The declared math rounds to the published probability.
WP-SCIENCE-0.2
Do not trust the percentage. Reproduce it.
Every public probability is rebuilt from a declared prior and dated evidence. Exact inputs are separated from context, unavailable data receives no directional weight and lowers coverage, and unvalidated weights are labeled honestly.
Required inputs with a retained value, period and revision state.
Two component models are tested; zero full forecast models are validated.
Calibration begins only after comparable forecasts resolve.
REPRODUCIBLE METHOD
Prior belief + declared evidence weight = published probability.
posterior = logistic(logit(prior) + Σ evidence contribution)Log-odds let independent evidence move a probability without crossing below 0% or above 100%. The current weights are expert research assumptions, not fitted coefficients. That distinction remains visible until backtesting and calibration exist.
ECONOMY · WP-ECO-0.1
49%PUBLISHEDWill U.S. all-items consumer-price inflation be at or above 3.0% year over year in December 2026?
PRIOR41%
+NET EVIDENCE+8 pts
=RECOMPUTED49%
MATH MATCHES
PRIOR BASIS
Pre-release base rate for a six-month threshold crossing, set before the June CPI input.
POINT-IN-TIME INPUTS100% exact · quality 100/100
Official primary sourceU.S. Bureau of Labor StatisticsConsumer Price Index, all urban consumers, all items
OBSERVATION3.5 percent year over yearAll-items CPI rose 3.5% over the 12 months ending June 2026.
PERIOD + REVISION2026-06First release · retrieved 2026-07-30
MODEL EFFECT+7.9 ptsProbability input · log-odds +0.32
Official modeled sourceIMFConsumer prices and World Economic Outlook context
OBSERVATIONExcludedMacroeconomic context only; no second CPI value is allowed to duplicate the BLS input.
PERIOD + REVISION2026-07Current release · retrieved 2026-07-30
MODEL EFFECT0 ptsContext only · log-odds 0
ReproducibleDeclared inputs recreate the published number.
Point-in-timeObservation and retrieval dates remain attached.
Component testedA historical component diagnostic is published below; the full model is still unvalidated.
Calibration pendingRequires resolved forecasts.
HISTORICAL COMPONENT DIAGNOSTIC
40.8%HISTORICAL-ONLY SHADOWDoes June inflation contain useful information about December?
Single-feature logistic regression with L2 regularization λ=0.15; leave-one-year-out validation. The feature improves probabilistic Brier score versus a constant base rate, but threshold accuracy does not beat always-NO. This tests one input—not the complete published forecast—and therefore does not replace the 49% public probability.
TRAINING WINDOW2001–202525 annual observations
LEAVE-ONE-YEAR-OUT BRIER0.1759Base-rate benchmark 0.2016
BRIER IMPROVEMENT12.7%Probabilistic improvement over a constant base rate
THRESHOLD ACCURACY72%Always-NO also scores 72%
Open all 25 historical years and caveats
YEARJUNE YoYDECEMBER YoYOUTCOME
20013.194%1.604%NO
20021.069%2.480%NO
20031.949%2.035%NO
20043.168%3.342%YES
20052.541%3.339%YES
20064.182%2.524%NO
20072.693%4.109%YES
20084.936%-0.022%NO
2009-1.229%2.814%NO
20101.122%1.438%NO
20113.502%3.062%YES
20121.654%1.760%NO
20131.716%1.513%NO
20142.059%0.653%NO
20150.180%0.639%NO
20161.079%2.051%NO
20171.641%2.130%NO
20182.808%2.002%NO
20191.671%2.320%NO
20200.717%1.320%NO
20215.296%7.174%YES
20228.979%6.405%YES
20233.071%3.316%YES
20242.970%2.871%NO
20252.680%2.653%NO
- This tests one historical feature, not the complete expert-weighted WorldPredicta model.
- Twenty-five annual observations are a small sample and include major structural breaks.
- The 72% threshold accuracy only matches an always-NO classifier; the useful improvement is probabilistic Brier score, not hit rate.
