Corrections

Every figure on this site is re-checked nightly against the data it rests on. A wrong figure or a sentence that stopped holding is listed here the day it was found, with what it read, what it reads now and the check that now catches it. Nothing is removed. To report an error, email corrections@usmacro.com with the page's link and what looks wrong.

0open today

28corrections since October 18, 2026

7,399claims checked at the latest re-check, November 8, 2026

94claims that moved at a re-check and were resolved

November 2026

Found November 6, 2026Resolved the same dayserved sentence · new data

The GDPNow page's count of strong final nowcasts left out the third quarter of 2026

Read: Tracking strong growth — 3% or more. Eighteen final nowcasts since 2011; GDP came in at 3% or more after 11 of them". The reading said "In the 18 quarters since 2011 whose final nowcast read this high, GDP came in at 3% or more in 11 and at 1% or more in 16.
Reads now: "Since 2011, 19 final nowcasts read here; GDP came in at 3% or more after 12 of them", and "In the 19 quarters … in 12 and at 1% or more in 17", on the data of November 6, 2026. The counts are read from the data.
Cause: The counts were typed for the quarters through the second quarter of 2026. The third quarter's final nowcast was 3.3%, and GDP for it came in at 3.4% on October 29, 2026, so "since 2011" now holds one more quarter. The page was out of date from October 29, 2026 until this fix is deployed.
Where: /indicator/gdpnow
Now caught by: the typed-window census of November 6, 2026, which re-read each count on a closed window against the open record · record CLOSED-WIN-1

Found November 6, 2026Resolved the same dayserved sentence · new data

The services payrolls page's count of stalled months went out of date when August was revised

Read: Stalling — hiring this weak has been unusual outside recessions: about one non-recession month in seven from 1950 to September 2026 (121 of 808)". The stalled reading said the same, ending "(121 of 808).
Reads now: "Stalling — hiring this weak has been unusual outside recessions: 120 of the 809 non-recession months since 1950", on the data of November 6, 2026. The two counts are read from the data, so they follow later months and revisions.
Cause: The counts were typed for a window ending in September 2026. The jobs report of November 6 revised August's services gain from 55K to 102K, out of the 0K–60K stall band, so the window now holds 120 such months, not 121. The page read this way from October 23, 2026; it was out of date from the data fetch of November 6, 2026 at 22:00 UTC until this fix is deployed.
Where: /indicator/services_payrolls
Now caught by: the claims re-check of November 6, 2026 at 22:25 UTC, after the site's data fetch at 22:00 UTC · record SVCPAY-1

Found November 4, 2026Resolved the same dayserved sentence · correction

The housing affordability page's stretched band listed only some of its years

Read: Stretched — the fourth fifth; on today's bands, 1991–92, 1994, 2005–07 and most months since mid-2022". The band's reading said the same: "Against today's bands, that was the case in 1991–92, 1994 and 2005–07, and in most months since mid-2022.
Reads now: "Stretched — the fourth fifth; on today's bands, 1978–79, 1986–87, 1991–96, 2000, 2002, 2004–07 and 2022–25". The reading: "Against today's bands, that was the case in 1978–79, 1986–87, 1991–96, 2000, 2002, 2004–07 and 2022–25." The years are read from the data on the band's own edges.
Cause: The list was typed. On the stored history, months in the band also fall in 1978–79, 1986–87, 1993, 1995–96, 2000, 2002 and 2004, which the list left out. The page read this way until this fix is deployed.
Where: /indicator/housing_affordability
Now caught by: the typed-window review of November 4, 2026, serving each band's years · record CLAIMS-TIDY

Found November 4, 2026Resolved the same dayserved sentence · correction

The single-family permits page's top band left out two of its years

Read: ~~Against today's bands, readings this high came in every month from 1998 through 2006, most of 1977–78, and 13 months of 2020–22, with scattered months in 1972, 1984–87 and 1993–94.~~
Reads now: "Against today's bands, readings this high came in 1972, 1977–79, 1984, 1986–87, 1993–94, 1998–2007 and 2020–22." The years are read from the data at the band's edge.
Cause: A typed list, read off the chart rather than the stored months. On the stored history one month of 1979 and three months of 2007 also sit at or above the band's edge of 1,110K, and 1985 has none. The page carried the shorter list until this fix is deployed.
Where: /indicator/sf_permits
Now caught by: the typed-window review of November 4, 2026, serving the band's years · record CLAIMS-TIDY

Found November 4, 2026Resolved the same dayserved data · correction

Two State of the Economy cards' one-month change sat a tenth from the sum printed beside it

