The One Lab
SCORECARD ARENAS DISCOVERY FOUNDATIONS FOR AGENTS
AR-01 / ARENA LOCKED · AWAITING FIRST GRADE

Housing / US

Every forecast in this arena is locked and none has been graded yet. Each one was written to a table that cannot be edited or backdated, before the figure it predicts existed, and each is now waiting on the FHFA to publish that figure. Nothing on this page is a claim about how accurate we are — it cannot be, yet. The first verdicts arrive when the periods below mature, and they will be published whether they are passes or misses.

FORECASTS LOCKED 30
GRADED 0
ACCURACY insufficient data — nothing has been graded
WHAT THIS
ARENA PREDICTS
For the United States and its nine Census divisions, the level of the FHFA Purchase-Only House Price Index at three, six and twelve months ahead, with a published interval. Each forecast is written to a write-once table before the outcome exists, and graded against the first release of the same series — the number the market itself saw, not a revised one.

How to read this

We said, on the record and in advance, that United States house prices would be +2.68% over 12 months. Nobody can edit that now. In ~Jul 2027 the official figure arrives and we publish whether we were right — including when we were not.

What the words mean →
A forecast of +2.68% over 12 months
Our best single guess at how much the price index moves between now and then. If today's index is 100 and we say +3%, we are guessing it reads about 103 when the period we are forecasting is finally published. It is a guess about an official statistic, not about the price of any particular house.
What "12 months" is counted from
From May 2026 — the most recent month the FHFA had published when we locked this — not from the day we locked it. Official price statistics arrive a couple of months late, so the newest real figure is always somewhat behind today. Counting the horizon from the data is what makes the forecast a forecast: we predict a month whose figure does not exist yet, and cannot, until the FHFA publishes it.
"Locked"
Written down before the answer existed, in a place that cannot be edited or deleted — not by us, not by anyone. The date stamped on it is put there by the database at the moment of writing, so it cannot be backdated. This is the whole point: anyone can look accurate by adjusting a prediction after the fact, and the only defence against that is making it impossible.
The range around the forecast
A single number pretends to a confidence nobody has, so each forecast carries a range. Read it as: if our method is working as intended, the true figure should land inside this range about 95% of the time. A wider range is not us hedging — it is us telling you how much we actually know. Ranges get wider the further ahead we look, because they should.
When this gets its grade
On ~Jul 2027, when the statistics agency publishes the figure for the period we forecast. We do not grade against a number we produced, and we do not go back and choose a kinder one later. Every forecast on this page shows its own verdict date, and the wait is real: a twelve-month forecast cannot be graded early, and we would rather say so than fill the gap with something that looks like progress.
"PENDING", "PASS", "MISS"
PENDING means the answer does not exist yet. PASS and MISS are the verdict once it does. Misses stay on the page permanently, next to the hits. A record with no misses in it is not a good record — it is an incomplete one.

The longer version: how the scorecard works, and why we show our misses →

01 / LIVE FORECAST PANEL
LOCKED · CURRENT VINTAGE 2026-08-04T22:05:04.485867+00:00
b1d86293 · UNITED STATES · 12M
454.285
95% interval 427.869 … 482.332
Index level. That is +2.68% against May 2026 — the last month the FHFA had published when this was locked, and the point the 12M horizon is measured from.
vintage b1d86293 → FOR May 2027 · VERDICT ~Jul 2027
OTHER DIVISIONS · SAME VINTAGE
West South Central 431.727
West North Central 440.840
South Atlantic 484.516
Pacific 454.678
New England 479.856
Mountain 626.838
Middle Atlantic 431.044
East South Central 435.197
East North Central 392.337
NEXT LOCKon the next FHFA publisher vintage — checked daily
02 / EVERY LOCK IN THIS CYCLE

Every forecast this arena has issued, in the order the run log prints them. Each row opens the locked record it came from. None has been graded, so none carries an outcome — a pending row shows what was committed, and nothing else.

DATA THROUGH is the last month the FHFA had published when the forecast was locked — the point the horizon is measured from. TARGET PERIOD is the month the forecast is about: data-through plus the horizon. EXPECTED VERDICT is roughly when the FHFA publishes that month's figure and the forecast can be graded, on their release calendar rather than ours, which is why it carries a tilde.

Read the three together. A horizon counts months of the index, not months on the wall clock. The FHFA publishes about two months in arrears, so a three-month forecast locked in early August runs from May's figure — the newest one that existed — to August's, which will not be published until around October. The forecast is neither late nor a short hop: it was written when August's number did not exist, and it still does not.

