Minerval

← evals

EvalWhere can reality check it?

Predictions

Twenty-two questions the world will settle, seeded as claims. When they resolve, the Steward's probabilities are scored against what happened.

22 questions, authored 2026-09-02not yet seeded into productionfirst resolutions 2026-10-15≈22 Steward runs to seed

How it works

A prediction over time: seeded as a claim, assessed by the Steward, credence frozen at the cutoff, then scored against the outcome.seededas an ordinary claimassessedSteward records a credencecutofflast credence before this countsresolvedthe world settles itscoredBrier, log, calibration

Each question is seeded as an ordinary claim that its The agent that owns a claim: breaks it into the claims it depends on, weighs the evidence, and records a verdict.agent assesses like any other. The The Steward's probability that a claim is true.constitution that is scored is the last one recorded at or before the cutoff: the actual resolution, or the scheduled date if that came first. A question with no credence in time counts as declined, not scored.

Scores: Brier, log score, a calibration curve with expected calibration error, per domain, and against market forecasts once those are attached (none yet). Code: prediction-score.ts.

Detail

The 22 questionsthe fixture, verbatim

Open a row for what counts as yes and where to look. File: manifest.json.

2026-10-15 · climate · Arctic sea ice extent will set a record-low September monthly average in 2026.

Resolves yes if: YES if the NSIDC Sea Ice Index September 2026 monthly average extent is below the September 2012 monthly average, the record low at the time of seeding.

Where to look: NSIDC Sea Ice Index (G02135), monthly average extent, Northern Hemisphere, September 2026 versus September 2012.

Note: Short horizon; the melt season is mostly over at seeding, which makes this a test of how the Steward handles late information.

2026-10-31 · science · The 2026 Nobel Prize in Physics will be shared by two or more laureates.

Resolves yes if: YES if the Royal Swedish Academy of Sciences names more than one laureate for the 2026 Nobel Prize in Physics.

Where to look: nobelprize.org, the official 2026 Physics announcement.

Note: Short horizon; a recurring, well-defined event with a stable base rate.

2026-10-31 · science · At least one 2026 Nobel Chemistry laureate will be affiliated with a United States institution.

Resolves yes if: YES if nobelprize.org lists a United States institution as an affiliation at the time of the award for any 2026 Chemistry laureate.

Where to look: nobelprize.org laureate pages for the 2026 Chemistry prize, the 'affiliation at the time of the award' field.

Note: Short horizon.

2026-10-31 · politics · The 2026 Nobel Peace Prize will be awarded to an organization rather than only to individuals.

Resolves yes if: YES if at least one 2026 Peace Prize laureate is an organization (alone or shared).

Where to look: nobelprize.org, the official 2026 Peace Prize announcement.

Note: Short horizon.

2026-11-10 · sports · The Los Angeles Dodgers will win the 2026 World Series.

Resolves yes if: YES if the Dodgers win the 2026 MLB World Series.

Where to look: MLB.com official results.

Note: A low-probability-by-construction question, to populate the low end of the calibration curve.

2026-12-15 · climate · A Category 4 or stronger hurricane will make U.S. landfall during the 2026 Atlantic season.

Resolves yes if: YES if the National Hurricane Center's records show a hurricane making landfall in a U.S. state or territory at Category 4 or 5 intensity (sustained winds at or above 130 mph) during the 2026 season.

Where to look: NHC tropical cyclone reports and advisories for the 2026 Atlantic season.

Note: Partly elapsed at seeding; still open through November.

2026-12-20 · economics · The federal funds target range's upper bound will be at or below 3.25% after the December 2026 FOMC meeting.

Resolves yes if: YES if the FOMC statement issued at the conclusion of its December 2026 meeting sets a target range whose upper bound is 3.25% or lower.

Where to look: federalreserve.gov, the FOMC statement for the December 2026 meeting.

Note: Monetary-policy path; a question markets price continuously, so a baseline is easy to attach later.

2027-01-02 · markets · The S&P 500 will close 2026 above its 2025 year-end close.

Resolves yes if: YES if the S&P 500's official closing level on the last trading day of 2026 exceeds its official close on 2025-12-31.

Where to look: S&P Dow Jones Indices official closing levels.

Note: A base-rate question with a well-known prior.

2027-01-02 · markets · Bitcoin will trade above US$150,000 at some point before 2027-01-01.

Resolves yes if: YES if CoinMarketCap's BTC/USD daily high exceeds 150,000 on any day from seeding through 2026-12-31.

Where to look: CoinMarketCap historical data for BTC, daily high, USD.

2027-01-04 · politics · Republicans will hold a majority of U.S. House seats when the 120th Congress convenes.

Resolves yes if: YES if, on the day the 120th Congress first convenes (scheduled 2027-01-03), the Clerk of the House's official party breakdown shows Republicans with at least 218 seats, vacancies counted as neither party.

Where to look: Office of the Clerk of the U.S. House of Representatives, official party division on the convening day.

Note: Resolves on the November 2026 midterms plus certification.