Read: The Inflation Regime card read "↑ 1.7 · vs Oct 2026" beside its why table's "Σ this month = +1.8 → the card's +1.7", and Reading Today said "this month's move is +1.7 pts". Forward Stress read "↑ 3.3" beside a sum of +3.2.
Reads now: Since the release of November 4, 2026 the change is worked out from October's end before rounding. On the November 4 data the Inflation Regime card reads "↑ 1.8" beside "Σ this month = +1.8 → the card's +1.8", and Forward Stress "↑ 3.2" beside "Σ this month = +3.2 → the card's +3.2". October settles on November 5, after which the cards read October as first published.
Cause: In a month's first days, before the prior month settles, the card's one-month change read that month's end from the daily record, which stores each day's score rounded to 0.1. The why table works from the same day's exact figure, as the table's note promises for every change. October 30's Inflation Regime reading was 29.75, stored as 29.8, so the change read 1.7 where the exact change was 1.8; Forward Stress's 55.548 was stored as 55.5, so its change read 3.3 for 3.2. This was the first month start since the cards began reading an unsettled prior month at its end.
Where: /state-of-the-economy
Now caught by: the nightly tests, on the November 4 data · record WHY-1B

Found November 4, 2026Resolved the same dayserved sentence · new data

The copper price page's record card called September 2026 the all-time high after October passed it

Read: ~~"All-Time High" and "The highest monthly average on the record, which begins in 1992." on the card dated September 2026, $14,474 a ton.~~
Reads now: "Record High at the Time" and "The second-highest monthly average on the record, which begins in 1992." The rank is read from the data, so it moves with later months.
Cause: The card's title and sentence were typed when September 2026 was the record. October 2026's monthly average, $14,508 a ton, arrived in the site's data fetch of November 4, 2026 at 14:00 UTC and passed it. The card read that way from the fetch until this fix is deployed.
Where: /indicator/copper_price
Now caught by: the claims re-check of November 4, 2026 at 18:42 UTC · record COPPER-1

Found November 3, 2026Resolved the same dayserved data · correction

The Time Machine could not load Forward Stress or the Inflation Regime for November 2

Read: ~~the State of the Economy for November 2, and for later days until the next stored score, did not load: the page's read failed instead of showing no score.~~
Reads now: since the release of November 3, 2026 at 20:17 UTC, those days load; a gauge the site could not score shows no score for that day. The scores are recomputed each night, and November 2 gets its two scores once a reading dated in November is stored.
Cause: On the first business day of a month, before any of their inputs has a reading dated in the new month, the two gauges cannot be scored for that day; the site stored the day with no score. The page's read for a past date then failed on that day instead of showing no score, so the State of the Economy for November 2, and for later days until the next stored score, did not load.
Where: /state-of-the-economy
Now caught by: the nightly tests, on a copy of the November 3 data · records READAPI-1, NULL-1

Found November 2, 2026Resolved the same dayserved sentence · correction

Two tooltips misnamed the Chauvet-Piger recession probability

Read: ~~the NY Fed's recession probability" and "Both read recession risk from the markets.~~
Reads now: the Chauvet-Piger recession probability" and "The EBP reads corporate bond spreads; the Chauvet-Piger probability reads economic activity: payrolls, industrial production, real income and real sales.
Cause: The series in both tooltips is the Chauvet-Piger smoothed recession probability, not the New York Fed's yield-curve model. The Chauvet-Piger model reads economic activity, not markets, as the recession probability page itself says. Both tooltips read this way from October 1, 2026 until this fix is deployed.
Where: 2 pages: /indicator/g7_cli · /indicator/ebp
Now caught by: the design partner's wording check during the Inflation Regime review · record IR-NOTE

Found November 2, 2026Resolved the same dayserved sentence · correction

The recession probability page stated figures the source has since revised

Read: ~~In 2001 it peaked near 30%." and "62.3% in June, then 2.5% in July 2009~~
Reads now: In 2001 it peaked at 28.1%, in June 2001." and "60.8% in June, then 2.8% in July 2009
Cause: The page's figures were typed into its text. The source re-estimates the model's whole history with each monthly release, and the release the site took in on November 2, 2026 moved them. The 2001 peak went from 30.4% (April 2001) to 28.1% (June 2001), so "near 30%" no longer held. June and July 2009 went from 62.3% and 2.5% to 60.8% and 2.8%. December 2000 went from 10.0% to 9.2%, so it left the Low band's list of months outside a recession ("December 2000–January 2001" now reads "January 2001"). The page stated the old values until this fix is deployed. Every figure on the page is now read from the data, so the next revision updates the page. The other statements, such as seven of the eight recessions above 50%, still hold.
Where: /indicator/chauvet_piger_recession_prob
Now caught by: the nightly claims re-check, on a copy of the November 2 data · record CP-REVISE

Found November 1, 2026Resolved the same dayserved sentence · correction

The GDPNow page called the 2020 shutdown quarter its record's largest move

Read: ~~It was the record's second-largest miss, in its largest move.~~
Reads now: It was the record's second-largest miss, in its deepest fall.
Cause: The Q2 2020 card said that quarter was the largest move in the GDPNow record. Over the quarters GDPNow has nowcast, the next quarter moved further: GDP rose 34.9% (annualized) in Q3 2020, against the 28.0% fall in Q2. Q2 2020 is the record's deepest fall, and its miss is the second-largest, as the card says. The sentence ran from the write-up's October 2026 release until this fix is deployed, and is now tested every night.
Where: /indicator/gdpnow
Now caught by: the registration of sentences that assert a record without a typed figure · record CLAIMS-WORDS-2