VINTAGE SERIES HORIZON FORECAST CHANGE INTERVAL DATA THROUGH TARGET PERIOD EXPECTED VERDICT GRADE
57f28975 East North Central 3M 383.823 +0.99% 377.961 … 389.777 May 2026 Aug 2026 ~Oct 2026 PENDING
a720b478 East North Central 6M 386.955 +1.81% 377.319 … 396.837 May 2026 Nov 2026 ~Jan 2027 PENDING
2339d9eb East North Central 12M 392.337 +3.23% 373.702 … 411.901 May 2026 May 2027 ~Jul 2027 PENDING
d02994e2 East South Central 3M 423.824 +0.56% 417.476 … 430.267 May 2026 Aug 2026 ~Oct 2026 PENDING
ceb0870d East South Central 6M 428.052 +1.56% 417.782 … 438.574 May 2026 Nov 2026 ~Jan 2027 PENDING
54a1ef5a East South Central 12M 435.197 +3.26% 415.312 … 456.035 May 2026 May 2027 ~Jul 2027 PENDING
19075ba8 Middle Atlantic 3M 418.287 +0.89% 411.066 … 425.635 May 2026 Aug 2026 ~Oct 2026 PENDING
7399c8f1 Middle Atlantic 6M 422.870 +2.00% 410.897 … 435.192 May 2026 Nov 2026 ~Jan 2027 PENDING
d411b483 Middle Atlantic 12M 431.044 +3.97% 407.889 … 455.514 May 2026 May 2027 ~Jul 2027 PENDING
ff01986b Mountain 3M 611.121 +0.53% 597.165 … 625.404 May 2026 Aug 2026 ~Oct 2026 PENDING
c699e4b6 Mountain 6M 616.466 +1.41% 589.961 … 644.161 May 2026 Nov 2026 ~Jan 2027 PENDING
1b41ce55 Mountain 12M 626.838 +3.11% 571.447 … 687.599 May 2026 May 2027 ~Jul 2027 PENDING
7053e6c1 New England 3M 461.896 +1.37% 451.778 … 472.241 May 2026 Aug 2026 ~Oct 2026 PENDING
7a5b4654 New England 6M 467.562 +2.61% 451.479 … 484.217 May 2026 Nov 2026 ~Jan 2027 PENDING
9c8abc48 New England 12M 479.856 +5.31% 449.805 … 511.914 May 2026 May 2027 ~Jul 2027 PENDING
4d8f88e0 Pacific 3M 461.945 −0.44% 451.543 … 472.586 May 2026 Aug 2026 ~Oct 2026 PENDING
ab566341 Pacific 6M 458.702 −1.14% 437.788 … 480.615 May 2026 Nov 2026 ~Jan 2027 PENDING
73207311 Pacific 12M 454.678 −2.00% 407.649 … 507.131 May 2026 May 2027 ~Jul 2027 PENDING
34ec67fd South Atlantic 3M 472.748 +0.69% 463.710 … 481.963 May 2026 Aug 2026 ~Oct 2026 PENDING
276a3098 South Atlantic 6M 476.501 +1.49% 459.808 … 493.800 May 2026 Nov 2026 ~Jan 2027 PENDING
f602c510 South Atlantic 12M 484.516 +3.20% 449.237 … 522.565 May 2026 May 2027 ~Jul 2027 PENDING
2595606d United States 3M 445.427 +0.68% 439.152 … 451.792 May 2026 Aug 2026 ~Oct 2026 PENDING
e1e174c7 United States 6M 448.305 +1.33% 436.085 … 460.868 May 2026 Nov 2026 ~Jan 2027 PENDING
b1d86293 United States 12M 454.285 +2.68% 427.869 … 482.332 May 2026 May 2027 ~Jul 2027 PENDING
6d7ad9bd West North Central 3M 431.461 +0.81% 424.893 … 438.131 May 2026 Aug 2026 ~Oct 2026 PENDING
f25b7a77 West North Central 6M 434.129 +1.44% 423.681 … 444.834 May 2026 Nov 2026 ~Jan 2027 PENDING
3ed0fbb7 West North Central 12M 440.840 +3.00% 420.442 … 462.228 May 2026 May 2027 ~Jul 2027 PENDING
30bd22cf West South Central 3M 425.541 +0.12% 419.486 … 431.684 May 2026 Aug 2026 ~Oct 2026 PENDING
146cfaf0 West South Central 6M 427.050 +0.47% 416.691 … 437.668 May 2026 Nov 2026 ~Jan 2027 PENDING
79c5ba1e West South Central 12M 431.727 +1.57% 411.397 … 453.062 May 2026 May 2027 ~Jul 2027 PENDING
03 / GRADED HISTORY

Every row links to the vintage's own page. Misses sit in the same table as passes, in issue order, never sorted away.

BELOW THE SEAM · MIGRATED v1 BACKTEST · FITTED AFTER THE FACT · COUNTS ONLY, NO RATES
SEAM · NOTHING BELOW THIS LINE WAS FITTED FORWARD RECORD BEGINS 2026-08-04 · NOTHING BELOW WAS LOCKED IN ADVANCE
INSUFFICIENT DATA
No forecast has been graded yet. The forecasts above are locked and waiting on the FHFA figures for the periods they name; the first verdict appears here on its own schedule, whether it is a pass or a miss. It cannot be hurried, and a forecast cannot be graded early.
04 / LIMITATIONS

Constitutional. This block is the same size class as the results panel above and is never collapsed, moved below the fold, or set in a smaller face.

