Best papers

The research record ranked by scientific interest: how strong the test design is, how much rides on the claim, and how honestly the failures are reported. Not ranked by confidence; several of the best entries here are negative results.

Source and date. Ranking, descriptions and the counts quoted inside each card are as written by Kimi (Moonshot AI) for the Neural Nations layout prototype, 19 November 2026, and are presented as written apart from punctuation and wording swaps to the site’s CAS lexicon. Check the linked page for current figures. Links point to the live pages on this site.

Prospective and pre-registered tests

The only tests that can’t be gamed retrospectively. Ranked first because the design does the defending.

1

PR-GAP-2: CAMS as a Leading Indicator of Societal Rupture

Pre-registeredFinding Updated Nov 2026 · 44 societies, 1900–2025, 126 ruptures

Leave-one-society-out cross-validation reaches AUC 0.740 for deficit crises once coordination transitions are correctly separated: using only structural coordination metrics, no political or economic covariates.

Why first: the strongest single empirical claim on the site: genuine out-of-sample structure, an explicit confound-handling move, and a headline number modest enough to believe. Read beside the full paper next.

2

PR-GAP-2: Structural Coordination and Rupture Morphology, Full Paper

Pre-registeredValidation Aug 2026 draft · with findings summary and six supporting analyses

The full academic paper behind the rupture study: panel validation across 44 societies, Robson Gauge cultural-grounding results, and a cross-system parsimony assessment. Supported by the findings note, a synthesis, a formalism audit, an NCT/LOSO report, a Robson-plane analysis and a general-audience version.

Why second: the closest thing CAMS has to a flagship manuscript, and it carries its own audit trail in public.

3

ADVERSARY-02 Results: a Pre-registered Test of CAMS

Pre-registered 6 Nov 2026 · 15 predictions, China + 7 firms, 2021–2024

Fifteen predictions sealed before any scoring: six passed, nine failed, and the built-in kill-switch (FC5) was not triggered. Written for new readers, with the scoreboard, scorer grades and limitations.

Why third: the cleanest demonstration of the project’s epistemic discipline, a 40% hit rate published with the same prominence as a win would get. That is what pre-registration is for.

4

Persistent National Topology in Coded Institutional Profiles

Pre-registeredBlind rounds 15 Nov 2026 · four blind rounds, 24 eras

Four pre-registered blind rounds asking whether a national signature persists in CAMS 8×4 profiles across 24 eras. Crisis-swamping and single-carrier explanations are falsified; Helm is flagged as σ’s blind spot.

Why fourth: a falsification-oriented design that reports what didn’t work and names the instrument’s own weak node, rarer and more valuable than another confirmation.

5

Predictions on Record: the 2027–2028 Register

Pre-registered Lodged Aug 2026 · 32 claims · scored when the years close

Thirty-two probability-weighted claims about the USA, China and Australia for 2027–2028, lodged at Zenodo and frozen. Currently the strongest reason to wait before judging the framework.

Why fifth: not a result but the decisive pending experiment, the entry that turns every other ranking on this page into a provisional one.

Validation studies and blind rounds

Does the instrument measure what it claims? Ranked by design quality and by how much of the framework each one stress-tests.

6

Graph-Theoretic Diagnosis of Societal Collapse: CAMS Validation Across 18 Societies

Validation 10 Jun 2026 · 10–2026 CE · 12 falsifiable predictions

Tests three core CAMS claims against 18 societies spanning 2,000 years: dimensional collapse, node coupling architectures, and crisis-signature matching, with 12 predictions stated up front.

Why sixth: the broadest historical validation and an early statement of the falsification culture that later work sharpened.

7

CAMS Temporal Inelasticity: Results

Validation 15 Aug 2026 · internal audit · Cuba, Iceland, ASPI pilots

Shows the JUNO v1.2 bond formula factors completely into node-level weights: B = w⊗w is rank-1, so the bond matrix carries no pair-specific information and λ₂ is algebraically constrained rather than empirically discovered.

Why seventh, and arguably first in importance: the most consequential single result on the site; it defunds an entire family of “network” claims. A model auditing itself this hard earns the right to be read.

8

JUNO/CAMS Blind-Analysis Protocol v1.1.1

Blind rounds Nov 2026 · 24 societies, 3 blind sets

Results of the cognitive-affective extension experiment: MRR improved 0.667 → 0.833, but the calibration gate failed on conflict states.