Found November 1, 2026Resolved the same dayserved sentence · correction

The Housing Cycle page printed internal series names in its stale-inputs line

Read: ~~Carried (stale under the cadence rule): [three internal series names, joined by commas] · coverage 67% current.~~
Reads now: Carried (stale under the cadence rule): Home prices (FHFA), Mortgage delinquency and Residential construction spending · coverage 67% current.
Cause: The line joined the inputs' internal names, not the names the page shows for them. It appeared whenever an input was stale under the cadence rule, from September 4, 2026 until this fix is deployed. It now reads each input's name from the page's own list of inputs.
Where: /housing-cycle
Now caught by: the site's internal-name scan, read row by row · record ASOF-MARKERS

Found November 1, 2026Resolved the same dayserved sentence · correction

The JOLTS layoffs page's moderate band left out November 2020

Read: ~~Moderate — 54 months since 2000, in 2000–05, 2007–10 and mid-2020" and "as in 54 months since 2000 (2000–05, 2007–10 and mid-2020)~~
Reads now: Moderate — 54 months since 2000, in 2000–05, 2007–10 and 2020" and "as in 54 months since 2000 (2000–05, 2007–10 and 2020)
Cause: The band's 2020 months are May, June and November, and "mid-2020" left November out. The years were typed beside a served count, so nothing checked them against the band's months. They are now served from the same band as the count.
Where: /indicator/jolts_layoffs
Now caught by: the design partner's review of the sentences that type a time window the data can outgrow · record SYNTH-17

October 2026

Found October 30, 2026Resolved the same dayserved sentence · correction

The State of the Economy's cards and What changed table read the months as first published

Read: ~~"↓ Deteriorating" and "↓ 2.2 · vs Sep 2026" on the Inflation Regime card, and "42.5 · 31.7 · 29.5", a change of −13.0, in the What changed table.~~
Reads now: "→ Stable" and "↓ 15.9 · vs Sep 2026" beside "Deteriorating rapidly", and "33.1 · 45.4 · 29.5", a change of −3.6, in the What changed table.
Cause: From October 29 the Reading Today sentence compared the months as first published, but the cards' badges and changes, the trajectory sentences and the table still read today's revised months. So the page told two stories about one gauge. They now read the same months: each earlier month as the page showed it when it settled, and today's live reading. Momentum's three-month change reads +3.2 (was +1.4), Forward Stress's +3.0 (was −1.9).
Where: /state-of-the-economy
Now caught by: the design partner's review of the October 30 correction · record SOE-VELOCITY

Found October 30, 2026Resolved the same dayserved sentence · correction

Reading Today judges a gauge's monthly move on the months as first published

Read: ~~Last month's 14.2-pt drop in Inflation Regime has not extended — this month's move is −2.2 pts, back inside the normal range.~~
Reads now: Inflation Regime is deteriorating exceptionally fast — down 15.9 pts in the past month, a pace seen in roughly 1% of months in this gauge's history.
Cause: The sentence compared today's revised months. The September core PCE print of October 29 restated September's score from 45.8 to 31.7, and the sentence read the restatement as "last month's drop" while the live gauge fell about 15 points in a day. On what the page had shown, September stood at 45.4 and the fall happened this month. The velocity sentences now compare the months as first published with today's live reading, so the figure moves with the reading.
Where: /state-of-the-economy
Now caught by: the owner's reading of the live page on October 29–30 · record SOE-VELOCITY

Found October 30, 2026Resolved the same dayarchive · correction

The scorecard's public ledger repository showed test data twice; the site's own record never moved

Read: "ledger 2026-10-28: 6 calls, 0 resolutions" and "ledger 2027-01-05: 0 calls, 3 resolutions" in the public repository, from October 29 05:43 to 16:46 UTC and again from October 30 05:43 UTC until the repair. The Wayback Machine captured that ledger once.
Reads now: the ledger's 11 lines as the site wrote them, restored by one ordinary commit; the five test-data commits stay in the repository's history and its README names them
Cause: The nightly check's tests on the production box ran as the site's user, and one test's night step fell back to the real clone and deploy key. The site's own ledger file was never touched; its hashes stand.
Where: /scorecard
Now caught by: the owner, reading the public repository's history; the archive now refuses every clone, key and remote outside the night step, and the nightly's tests get a scratch home of their own · record SCORE-3