WHAT THIS ARENA CANNOT TELL YOU
L-01
This arena has no forward track record. Its forecasts are locked and none has been graded, so there is no accuracy figure on this page — not a provisional one, and not a favourable early read. The first grades mature on the FHFA's publication schedule, roughly two months after each period ends.
L-02
The historical results shown on this site for this arena are a BACKTEST, migrated from v1. They were fitted after the fact against data that already existed, and no backtest is evidence that a forecast made in advance will be right. They are never mixed with, averaged into, or shown as continuous with the locked forecasts above.
L-03
The migrated backtest has unmeasured completeness. v1 graded where its loop ran and matured, and no committed artifact states coverage per model, horizon or series — so an ungraded forecast in that set is not a miss, and the denominator of any rate over it is unestablished. Counts of graded rows are published; rates are not, and will not be until that is fixed.
L-04
v1's forecast grid was mutated roughly 129,960 times before write-once triggers existed on 2026-06-26, and its pre-mutation state is unrecoverable. Its published ensemble was adopted after the combiner it replaced was found degenerate on this same backtest, and its turning-point anchors were widened one day after grading began. Model selection followed the outcomes. This is why that history is classed as backtest and carries no rate.
L-05
Consecutive locks of the same target period are correlated forecasts, not independent trials. Every declared series is forecast at every lock at three, six and twelve months, so once grading begins, a count of grades over this arena counts each realized month up to three times. The count of distinct realized months is published beside every grade count for exactly this reason.
L-06
A horizon here counts months of the index, not months on the calendar. Each forecast is measured from the last month the FHFA had published when it was locked — shown as DATA THROUGH on every row — and the FHFA publishes about two months in arrears. So a three-month forecast locked in early August runs from May's figure to August's. What is guaranteed, and independently checked on every lock, is that the target month's figure did not exist when the forecast was written. What is not claimed is three months of calendar distance between the lock and the answer.
L-08
Metropolitan areas are excluded, not missing. This arena covers the United States and the nine Census divisions — the coverage v1 proved. No metro forecast exists to show, and none is implied by the divisions above.
L-09
FHFA figures reach us through FRED/ALFRED rather than directly from the FHFA, because the FHFA publishes the current index but not an archive of what each past release said — and grading a forecast against the data available when it was made requires that archive. The redistribution is recorded on every source row. ALFRED's vintages for these series begin 2012-05-23; earlier vintages do not exist to be had.
05 / ISSUE CADENCE
ISSUE CADENCE monthly — triggered by the FHFA/ALFRED publisher vintage, not by the calendar
HORIZONS 3, 6 and 12 months, for every declared series, at every lock
Consecutive locks of the same target period are correlated forecasts, not independent trials. A grade count over this arena is not a count of independent tests: summing across the three horizons counts each realized month three times. Wherever grades render, the count of distinct realized months renders beside the count of grades.
06 / CALIBRATION
INSUFFICIENT DATA
A calibration curve needs graded forecasts to plot. These forecasts are locked but none has matured, so there is no curve — not a flat one, and not a shape borrowed from anywhere else.
07 / PROVENANCE & ENDPOINTS
PRIMARY SOURCE Federal Housing Finance Agency (FHFA)
RETRIEVED VIA FRED / ALFRED — real-time vintage archive, redistribution disclosed
INDEX FHFA House Price Index, Purchase-Only, monthly, seasonally adjusted
SERIES RESOLUTION resolved per run from FRED metadata; no series ID is pinned
COVERAGE United States + 9 Census divisions; metros excluded
MODEL SPEC v1.1 — AR(12) on log first differences, pre-registered 2026-08-04
ACTUALS first_print (headline) · settled_t_plus_12 (companion)
VINTAGES FROM 2012-05-23 (earliest ALFRED vintage for these series)
ISSUE CADENCE monthly — triggered by the FHFA/ALFRED publisher vintage, not by the calendar
HORIZONS 3, 6 and 12 months, for every declared series, at every lock
THE v1 API HAS NOT BEEN CUT OVER It still serves the design fixture set and stamps sample:true on every payload. The thirty locked forecasts on this page are not in it yet. Read them here, or from the vintage pages, until it is.
GET /v1/vintages?arena=housing_us Sample fixtures, labelled sample:true. Will serve this arena's locked vintages when the Worker is cut over.
GET /v1/calibration?arena=housing_us Sample fixtures. No calibration exists for this arena in any case — nothing has been graded.
The One Lab · Toronto · Every number on this site resolves to a primary source or a graded model output. Where it can't, we say so. ARENA AR-01 · NEXT LOCK ON THE NEXT FHFA PUBLISHER VINTAGE — CHECKED DAILY