Predictions for 2027 and 2028

A forecast the instrument can fail

CAMS reads a society as eight cooperating functions and scores each on coherence, capacity, stress and abstraction. JUNO turns those scores into a handful of summary quantities and a regime label. These predictions say what those quantities and labels will look like for the United States, China and Australia in 2027 and 2028, with a probability on each claim and a rule fixed in advance for how it will be scored.

It is a thermometer reading projected forward, not a crystal ball. It makes no claim about events, causes or the intentions of any government, and it takes no political side. The scores are model-generated readings of the documentary record, and JUNO v1.2 has not been externally validated. Scoring these predictions is part of the test that would begin to change that.

The eight functions

What we expect, in plain terms

On the overall score, we make no call on whether it rises or falls. The model finds about a coin flip in every society, and the register says so.

Where each dataset puts 2028

Diamond: latest scored value. Dot: 2028 median. Bar: 80% interval. Dashed lines at 4 and 10 are V̄ thresholds in the regime rules; the dotted red line at 0 marks the Freeze/Collapse side. Datasets for the same society differ in level, not just in noise.

Capacity intact, coordination impaired

High stress and a wide gap between capacity and coordination persist. The model has no view on whether the overall score rises or falls, but it expects no return to a “Strained” or better reading by 2028 and expects Shield to stay among the two strongest functions.

“Our probability” is the primary-dataset figure capped to the range 5–95% as a model-risk haircut, so a dataset showing 100% is stated as 95%. Type A: the probability varies by 25 points or less across datasets. Type B: it varies by more. Type C: conditional.

Where the four datasets disagree. The four datasets agree on structure and disagree on level. Shield is in the top two on all four and Lore is among the lowest three at the 2026 base on all four. Helm is the single weakest node on two datasets (MARKER, M5N), second weakest on MFIN and fourth lowest on UNIFIED. 2026 mean V̄ ranges from 0.9 (M5N) to 6.15 (MFIN), and the 2026 regime label is Systemic Crisis (UNIFIED, M5N), Phantom Type II (MARKER) or Local Node Failure (MFIN). On M5N the base sits near the Freeze/Collapse boundary, so that label has a 19–26% probability there and 0–2% on the others. The MARKER envelope is floored at 0.3 and is not a measure of scorer spread; the M5N envelope is.

Agreement on shape, disagreement on level

The datasets disagree about China’s level, so the register separates the claims that should hold whichever dataset is used (the weakest functions are the labouring base and resource allocation; Archive stays strong) from those that depend on it (the regime label).

Where the datasets disagree. The datasets agree on shape and disagree on level. Hands and Stewards sit in the bottom three on every dataset (79–91%) and Archive stays high. UNIFIED reads Stable Adaptive at V̄ 13.5, CAMS1900 reads Stable Adaptive at 11.6, and the newest evidence-gated run (M5N) reads Strained at 8.0. The 5.5 NV gap between UNIFIED and M5N is about 35 standard errors, so it is systematic, not scorer noise, and it widens after 2020 (mean gap 1.3, 2.0, 4.1 NV by decade). It sits in Coherence and Abstraction, not Stress. Whether that reflects thin external evidence on opaque systems, a different scoring protocol, or real organisational difference cannot be settled from these data; the M5N NA rates would help.

Balanced on a boundary

Australia is balanced near a regime boundary. The model leans toward persistence, expects Archive to stay among the strongest functions, and flags that the July 2026 Craft-recovery claim holds or fails depending on the dataset used.

Where the datasets disagree. Australia sits close to a regime boundary on every dataset, so small level differences flip labels (regimes agree in only 26% of overlap years between UNIFIED and M5N). UNIFIED reads Strained at V̄ 8.9, M5N Strained at 9.9 (15% of base-noise draws read Stable Adaptive) and CAM5_1996 Stable Adaptive at 11.8. M5N reads Australia higher than UNIFIED, mainly through lower Stress, the opposite direction to China. Archive in the top three is the one node-level feature all three datasets share. The UNIFIED envelope is largely carried forward (19 distinct blocks in 152 years) and understates Abstraction spread; the finding that its base is Strained holds with that spread floored at M5N levels.