- Seasonally adjusted CPI levels can be revised. A production backtest must preserve the vintage available on each historical forecast date.
Scientific limitations and update rule
Recompute after each first-published BLS all-items CPI release; never use a later revision to rewrite the historical run.
- Energy shocks and methodology revisions can move headline inflation abruptly. This initial model has no published WorldPredicta track record yet.
- Likelihood weights are declared expert research weights, not fitted coefficients.
- The result is not called calibrated until enough comparable forecasts resolve.
SUPPLY CHAIN · WP-FLOW-0.1
36%PUBLISHEDWill a tracked strategic sea lane record a 7-day average transit decline of at least 20% before December 31, 2026?
PRIOR30%
+NET EVIDENCE+6 pts
=RECOMPUTED36%
MATH MATCHES
PRIOR BASIS
Pre-release disruption base rate for a large 20% seven-day transit decline across the monitored passage set.
POINT-IN-TIME INPUTS100% exact · quality 85/100
Official modeled sourceIMF PortWatchPort and strategic chokepoint vessel-flow indicators
OBSERVATION4 passages on flow watch out of 13Four of 13 monitored passages were on flow watch; two were classified as disrupted in the July 26 snapshot.
PERIOD + REVISION2026-07-26Current release · retrieved 2026-07-30
MODEL EFFECT+6 ptsProbability input · log-odds +0.274
ReproducibleDeclared inputs recreate the published number.
Point-in-timeObservation and retrieval dates remain attached.
Backtest pendingNo historical performance is claimed.
Calibration pendingRequires resolved forecasts.
Scientific limitations and update rule
Recompute from complete weekly passage releases only; source outages and partial weeks contribute zero evidence.
- AIS coverage, vessel classification and calendar effects vary. A traffic decline does not by itself identify the cause.
- Likelihood weights are declared expert research weights, not fitted coefficients.
- The result is not called calibrated until enough comparable forecasts resolve.
CLIMATE · WP-CLIMATE-0.1
78%PUBLISHEDWill 2026 finish among the three warmest calendar years in the ERA5 global temperature record?
PRIOR70%
+NET EVIDENCE+8 pts
=RECOMPUTED78%
MATH MATCHES
PRIOR BASIS
Pre-release annual-rank prior informed by recent ERA5 annual rankings, without using local heat markers.
POINT-IN-TIME INPUTS100% exact · quality 85/100
Official modeled sourceCopernicus Climate Change ServiceERA5 global surface air temperature
OBSERVATION2 monthly global rank; anomaly details retained in observationJune 2026 ranked second warmest globally; it was 0.56°C above 1991–2020 and the trailing 12 months were 1.43°C above 1850–1900.
PERIOD + REVISION2026-06Modeled estimate · retrieved 2026-07-30
MODEL EFFECT+8 ptsProbability input · log-odds +0.419
Prototype contextOpen-MeteoCurrent and recent city-level temperature observations
OBSERVATIONExcludedUnavailable at the publication cutoff; city observations are excluded from the global-rank probability.
PERIOD + REVISION2026-07-30Unavailable · retrieved 2026-07-30
MODEL EFFECT0 ptsContext only · log-odds 0
ReproducibleDeclared inputs recreate the published number.
Point-in-timeObservation and retrieval dates remain attached.
Backtest pendingNo historical performance is claimed.
Calibration pendingRequires resolved forecasts.
Scientific limitations and update rule
Recompute after each Copernicus monthly bulletin using the first published ERA5 values and the same declared annual-rank rule.
- The current WorldPredicta formula is a baseline research model and has not been calibrated across a long hindcast. Local heat markers are context, not a global thermometer.
- Likelihood weights are declared expert research weights, not fitted coefficients.
- The result is not called calibrated until enough comparable forecasts resolve.
ECONOMY · WP-ECO-0.1
44%PUBLISHEDWill the IMF's first 2027 outlook update forecast world real GDP growth below 3.0% for 2027?