Why eighth: an honest mixed result that marks its own boundary, exactly where a blind protocol earns trust.

9

JUNO/CAMS Blind Analysis Challenge

Blind rounds Open now · 8 anonymised societies, 1750–2026

The open challenge: load an anonymised dataset and the protocol prompt into any capable LLM, run the five-stage analysis, and compare against the sealed key before unblinding.

Why ninth: not a finding but an invitation, the entry point for any outsider who wants the strongest form of evidence the project can offer: their own run.

10

An Accumulating Case for CAMS: the Personality Experiment

Blind rounds 7 Nov 2026 · recognition counts start closed

A possible unification of meaning and metabolism, offered with the mechanism explicitly not treated as demonstrated: and a participant route that keeps the November grades undisclosed so you can take part clean.

Why tenth: the most methodologically careful piece of speculative work on the site: it builds a disclosure firewall into its own design. The speculation is labelled speculation.

11

Ensemble Validation: CAMS v2.3

Validation May 2026 · Germany & USA 1880–2026

The Wright Thesis confirmed with qualifications: variance reduction, SNR improvement, contestation index and κ-uncertainty bounds for ensemble scoring.

Why eleventh: the measurement-theory foundation under every score the site publishes, unglamorous and load-bearing.

12

Operator Portability in CAMS: Synthesis & Verification Trail

Validation Nov 2026 · two independent computational audits

Canonical operators are structurally portable across corporate and national domains but parametrically non-invariant: confirmed twice independently. The companion Soil-Bond Index paper is explicitly paused pending rework.

Why twelfth: the cross-scale claim tested with replication, plus a rare public example of a result being stood down rather than quietly left standing.

13

P4 Retrospective Validation: Does the Crisis Detector Work on Firms?

Validation 10 Nov 2026 · interim · 3 cases, 6 matched controls

A rule sealed on fifteen national crises was run unchanged on Boeing, Theranos and WorldCom against matched survivors. The controls set it off almost as often as the cases.

Why thirteenth: a clean negative that bounds the portability claim, published as prominently as the positives. This is the audit culture working.

15

CAMS v1.0.1: Instrument Test Battery

Validation Jun 2026 · 2,532 society-years verified

The full test battery for the v1.0.1 instrument: corpus verification, JUNO formalism audit, blind scoring control, and the first pre-registered prospective predictions.

Fifteenth: superseded in parts by later versions but documents the instrument’s quality bar at the time, useful for tracing what changed and why.

16

Carbon’s Evaluation of CAMS

Review 11 Nov 2026 · external AI evaluation, checks included

An outside AI system reads the synthesis and the papers and judges the scientific value; the report is presented as written, followed by Neural Nations’ own claim-by-claim checks.

Why sixteenth: the nearest thing to peer review the project currently has, and the meta-check format is more honest than most human review cycles.

Findings from the datasets

What the numbers say, once the instrument’s limits are on the table. In-sample unless marked; the distinction is marked throughout.

17

The Synchronised Slide: Global Stress After 2000

Finding 23 Oct 2026 · JUNO v3 common-factor analysis · external series: IMF, World Bank, NASA GISTEMP, FRED

Stress synchrony jumps after 2000 to war-era levels without a world war, carried by a new near-uniform global factor. Of the drivers tested, only the public-debt build-up tracks it; the 1920s–30s parallel arrives with the rate sign inverted.

Why seventeenth: the most scientifically interesting descriptive finding, an unexpected, testable macro claim with the comparison series named, and the inverted-sign parallel flagged rather than smoothed over.

18

Organisational Pathologies: Interim Results

Finding 20 Oct 2026 · six frozen signatures · 37 JUNO v3 societies

The failures we recognise in families and offices recur at national scale: twelve of eighteen named historical windows hold, all four windowed signatures cluster beyond chance, and Late Abstraction Collapse precedes big falls.

Why eighteenth: the cross-scale claim that would matter most if it survives replication, and “twelve of eighteen” is reported as twelve of eighteen.

19

US Legitimacy Crisis × CAMS Stress: Study 2

Finding Nov 2026 · 1996–2025 panel

CAMS Stress leads Gallup and Pew institutional-trust series by 2–3 years: the exact opposite of the fiscal finding in Study 1.