2027-01-04 · politics · Republicans will hold at least 50 U.S. Senate seats when the 120th Congress convenes.

Resolves yes if: YES if the Senate's official party division on the convening day of the 120th Congress shows 50 or more Republican senators (independents caucusing with either party do not count as Republicans).

Where to look: senate.gov party division for the 120th Congress on its convening day.

Note: Sibling of the House question; tests whether credences on correlated questions cohere.

2027-01-05 · technology · SpaceX will catch a Starship upper stage with the launch tower arms during 2026.

Resolves yes if: YES if SpaceX's official flight recap for any Starship flight in 2026 states that the ship (the upper stage, not the Super Heavy booster) was caught by the tower.

Where to look: spacex.com flight updates for Starship flights in 2026.

Note: Engineering milestone with a crisp observable.

2027-01-15 · economics · Euro-area annual HICP inflation for December 2026 will be at or below 2.0% in Eurostat's flash estimate.

Resolves yes if: YES if Eurostat's flash estimate of euro-area annual HICP inflation for December 2026 is 2.0% or lower.

Where to look: Eurostat euro indicators flash estimate for December 2026 (published early January 2027).

2027-01-15 · business · Tesla will report more than 1.8 million vehicle deliveries for calendar 2026.

Resolves yes if: YES if Tesla's official production and deliveries release for Q4 2026 puts full-year 2026 deliveries above 1,800,000.

Where to look: Tesla investor relations, the Q4 and FY 2026 production and deliveries press release.

2027-01-31 · climate · NASA GISS will report the 2026 global mean surface temperature anomaly at or above 1.20 °C versus 1951–1980.

Resolves yes if: YES if the GISTEMP v4 annual land-ocean temperature index for calendar 2026, as first published, is at or above 1.20 °C relative to the 1951–1980 mean.

Where to look: NASA GISS GISTEMP v4 annual mean anomaly table (the January release covering 2026), first published value.

Note: A threshold near recent annual values, so the question is live rather than settled.

2027-01-31 · economics · U.S. CPI-U 12-month inflation for December 2026 will be below 3.0% as first published.

Resolves yes if: YES if the BLS first-published 12-month percent change in CPI-U, all items, not seasonally adjusted, for December 2026 is below 3.0%.

Where to look: BLS Consumer Price Index news release for December 2026, first print (revisions ignored).

2027-01-31 · economics · China's preliminary 2026 GDP growth will be reported at or above 4.5%.

Resolves yes if: YES if the National Bureau of Statistics of China's preliminary estimate of real GDP growth for calendar 2026 is 4.5% or higher.

Where to look: NBS of China, the January 2027 preliminary GDP release for 2026.

Note: Official statistic as published; the question is about the reported number, not the true growth rate.

2027-07-01 · science · The James Webb Space Telescope will be conducting science observations on 2027-06-30.

Resolves yes if: YES if JWST is in normal science operations (not in an extended safe mode or decommissioned) on 2027-06-30 per STScI/NASA status reports.

Where to look: STScI JWST observatory status and NASA mission updates.

Note: A high-probability-by-construction question, to populate the high end of the calibration curve.

2027-09-02 · science · A magnitude 8.0 or greater earthquake will occur somewhere on Earth between 2026-09-02 and 2027-09-01.

Resolves yes if: YES if the USGS earthquake catalog contains any event with reviewed magnitude 8.0 or greater in that window.

Where to look: USGS Earthquake Catalog (earthquake.usgs.gov), reviewed magnitudes.

Note: A pure base-rate question; good for checking whether the Steward reasons from frequency.

2027-09-02 · health · The WHO will declare a new Public Health Emergency of International Concern between 2026-09-02 and 2027-09-01.

Resolves yes if: YES if the WHO Director-General declares a PHEIC under the International Health Regulations in that window, for an event not already under a PHEIC declaration at seeding.

Where to look: WHO news releases and IHR Emergency Committee statements.

2027-09-02 · economics · The NBER will announce a U.S. business-cycle peak dated in 2026 before 2027-09-01.

Resolves yes if: YES if, before 2027-09-01, the NBER Business Cycle Dating Committee announces a peak (the start of a recession) with a peak month in calendar 2026.

Where to look: NBER Business Cycle Dating Committee announcements (nber.org/research/business-cycle-dating).

Note: Announcement lag is part of the question, deliberately.

2028-01-02 · politics · The United Kingdom will hold a general election before 2028-01-01.

Resolves yes if: YES if polling day of a UK general election falls on or before 2027-12-31.

Where to look: UK Parliament / Electoral Commission records.

Note: Longest horizon in the set.

What it cannot show

Anything about claims that never resolve, which is most of the graph. The questions are in-house, several are correlated, and no market baseline is attached, so the curve will be thin for a long time.

Run it
npm run predictions -- list
npm run predictions -- seed --corpus --drain        # assess the seeds now (LLM spend)
npm run predictions -- resolve <id> yes|no --note="how"
npm run predictions -- score