PRIOR50%
+NET EVIDENCE-6 pts
=RECOMPUTED44%
MATH MATCHES
PRIOR BASIS
Neutral pre-release prior because the question predicts a future institutional forecast rather than realized GDP.
POINT-IN-TIME INPUTS100% exact · quality 85/100
Official modeled sourceInternational Monetary FundWorld Economic Outlook
OBSERVATION3.4 percent real GDP growthThe July 2026 update projected 3.4% world real GDP growth for 2027, 0.4 percentage points above the forecast threshold.
PERIOD + REVISION2027 projection in July 2026 updateCurrent release · retrieved 2026-07-30
MODEL EFFECT-6 ptsProbability input · log-odds -0.241
Official primary sourceWorld BankWorld Development Indicators
OBSERVATIONExcludedCountry observations are retained as context but are not double-counted against the IMF forecast being predicted.
PERIOD + REVISIONlatest country-specific annual releasesCurrent release · retrieved 2026-07-30
MODEL EFFECT0 ptsContext only · log-odds 0
ReproducibleDeclared inputs recreate the published number.
Point-in-timeObservation and retrieval dates remain attached.
Backtest pendingNo historical performance is claimed.
Calibration pendingRequires resolved forecasts.
Scientific limitations and update rule
Recompute only when the IMF publishes a new WEO projection or a qualifying global shock input is added under a versioned rule.
- This forecasts another institution's forecast, not final realized GDP. Large revisions are common and country coverage is uneven.
- Likelihood weights are declared expert research weights, not fitted coefficients.
- The result is not called calibrated until enough comparable forecasts resolve.
FOOD · WP-FOOD-0.1
39%PUBLISHEDWill the FAO Food Price Index for December 2026 be at least 8% above its December 2025 level?
PRIOR35%
+NET EVIDENCE+4 pts
=RECOMPUTED39%
MATH MATCHES
PRIOR BASIS
Pre-release base rate for an 8% December-over-December rise in the global food basket.
POINT-IN-TIME INPUTS100% exact · quality 100/100
Official primary sourceFood and Agriculture OrganizationFAO Food Price Index
OBSERVATION130.3 index points; December 2025 baseline 124.3The June 2026 index was 130.3 versus 124.3 in December 2025, a 4.8% increase toward the forecast's 8% threshold.
PERIOD + REVISION2026-06; comparison baseline 2025-12First release · retrieved 2026-07-30
MODEL EFFECT+3.9 ptsProbability input · log-odds +0.168
Prototype contextOpen-MeteoCurrent temperature anomalies for selected cities
OBSERVATIONExcludedUnavailable at the publication cutoff and excluded because selected cities are not a crop-yield model.
PERIOD + REVISION2026-07-30Unavailable · retrieved 2026-07-30
MODEL EFFECT0 ptsContext only · log-odds 0
ReproducibleDeclared inputs recreate the published number.
Point-in-timeObservation and retrieval dates remain attached.
Component testedA historical component diagnostic is published below; the full model is still unvalidated.
Calibration pendingRequires resolved forecasts.
HISTORICAL COMPONENT DIAGNOSTIC
30.9%HISTORICAL-ONLY SHADOWDoes the June food-price path predict an 8% full-year rise?
Single-feature logistic regression with L2 regularization λ=0.15; leave-one-year-out validation. This single feature underperforms a constant base-rate probability and does not improve threshold accuracy. It is a warning against over-weighting the midyear index. This tests one input—not the complete published forecast—and therefore does not replace the 39% public probability.