Why nineteenth: a lead-lag result against external survey data is genuinely testable, and the open contradiction with Study 1 is the interesting part; two probes, two answers, both published.

20

Attractor Coherence, Revisited

Finding 9 Nov 2026 · 53 societies, 47,960 node-years · labelled: consistency check, not a blind test

An old experiment re-run on the combined JUNO dataset after its first blind prospective test failed: do societies break in the same places every time?

Why twentieth: valuable as candour, a failed replication revisited on fuller data, with the “not blind” label worn on the sleeve.

21

Robodebt and the Australian State

Pre-registered 14 Nov 2026 · retrospective · scorers not blind: flagged on the page

A question register timestamped before any score was examined, run by Kimi against three hash-frozen CAMS panels with three-scorer ensembles; presented as written, with checks against the score files.

Why twenty-first: a well-constructed register study of a live Australian wound; its own authors downgrade it (“retrospective, scorers not blind”), the ranking follows their label.

Theory, specifications and synthesis

The scaffolding: canonical definitions, formalisms and the current statement of the framework. Reference material more than findings.

22

Canonical Definitions + Bond Strength spec

Theory & specs Oct / May 2026 · the authoritative reference

Authoritative definitions of the four metrics, Node Value, and the derived composites (σ, bond strength, Φ, 𝒫(t)/𝒴(t)): with the bond-strength formula, scale, interpretation and version history.

Why twenty-second: every claim on this page reduces to these definitions; they are the contract between the project and its critics. Low drama, highest dependency.

23

JUNO v1.2-Final: Production Specification

Theory & specs Nov 2026 · follows the 3,002 society-year blind-corpus test

Regime classifier, numerical policy and reference ranges for the production formalism: the current scoring machinery behind the datasets.

Why twenty-third: the spec that operationalises everything; read after the Inelasticity audit, which limits what its network quantities can mean.

24

The Natural History of Civilisations + the book

Theory & specs 8 / 31 Oct 2026 · master essay · CC BY 4.0

The current statement of the framework: an alien naturalist, eight functions, culture as flower, meaning and metabolism, with the supporting document linking each claim to its evidence, and the book-length personal account with free PDF/EPUB.

Why twenty-fourth: the best-written prose on the site and the frame everything else hangs in, but as science it is the hypothesis, not the test, so it stands behind the tests in this ranking.

25

Why Eight Nodes? A Rate-Separation and Coupling Argument (Mar 2026 draft, in the book’s research thread)

Theory & specs Mar 2026 draft · first-principles derivation

A derivation of the 8-node architecture from multi-timescale coordination mathematics, introducing the Coordination Phase Transition Theorem and the criticality index 𝒳(t).

Why twenty-fifth: the most ambitious theoretical move, and still a draft. If the derivation survives formal scrutiny it justifies the architecture; until then it is the framework’s biggest open bet.

Review, record and response

The layer that lets outsiders check the work.

26

CAMS Human–AI Scientific Record

Review 30 Oct 2026 · Zenodo DOI · checked with Claude Opus 5.5

Preservation package with hash verification: all 13 files match, and the blind-rater agreement (r 0.89–0.93, ICC 0.869) reproduces from the raw CSVs.

Why twenty-sixth: reproducibility infrastructure rather than a result, but it is what makes every other ranking on this page checkable.

27

Blind Testing & Validation: the hub

Review Updated 13 Nov 2026 · the honest front door

“CAMS is not externally validated. The out-of-sample holdout is locked and unscored. Everything here is in-sample or pre-registered: the distinction matters and is marked throughout.” Counts, challenges and categories on one page.

Why twenty-seventh: the single best page for a scientist arriving cold; in this layout its candour is the standard every other card is ranked against.

28

Open Validation Invite + Response to Critics

Review 2026 · standing invitation

The standing invitation to scrutinise the framework, and the collected responses to published criticism.

Why twenty-eighth: process pages, necessary, and they close the loop between claim and scrutiny.

Reading order for a scientist in a hurry: Blind Testing hub → ADVERSARY-02 → Temporal Inelasticity → PR-GAP-2 rupture study → Synchronised Slide → the 2027–2028 register. Ninety minutes, and you will know exactly how much of the rest to believe.