- CRP-034 is the only catalog entry reporting physical measurements: one
five-minute ECG recording from one patient. It discriminates windows and
controls within that recording; it does not compare independent biological
source ontologies or measure O/0 experiences.
- T1's overall verdict is INCONCLUSIVE, with an LTI-specific manifold result
linked to conventional unobservable-state freedom; its R6 baseline failed.
- T2's original aggregate hypothesis is UNSUPPORTED. Its later 15-seed follow-up
reports a strict-signature rate of 9/15 for A3 on D1 beta=128 versus 0/15
for a linear null, and a linearly-hidden signature of 10/15 on D3 beta=8
versus 3/15. These motivate further testing; they are not physical validation.
Methodological issues that affect interpretation
1. T2's mean report bypasses stochastic channel noise. In metrics.py,
skill_report is measured on d(mu(x)), while skill_perception is measured
on mu(x)+sigma(x)*epsilon. Low I(X;Z) constrains information transmitted
through Z, not noise-free access to mu. Reportability from d(mu) therefore
does not by itself establish reportability through a collapsed channel.
2. T2's original linear_probe_accuracy fits ridge weights on the same feature
rows and labels used to score accuracy. This is resubstitution accuracy,
not held-out generalization. R2/R2a also calculate split-set scores, which
is an improvement, but their original H_STRICT count retains full-set
scores. The historical results remain intact; the new test uses held-out
observations throughout.
3. The historical A3 latent dimension is four while the linear null uses two.
Their difference does not isolate nonlinear architecture from latent width.
The new study matches latent dimension; total parameter counts still differ.
4. More model seeds on the same dataset quantify initialization sensitivity,
not robustness across independently generated data distributions. Fresh
dataset draws are needed as experimental units.
5. A classifier AUC inside [0.30,0.70] is not an equivalence test or proof that
two physical explanations are identical. Matched-source studies establish
discrimination performance under particular constructions and measurements.
The duplicated-input AUC inversion reported by CRP-034 also warrants a
split-design audit before attributing it to an implementation bug.
6. The public T2 manifest names execution files under _internal; that directory
is excluded by the production build. Its listed src files are not inside
the T2 public package locally. A self-contained download is needed for the
next result. Internal reanalysis and architecture variants should not be
described as independent laboratory replication.
7. The July claim registry predates the latest T-series work. It needs an
eventual versioned update to cover later claims. This audit adds a dated
account without silently rewriting historical conclusions.
Selected continuation
T2-CHANNEL-001 directly tests whether strong reporting survives passage through
the single-sample stochastic bottleneck. It freezes 30 independent datasets,
matched-latent nonlinear and linear models, 72 fits, held-out classifiers,
analytic KL bounds, negative controls, positive controls and decision rules.
All execution sources are public inside the study package. Its local hash lock
is prospective but is not an external preregistration service.
The result must be published even if the mean-path explanation wins. Such a
result narrows the original interpretation and gives a better-founded next
question; it does not refute every possible operationalization of O/0.
What would count as stronger evidence
A physical or human-experience claim needs a precise observable that differs
between O/0 and a named conventional account, a prospective protocol, controls,
appropriate participant procedures when people are studied, and independent
replication. A simulated observer is not evidence of a person's experience.
An attribute-free source with no differing observable predictions cannot be
identified by this program's measurements. Mathematical analogies in the source
document are not physical demonstrations; a shared symbol or a singular limit
does not establish equality of zero and infinity.
Conceptual provenance is not empirical support. This audit evaluates the
available evidence without assuming the framework's truth or falsehood.
Sources
- Live catalog: https://ozonearchive.online/science/catalog.json
- Scope: https://ozonearchive.online/science/O0-SCOPE-001.html
- Claim registry: https://ozonearchive.online/science/O0-SCOPE-002.html
- ECG: https://ozonearchive.online/science/O0-CRP-034.html
- T1: https://ozonearchive.online/science/T1-EMPTY-THRONE.html
- T2: https://ozonearchive.online/science/T2-NETI-NETI.html
- Variational information bounds: https://arxiv.org/abs/1612.00410
- Evaluation methodology: https://scikit-learn.org/stable/common_pitfalls.html