TRAINING WINDOW1991–202535 annual observations
LEAVE-ONE-YEAR-OUT BRIER0.1954Base-rate benchmark 0.191
BRIER IMPROVEMENT-2.3%Probabilistic improvement over a constant base rate
THRESHOLD ACCURACY74.3%Always-NO also scores 74.3%
Open all 35 historical years and caveats
YEARJUNE VS PRIOR DECDEC VS PRIOR DECOUTCOME
1991-1.447%3.376%NO
19922.488%-4.666%NO
19930.816%5.546%NO
19941.236%13.292%YES
19952.729%6.685%NO
19963.197%-8.568%NO
1997-1.678%-4.755%NO
1998-5.433%-8.664%NO
1999-11.897%-16.238%NO
20004.031%5.182%NO
20010.730%-0.365%NO
2002-5.495%2.564%NO
20030.714%12.321%YES
20045.564%4.928%NO
20052.121%5.758%NO
20062.292%12.894%YES
200715.990%45.558%YES
200815.693%-24.760%NO
20098.227%16.802%YES
2010-4.861%28.373%YES
20114.173%-5.796%NO
2012-5.004%0.738%NO
2013-1.547%-3.746%NO
20140.846%-10.998%NO
2015-10.076%-17.300%NO
20167.816%9.540%YES
20172.623%1.154%NO
20180.415%-4.564%NO
20193.696%9.565%YES
2020-7.639%7.639%NO
202115.576%23.226%YES
202216.455%-0.449%NO
2023-7.588%-10.518%NO
20241.595%6.885%NO
20250.628%-2.200%NO
- This tests one historical feature, not the complete expert-weighted WorldPredicta food-price model.
- The negative Brier improvement means this feature should not be treated as validated predictive skill.
- Commodity shocks can reverse sharply between June and December; 2008, 2010 and 2022 illustrate that instability.
- FAO can revise recent monthly index values. A production backtest must retain the first-published vintage available at each historical forecast date.
Scientific limitations and update rule
Recompute after each first-published FAO monthly index; weather context contributes zero until a validated crop-weather model exists.
- The present heat layer is city-based and is not yet a global crop-weather or yield model.
- Likelihood weights are declared expert research weights, not fitted coefficients.
- The result is not called calibrated until enough comparable forecasts resolve.
TRAVEL · WP-MOVE-0.1
56%PUBLISHEDWill international tourist arrivals in 2026 exceed the 2019 level by at least 5% in UN Tourism's first full-year estimate?
PRIOR56%
+NET EVIDENCE0 pts
=RECOMPUTED56%
MATH MATCHES
PRIOR BASIS
Expert base-rate prior published without a qualifying 2026 arrivals input; this is intentionally low-maturity.
POINT-IN-TIME INPUTS0% exact · quality 0/100
Official modeled sourceUN TourismWorld Tourism Barometer
OBSERVATIONExcludedNo complete 2026 global-arrivals estimate was publicly verified at the model cutoff; the last located Q1 2025 release is too stale for a probability update.
PERIOD + REVISION2026 estimate unavailable at cutoffUnavailable · retrieved 2026-07-30
MODEL EFFECT0 ptsResolution source · log-odds 0
Prototype contextOpenSky and Airplanes.liveObserved aircraft state vectors used as movement context
OBSERVATIONExcludedAircraft state vectors are excluded from the probability because they do not measure passengers, trip purpose or occupancy.
PERIOD + REVISION2026-07-30Unavailable · retrieved 2026-07-30
MODEL EFFECT0 ptsContext only · log-odds 0
ReproducibleDeclared inputs recreate the published number.
Point-in-timeObservation and retrieval dates remain attached.
Backtest pendingNo historical performance is claimed.
Calibration pendingRequires resolved forecasts.
Scientific limitations and update rule
Hold the probability unchanged until a qualifying UN Tourism 2026 arrivals estimate is published; aircraft counts remain context only.
- Aircraft positions do not measure passengers, purpose of travel or occupancy. UN Tourism estimates can be revised.
- Likelihood weights are declared expert research weights, not fitted coefficients.
- The result is not called calibrated until enough comparable forecasts resolve.
Reproducible does not mean validated.
This release strengthens traceability and prevents hidden math. It does not create a historical track record. The next scientific gate is a point-in-time backtest that only uses information available on each simulated forecast date.