Found October 29, 2026Resolved the same dayfigures · new data

The G7 CLI page's revision figures are served from the archive of first prints

Read: ~~the August 2026 figure arrived on September 12" · "of the 32 first prints the site has archived since early 2024, 29 were later changed, one by nearly a point (March 2025, from 100.6 to 99.6)~~
Reads now: "the September 2026 figure arrived on October 15, 2026" · "of the 33 first prints the site has archived since early 2024, 33 were later changed, the largest by 1.0 points (March 2025, from 100.6 to 99.6)". Each figure is served from the archive of first prints, so the next vintage moves it.
Cause: The OECD's vintage of October 29 added September 2026 and revised every month since January 2024; the sentence typed its counts, which were true when written on October 3 and wrong from that morning. The archive's start, "April 2024" and "from 2024", is served the same way; the arrival day is the one the archive holds.
Where: /indicator/g7_cli
Now caught by: the claims re-check of October 29 (the page's "nearly all" claim broke on a count the sentence never typed; its test now reads the sentence's own words) · record G7-CLAIMS

Found October 27, 2026Resolved the same dayfigures · correction

A gauge figure prints at its page's precision, never on a band edge it has not reached

Read: ~~"neutral · 59.95 / 100" on the landing and the State of the Economy card; "60.0" beside the neutral reading in the What-changed table~~
Reads now: "59" on the landing and the card; "59.9" in the What-changed table
Cause: Forward Stress stood just under the healthy edge at 60; the first rule kept adding decimals until the figure left the edge, and the table rounded to 60.0 beside the word neutral.
Where: 3 pages: / · /how-to-read · /state-of-the-economy
Now caught by: the owner's review of the live reading; now the gauge-figure rule's property test and the nightly's gauge-figures check · record READ-2

Found October 27, 2026Resolved the same dayfigures · correction

Eight market pairs tried: one held, seven did not; the home strip's count renamed

Read: ~~All 8 classic market pairs tried, such as copper against gold, did not hold up (the record)." · "43 indicators feed the three composites~~
Reads now: Of 8 classic market pairs tried, such as copper against gold, 1 held (the yield curve) and 7 did not (the record)." · "43 indicators feed the site's composites · 23 of them feed the three gauges
Cause: The failure count took in a relationship retired before the September test and left out the yield curve, which held; the home strip called the indicators feeding any composite "the three composites".
Where: 5 pages: /methodology · /about · /yield-curve/methodology · /validation · /
Now caught by: an outside review of the site's copy; both counts are now served from one function each and tested · record COPY-FIX-3

Found October 27, 2026Resolved the same daywords · correction

Internal names gone from the pages, each said in words

Read: ~~computed daily by the [an internal job's name] backfill job.~~
Reads now: computed daily by the site from the two published rates.
Cause: Stored-series keys, job names, file names and record paths sat in notes, update lines, table cells and tooltips that the earlier guard, which read only registered sentences, never saw; 291 on 26 pages.
Where: 26 pages
Now caught by: an outside reviewer's repeated findings; now a scan of every rendered page's text and attributes, in the test suite and the nightly · record IDENT-1

Found October 27, 2026Resolved the same daydates · correction

GDPNow's "as of" names the vintage its shown value comes from

Read: ~~"nowcast · as of" followed by the day the site received the value, whenever a newer vintage was on file that the site had not stored~~
Reads now: "nowcast · as of" followed by the date of the vintage whose value the cell shows, whatever newer vintages are on file
Cause: The stamp read the newest vintage on file only when the stored value equalled it, so a newer vintage the site had not stored yet pushed the date back to the day the value arrived.
Where: 2 pages: /heatmap · /indicator/gdpnow
Now caught by: a review of the stamp after the vintages backfill; pinned by a test that the cell's value and its date come from one vintage row · record FOLLOWUP-1

Found October 27, 2026Monitoringdata feeds · new data

The commercial-paper spread's nonfinancial rate has paused in the Fed's series

Status: The nonfinancial rate has had no print since October 13 while the financial one continues; the health page names the spread until the Fed prints again, under a 54-business-day allowance from its own gap record.
Cause: The rate is a thinly traded index the Fed leaves blank on days without a print; its record holds gaps three to five times this long, every one followed by fresh prints, so nothing was changed.
Where: 2 pages: /indicator/cp_fin_nonfin_spread · /heatmap
Now caught by: the staleness rule on the health page, which measures each series' age against its allowance · records STALE-2, FOLLOWUP-1

Found October 26, 2026Resolved the same daydates · correction

GDPNow's date line stood at "updated Jul 30" while the nowcast moved

Read: ~~Data through: Q3 2026 · updated Jul 30~~
Reads now: Q3 2026 nowcast · as of Oct 20
Cause: A nowcast is a vintage series stored as one row per quarter, so each new value overwrote the last and the line kept the day the row was first written; every vintage is now kept and the line names the newest.
Where: 2 pages: /heatmap · /indicator/gdpnow
Now caught by: an outside report of a stale date line; now the cadence staleness rule on the health page, which allows a nowcast ten days · record DATA-STALE-1

Found October 26, 2026Resolved the same daydates · correction

"Last published" on the State of the Economy named the data month, not the release month

Read: ~~last published Apr 2026~~
Reads now: last published Aug 2026
Cause: The line printed each carried input's observation month under the words "last published"; it now reads the day the input's newest observation was published, from the one release record, for every carried input.
Where: /state-of-the-economy
Now caught by: an outside report comparing the line with the heatmap's "updated" date; the fixture test that pinned the old month was itself the bug and moved · record DATA-STALE-1