Base states by dataset

The table is the heart of the uncertainty. Reading across any society’s rows, the starting point is not agreed, which is why a single regime verdict would overstate what the instrument knows.

Limits

• The scores are model-generated (AI-rater ensembles reading documentary evidence). Ensemble tightness certifies precision, not ground truth, and shared model bias is an unquantified confound.
• The four USA, three China and three Australia datasets disagree by more than the ±1 NV between-run margin the formalism cites. The register does not choose between them. It reports every claim on every available dataset and scores only the primary dataset as a single verdict.
• The simulation has no directional edge: with drift and mean-reversion pooled, medians sit near the base state. The register makes no call on whether V̄ rises or falls.
• Noise covariance is estimated from a small number of anchor-to-anchor changes (12 to 226 scored states) and shrunk toward the diagonal. Intervals are indicative.
• Claims are correlated (many share a node, a threshold or a path). Hit rates and Brier scores are not independent trials.
• Bond strength in JUNO v1.2 is algebraically rank-1, so B̄ and λ₂ carry no pair-specific topology. Claims on B̄ restate stress and coupling quality. Network pictures are illustrative.
• Phantom Type II has no mean-viability condition and fires on s_min < −0.65, so simulation noise raises its share with horizon. The 2028 Phantom share is partly a noise artefact.
• JUNO v1.2 is not externally validated. Scoring this register is part of the prospective test the formalism names, together with the frozen 28 August 2026 register.

Rules fixed in advance

• When: each year’s claims are scored after that year closes, from the first complete five-scorer ensemble scoring, with the dataset name and commit recorded.
• Primary dataset: the unified dataset named by the frozen register. If it has no 2027 or 2028 row, the claim is scored on whatever exists and marked as such.
• Tier A claims get a hit or miss on the primary dataset and are also reported on every other. Tier B claims get a single verdict only if every available dataset agrees; otherwise they are recorded as dataset-dependent, which counts as neither.
• Brier scores are reported for all claims and for Tier A separately. Claims are correlated, so they are not independent trials.
• No edits. Thresholds, wording and probabilities are fixed at deposit. Corrections go in a dated erratum.

What is happening, and is not modelled

United States. Press reports on 4 November 2026 (for example the ABC News projection) projected Democratic control of both chambers of Congress after the 3 November midterms, with seat counts differing between sources while counting continued. (ABC News live updates, 4 Nov 2026)

China. The Associated Press reported on 29 September 2026 that second-quarter 2026 growth was 4.3% against a 2026 target of 4.5–5%, and that new measures to support the property sector and economy had been announced. Outlets continue to describe a prolonged property downturn and weak domestic demand. (AP via WRAL, 29 Sep 2026, Outlook Business, Q2 2026 GDP)

Australia. The Reserve Bank of Australia raised the cash rate by 25 basis points to 4.85% on 3 November 2026, its fifth increase of the year, on a 6–3 vote, and projected headline inflation peaking around 4.5% and staying above 3% until late 2027, with GDP growth around 1.6% in the year to December 2026. (investingLive, RBA 3 Nov 2026, The Adviser, RBA 3 Nov 2026)

None of this enters the simulation. If an event inside the window moves the scores, that is information about the world, not a reason to reinterpret the claims.

How this fits with what is already on the record

This register does not replace the six-nation predictions of 12 July 2026 or the twelve claims locked on 10 June 2026 and frozen on 28 August 2026. Those stay in force and are scored separately. The 12 July memo was issued under a v1.0-labelled classifier table; this register and the frozen register’s scoring protocol use JUNO v1.2-Final. Where a claim here overlaps an earlier one (US-4, CN-2, CN-6, AU-2, AU-3), it is flagged and the earlier wording governs the earlier claim.

Archive, data and code

The full register, every probability by dataset, the input data and the code that regenerates every number are deposited together. Archive DOI: 10.5281/zenodo.24046639. Upstream repository: KaliBond/wintermute.

Suggested citation: McKern, K. (2026). CAMS/JUNO v1.2 prospective predictions 2027–2028: United States, China and Australia (Version 1.0.0). Neural Nations / ComplexityWorkz. Text and data CC BY 4.0; code MIT.