01Official
“Over the last 12 months, the all items index increased 2.7 percent before seasonal adjustment.”
Inflation dataUnited StatesResolved case
The machinery win:
the CPI tie, December 2025
The question
Will the first-published all-items CPI 12-month change for December 2025 be at or above November's first print?
Resolution rule — “at or above” includes an exact tie. Both monthly prints are compared as first published, before later revisions.
Our call · frozen 30 Dec 2025
87%
computed from level estimates
Direct-answer median
52%
the same analysts, asked YES or NO
Outcome · 13 Jan 2026
YES
2.7 versus 2.7 · an exact tie
Forecast error
0.017
direct answering: 0.230
Brier error: squared distance between the forecast and what happened. 0 = perfect, 0.25 = coin-flip guessing. Lower is better.
01 / The setup
A tie counts as YES.
The question is not a coin flip, even when the economy is. CPI prints round to one decimal place, and the rule says “at or above.” A no-change world therefore resolves YES.
The panel was asked twice: first for the binary outcome, then for the CPI level and uncertainty. Only the second route forced the rounding geometry into the answer.
Timeline / what was knowable when
-
November first print published at 2.7%.
-
Evidence frozen at T−14; both question forms registered.
-
December first print due; the question resolves.
02 / The frozen evidence
The public pack,
reduced to its load-bearing facts.
Two labeled excerpts from the evidence frozen on 30 December 2025. How packs are assembled stays private.
02Threshold
Known at freeze
November first print
2.7%
December first print
Below 2.7%NO
At or above 2.7%YES
Registered before December's result was known. An exact tie would count as YES.
Public excerpt only · complete record available through the audit trail
03 / The panel
Same models. Same evidence.
Different question shape.
Each seat answered twice. Direct YES probabilities clustered around a coin flip; level estimates implied substantially more probability above the 2.7 threshold.
Direct P(YES) · median0.52
Asked for the binary answer
Computed P(YES) · median0.87
Derived from level + uncertainty
SeatDirect P(YES)Level estimateComputed P(YES)
Analyst seat 1
0.502.8 ±0.150.84
Analyst seat 2
0.652.7 ±0.300.57
Analyst seat 3
0.552.9 ±0.200.89
Analyst seat 4
0.542.9 ±0.200.89
Analyst seat 5
0.422.8 ±0.200.77
Analyst seat 6
0.432.9 ±0.200.89
The calculator did not add new evidence. It translated each seat's stated level and uncertainty into the probability of meeting the registered 2.7 threshold.
04 / Why this was rare
The arithmetic was simple.
The intuition was not.
The models knew the number — mostly 2.8 or 2.9. They did not price what their own number implied.
The rule was asymmetric.
Below 2.7 resolves NO. Exactly 2.7 and everything above it resolve YES. A tie is not neutral.
Below 2.7
NO2.7 or above
YES
NO2.7 or above
YES
Rounding widened the YES region.
The published rate is rounded to one decimal place. A range of underlying values can land on the same 2.7 print.
2.652.72.74
The level question exposed the implication.
Direct answers sat near 0.50. Converting the models' own level estimates lifted the median to 0.87.
DirectBinary intuition0.52
ComputedLevel geometry0.87
05 / The resolution
The print landed
exactly on the line.
The December release arrived on 13 January 2026. It was not available to the panel and is shown only to resolve and score the registered forecast.
Resolved · Yes
YESDecember first print
13 Jan 2026
13 Jan 2026
November's first-published 12-month change was 2.7%. December's first print was also 2.7%. Because the question asked “at or above,” the exact tie resolved YES.
2.7 = 2.7 · the tie counts
Resolution source · byte-for-byte “The all items index rose 2.7 percent for the 12 months ending December, the same increase as over the 12 months ending November.” Wayback capture · 14 Jan 2026 ↗
(1 − 0.87)² = 0.017
Direct answering scored 0.230. Lower is better.
Calibration lab
Price the threshold yourself.
The threshold is 2.7. The panel's level estimates were 2.8–2.9 with roughly ±0.2 uncertainty. Set your probability that December landed at or above 2.7.
Error is (your probability − outcome)². The realized outcome is YES = 1; lower is better.
How we kept it honest
The result was screened for memory
before it counted.
A clean historical test needs more than an old evidence cutoff. The same question was first asked bare, without the pack, to check whether the outcome was already in model memory.
No analyst stated December's first print. One confident 2.9 guess was wrong — noise, not memory.
Every displayed evidence line is sourced and dated on or before the freeze.
The outcome was used only after the registered panel reads were recovered.
The registration commit predates the build; results are recorded at 95d0de8.
Registered question · verbatim
Will the first-published all-items CPI 12-month change for December 2025 be at or above November's first print?
Technical receiptRegistration, results and evidence record
- Registration
35daea5- Results
95d0de8- Artifact family
research_artifacts/policysim/cpical1/- Pack record
cpical_anchors.json · cpical-2026-01-13
Committed before the build.
Recorded result set.
Case evidence and recall-screen records.
Frozen evidence reference; public page shows excerpts only.
Verification trail
Request the trail
The separate public trail repository is being prepared. Verification materials are available on request.