Found October 26, 2026Resolved the same daydates · correction

The Buffett sigma's "Data through" claimed a month the data had not finished

Read: ~~Data through: October 2026 · updated Oct 1~~
Reads now: Data through: Oct 23, 2026 · updated Oct 23
Cause: The open month's row is rewritten every trading day from the last daily reading, so "October 2026" named an unfinished month and "updated Oct 1" the row's first write, three weeks before the value shown.
Where: 2 pages: /heatmap · /indicator/buffett_trend_deviation
Now caught by: the same outside report; "Data through" now prints at the precision of the newest observation the reading uses, checked by the display audit · record DATA-STALE-1

Found October 26, 2026Resolved the same dayfigures · correction

Forward Stress read "neutral · 60 / 100": the figure contradicted its word

Read: ~~neutral · 60 / 100~~
Reads now: neutral · 59 / 100
Cause: The stored score 59.4983 was rounded twice (to 59.5, then to 60) and the marker on /how-to-read was drawn at 60, inside the healthy band, under the word neutral; the figure is now rounded once from the stored value.
Where: 2 pages: / · /how-to-read
Now caught by: the nightly check's first night on the live site, failing the test that the landing's word matches the ladder at its figure · record READ-2

Found October 26, 2026Resolved the same daywords · correction

Three indicator addresses renamed to the series each page shows

Read: ~~/indicator/ny_fed_recession_prob" · "/indicator/auto_delinquency" · "/indicator/ted_spread~~
Reads now: /indicator/chauvet_piger_recession_prob" · "/indicator/all_loans_delinquency" · "/indicator/sofr_tbill_spread
Cause: Each address carried a stored key naming a series the page does not show; the pages serve the Chauvet–Piger recession probability, delinquency on all bank loans and the SOFR–T-bill spread, and the old addresses redirect.
Where: 3 pages: /indicator/chauvet_piger_recession_prob · /indicator/all_loans_delinquency · /indicator/sofr_tbill_spread
Now caught by: an outside reviewer; pinned by tests that each old address answers a redirect to its new one, with any Time Machine date kept · record SLUG-1

Found October 25, 2026Resolved the same dayfigures · correction

The count of indicators feeding the composites was typed, not counted

Read: ~~45 of them feed the three composites~~
Reads now: 43 of them feed the three composites
Cause: The figure (46 on dev, 45 on the live site) was a hand count from an earlier audit, typed with no code behind it; the count is now read from the engines' own rosters, published heatmap indicators only.
Where: 2 pages: / · /about
Now caught by: the typed-figure rule — a figure no code reproduces is stopped at the gate — and a data test that the strip's count equals the rosters' · record LP-B1

Found October 23, 2026Resolved the same daywords · correction

Release-schedule phrases claimed a timing the record did not keep

Read: ~~Monthly (released ~2 weeks after month end)~~
Reads now: Monthly
Cause: An unqualified schedule phrase claims the timing for every release; none held for more than 85% of the stored observations on the corrected calendar, so each is cut to its cadence and the parts that hold.
Where: 43 indicator pages · /data-sources
Now caught by: the sweep of schedule phrases against the first-print calendar, read under the margin rule (an unqualified phrase must hold above 96%) · record LIST-CLOSE-1

Found October 23, 2026Resolved the same dayfigures · correction

Counts at a band's edge follow the band's own rule; AAII neutral weeks 788 → 765

Read: ~~When bulls and bears were within 10 points of each other — 788 weeks since~~
Reads now: When bears were up to 10 points ahead or bulls less than 10 ahead — 765 weeks since
Cause: The neutral band is strict at +10 (the bullish band holds +10), but the count read "10 or below" and took in 23 weeks sitting exactly on the edge; every count at an edge now uses its band's side, and the words say which.
Where: 16 indicator pages
Now caught by: a census of every served count's bound against its band's rule, and a test that generates every band label from the colour rule · record LADDER-LABEL-1

Found October 21, 2026Resolved the same daydates · correction

Release dates from the first-print record: no reading dated before it was published

Read: ~~Each month's reading is published about 5 weeks after the month ends.~~
Reads now: Each month's reading has come about a month after the month ends since 2019 (a median 32 days from 2019 through 2025; about nine weeks, a median 62 days, from 2012 through 2018).
Cause: For 91 series the calendar dated each month by a lag measured on the site's own few receipts, shorter than the producer's; 1,813 observations were dated before their release and 23 served sentences moved.
Where: 12 indicator pages · /time-machine
Now caught by: measuring the calendar against ALFRED's first prints for every FRED-backed series; the release record is now generated from that archive and tested · record CAL-2

Found October 21, 2026Resolved the same daydates · correction

ADP dated by its own release record; its timing sentence says "usually"

Read: ~~Monthly (first Wednesday of each month)~~
Reads now: Monthly, usually the Wednesday two days before the BLS jobs report
Cause: The calendar put ADP on the 5th of the following month from the site's five receipts, on ADP's day in 11 of 50 months and before it in 6; the phrase held for 61% of months even on the corrected calendar.
Where: 2 pages: /indicator/adp_private_payrolls · /data-sources
Now caught by: an earlier gate's finding that the phrase held for 14.5% of months; the calendar is now measured against ALFRED's record of ADP's 189 releases · record ADP-CAL-1

Found October 19, 2026Resolved the same dayfigures · correction

Monthly and weekly sentences re-read; recession lead times counted per recession

Read: ~~It was above +0.5, and already published, before 2 of the 7 recessions since 1973 began.~~
Reads now: It was above +0.5, and already published, before 1 of the 7 recessions since 1973 began.
Cause: The registered test read the latest print before each recession's peak month, not before its first month; of 98 sentences, 45 figures are now computed, 42 dated to their window and 4 corrected.
Where: 33 indicator pages
Now caught by: the census of monthly and weekly sentences re-read against the record, with a data test on every sentence and the lead counts computed recession by recession · record CLAIMS-DRIFT-2D

Found October 18, 2026Resolved the same daydata feeds · correction

A revision older than the read window left a step in real retail sales

Read: ~~By December nominal growth had slowed to 3.5% and real sales were 1.2% lower: inflation had supplied all of the nominal gain.~~
Reads now: By December nominal growth had slowed to 3.5% and real sales were 2.0% lower: inflation had supplied all of the nominal gain.
Cause: FRED revised the series back to 2015 on October 14, but the nightly read only the last twelve observations, so the stored history kept a step of −4.2% where FRED had −0.2%. The read now takes 24 and refreshes on an edge revision.
Where: 2 pages: /indicator/real_retail_sales · /indicator/nfci
Now caught by: a check of every stored FRED series against FRED's current vintage (128 months differed); the five figures are now computed from the data, true on the old and the refreshed record alike · records DOC-19, INGEST-EDGE-1

Claims that moved

Every sentence on the site with a figure in it is a registered claim with a test. At each stored re-check every test is run again; a claim moves when its verdict changes or the figure it pins moves. The table lists every stored re-check, newest first; "the list" opens a re-check's moved claims beneath it.

Re-check	Reason	Checked	Moved	Corrected	New data	Open	List
November 8, 2026	nightly	7,399	0	0	0	0	—
November 7, 2026	manual	7,399	0	0	0	0	—
November 7, 2026	manual	7,398	0	0	0	0	—
November 7, 2026	nightly	7,371	0	0	0	0	—
November 6, 2026	manual	7,371	0	0	0	0	—
November 6, 2026	manual	7,339	2	0	2	0
Services Payrolls · the verdict changed · held → broken · new data Services Payrolls · the verdict changed · held → broken · new data
November 6, 2026	manual	7,301	0	0	0	0	—
November 6, 2026	nightly	7,301	0	0	0	0	—
November 5, 2026	manual	7,301	0	0	0	0	—
November 5, 2026	manual	7,301	0	0	0	0	—
November 5, 2026	manual	7,301	0	0	0	0	—
November 5, 2026	manual	7,301	0	0	0	0	—
November 5, 2026	manual	7,242	0	0	0	0	—
November 5, 2026	manual	7,226	0	0	0	0	—
November 5, 2026	nightly	7,182	0	0	0	0	—
November 4, 2026	manual	7,156	0	0	0	0	—
November 4, 2026	manual	7,157	5	0	5	0
Private Payrolls · the verdict changed · held → broken · new data Copper Price · the verdict changed · held → broken · new data Copper Price · the verdict changed · held → broken · new data Copper Price · the verdict changed · held → broken · new data Copper Price · the verdict changed · held → broken · new data
November 4, 2026	nightly	7,126	0	0	0	0	—
November 3, 2026	manual	7,126	0	0	0	0	—
November 3, 2026	manual	7,126	0	0	0	0	—
November 3, 2026	manual	7,117	0	0	0	0	—
November 3, 2026	manual	7,117	0	0	0	0	—
November 3, 2026	manual	7,126	0	0	0	0	—
November 3, 2026	nightly	7,110	0	0	0	0	—
November 2, 2026	manual	7,110	0	0	0	0	—
November 2, 2026	manual	7,110	0	0	0	0	—
November 2, 2026	manual	7,106	0	0	0	0	—
November 2, 2026	monthly	6,970	8	7	1	0
OECD Leading Indicator (US) · the figure moved · Jun 2020: 2.00 → 1.99 · new data Smoothed Recession Probability · the verdict changed · held → broken · corrected Smoothed Recession Probability · the verdict changed · held → broken · corrected Smoothed Recession Probability · the figure moved · inside 13 → 14; count 15 → 16 · corrected Smoothed Recession Probability · the verdict changed · held → broken · corrected Smoothed Recession Probability · the verdict changed · held → broken · corrected Smoothed Recession Probability · the verdict changed · held → broken · corrected Smoothed Recession Probability · the verdict changed · held → broken · corrected
November 2, 2026	nightly	6,970	1	0	1	0
OECD Leading Indicator (US) · the figure moved · Jun 2020: 2.00 → 1.99 · new data
November 1, 2026	manual	6,970	1	0	1	0
OECD Leading Indicator (US) · the figure moved · Jun 2020: 2.00 → 1.99 · new data
November 1, 2026	manual	6,970	1	0	1	0
OECD Leading Indicator (US) · the figure moved · Jun 2020: 2.00 → 1.99 · new data
November 1, 2026	manual	6,964	1	0	1	0
OECD Leading Indicator (US) · the figure moved · Jun 2020: 2.00 → 1.99 · new data
November 1, 2026	manual	6,964	4	0	4	0
Valuations · the verdict changed · held → broken · new data Valuations · the verdict changed · held → broken · new data Valuations · the verdict changed · held → broken · new data OECD Leading Indicator (US) · the figure moved · Jun 2020: 2.00 → 1.99 · new data
November 1, 2026	nightly	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 31, 2026	manual	6,959	0	0	0	0	—
October 30, 2026	manual	6,959	0	0	0	0	—
October 30, 2026	manual	6,959	0	0	0	0	—
October 30, 2026	manual	6,959	2	0	2	0
Real PCE Services · the figure moved · 2.460 → 2.465 · new data Real PCE Services · the figure moved · 2.460 → 2.465 · new data
October 29, 2026	manual	6,959	2	0	2	0
Real PCE Services · the figure moved · 2.460 → 2.465 · new data Real PCE Services · the figure moved · 2.460 → 2.465 · new data
October 29, 2026	manual	6,959	2	0	2	0
Real PCE Services · the figure moved · 2.460 → 2.465 · new data Real PCE Services · the figure moved · 2.460 → 2.465 · new data
October 29, 2026	manual	6,955	3	0	3	0
G7 Leading Indicator (OECD) · the verdict changed · held → broken · new data Real PCE Services · the figure moved · 2.460 → 2.465 · new data Real PCE Services · the figure moved · 2.460 → 2.465 · new data
October 28, 2026	manual	6,955	0	0	0	0	—
October 28, 2026	manual	6,955	0	0	0	0	—
October 27, 2026	manual	6,955	0	0	0	0	—
October 27, 2026	manual	6,956	0	0	0	0	—
October 26, 2026	manual	6,957	0	0	0	0	—
October 26, 2026	manual	6,957	0	0	0	0	—
October 24, 2026	deploy	6,957	0	0	0	0	—
October 23, 2026	deploy	6,957	0	0	0	0	—
October 23, 2026	deploy	6,957	0	0	0	0	—
October 23, 2026	deploy	6,957	0	0	0	0	—
October 22, 2026	deploy	7,031	0	0	0	0	—
October 22, 2026	deploy	7,031	0	0	0	0	—
October 21, 2026	deploy	6,951	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 20, 2026	deploy	6,887	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 20, 2026	event	6,877	3	3	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected Single-Family Permits · the verdict changed · held → broken · corrected
October 20, 2026	deploy	6,877	3	3	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected Single-Family Permits · the verdict changed · held → broken · corrected
October 19, 2026	deploy	6,877	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 19, 2026	deploy	6,877	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 18, 2026	deploy	6,758	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 18, 2026	event	6,345	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 18, 2026	deploy	6,345	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 16, 2026	deploy	4,589	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 16, 2026	deploy	4,589	2	2	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Capacity Utilization · the figure moved · 79.81 → 79.80 · corrected
October 15, 2026	deploy	4,340	6	6	0	0
5yr Breakeven Inflation · the figure moved · count 283 → 284; correlation 0.62 → 0.63 · corrected Real Retail Sales · the verdict changed · held → broken · corrected Real Retail Sales · the verdict changed · held → broken · corrected Existing Home Sales · the verdict changed · held → broken · corrected Existing Home Sales · the verdict changed · held → broken · corrected Nasdaq 100 RSI · the verdict changed · held → broken · corrected
October 13, 2026	deploy	3,890	0	0	0	0	—
October 12, 2026	deploy	3,886	0	0	0	0	—
October 11, 2026	deploy	3,100	17	6	11	0
State of the Economy · the figure moved · 16,817 → 16,834 · new data State of the Economy · the figure moved · −0.03 → −0.01 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.68 → 3.59 · new data OECD Leading Indicator (US) · the figure moved · 32 → 33 first prints · new data OECD Leading Indicator (US) · the verdict changed · held → broken · new data OECD Leading Indicator (US) · the verdict changed · held → broken · corrected OECD Leading Indicator (US) · the verdict changed · held → broken · corrected OECD Leading Indicator (US) · the verdict changed · held → broken · corrected OECD Leading Indicator (US) · the verdict changed · held → broken · corrected OECD Leading Indicator (US) · the verdict changed · held → broken · corrected OECD Leading Indicator (US) · the verdict changed · held → broken · corrected OECD Leading Indicator (US) · the figure moved · 32 → 33 first prints · new data Jobless Claims (Initial) · the figure moved · 3,117 → 3,118 first prints · new data Copper Price · the figure moved · Jun 2026: 13,552.04 → Sep 2026: 14,474.00 · new data
October 11, 2026	deploy	3,092	7	0	7	0
State of the Economy · the figure moved · 16,817 → 16,834 · new data State of the Economy · the figure moved · −0.03 → −0.01 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.68 → 3.59 · new data Jobless Claims (Initial) · the figure moved · 3,117 → 3,118 first prints · new data
October 10, 2026	deploy	3,092	7	0	7	0
State of the Economy · the figure moved · 16,817 → 16,834 · new data State of the Economy · the figure moved · −0.03 → −0.01 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.68 → 3.59 · new data Jobless Claims (Initial) · the figure moved · 3,117 → 3,118 first prints · new data
October 10, 2026	manual	3,085	8	0	8	0
State of the Economy · the figure moved · 16,817 → 16,834 · new data State of the Economy · the figure moved · −0.03 → −0.01 · new data Business Loans · the verdict changed · held → broken · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.68 → 3.59 · new data Jobless Claims (Initial) · the figure moved · 3,117 → 3,118 first prints · new data
October 9, 2026	manual	3,085	7	0	7	0
State of the Economy · the figure moved · 16,817 → 16,834 · new data State of the Economy · the figure moved · −0.03 → −0.01 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.68 → 3.59 · new data Jobless Claims (Initial) · the figure moved · 3,117 → 3,118 first prints · new data
October 8, 2026	manual	3,085	8	2	6	0
State of the Economy · the figure moved · 16,817 → 16,831 · new data State of the Economy · the figure moved · −0.03 → −0.01 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.677 → 3.684 · new data WTI Crude Oil · the figure moved · count 61 → 60 · corrected WTI Crude Oil · the figure moved · count 61 → 60 · corrected
October 7, 2026	event	3,080	7	2	5	0
State of the Economy · the figure moved · 16,817 → 16,831 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.677 → 3.684 · new data WTI Crude Oil · the figure moved · count 61 → 60 · corrected WTI Crude Oil · the figure moved · count 61 → 60 · corrected
October 7, 2026	event	3,080	7	2	5	0
State of the Economy · the figure moved · 16,817 → 16,831 · new data Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.677 → 3.684 · new data WTI Crude Oil · the figure moved · count 61 → 60 · corrected WTI Crude Oil · the figure moved · count 61 → 60 · corrected
October 7, 2026	event	1,993	6	2	4	0
Excess Bond Premium · the figure moved · +3.40 → +3.41 · new data Excess Bond Premium · the figure moved · +1.19 → +1.19, the same as printed · new data Excess Bond Premium · the figure moved · −1.08 → −1.08, the same as printed · new data GDPNow Nowcast · the figure moved · Q3 2026: 3.677 → 3.684 · new data WTI Crude Oil · the figure moved · count 61 → 60 · corrected WTI Crude Oil · the figure moved · count 61 → 60 · corrected
October 5, 2026	event	1,989	0	0	0	0	—
October 5, 2026	event	1,989	45	36	9	0
Manufacturing Weekly Hours · the verdict changed · held → broken · corrected Manufacturing Weekly Hours · the figure moved · 1,136 → 1,137 first prints · new data Fed Funds Rate · the verdict changed · held → broken · corrected Fed Funds Rate · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · new data Heavy Truck Sales · the verdict changed · held → broken · corrected Heavy Truck Sales · the verdict changed · held → broken · corrected Services Weekly Hours · the verdict changed · held → broken · corrected Services Weekly Hours · the verdict changed · held → broken · corrected Services Weekly Hours · the verdict changed · held → broken · corrected Services Weekly Hours · the verdict changed · held → broken · corrected ADP Private Payrolls · the verdict changed · held → broken · corrected ADP Private Payrolls · the verdict changed · held → broken · corrected Goods Payrolls · the verdict changed · held → broken · corrected Job Openings (JOLTS) · the figure moved · 2.96 → 2.98 · corrected JOLTS Quit Rate · the figure moved · 2.96 → 2.98 · corrected JOLTS Quit Rate · the figure moved · 2.96 → 2.98 · corrected Labor Force Participation · the verdict changed · held → broken · new data Labor Force Participation · the verdict changed · held → broken · corrected Part-time for Economic Reasons · the verdict changed · held → broken · new data Part-time for Economic Reasons · the verdict changed · held → broken · corrected Nonfarm Payrolls · the verdict changed · held → broken · corrected Nonfarm Payrolls · the verdict changed · held → broken · corrected Nonfarm Payrolls · the verdict changed · held → broken · corrected Nonfarm Payrolls · the figure moved · count 143 → 144 · corrected Nonfarm Payrolls · the figure moved · 1,052 → 1,053 first prints · new data Private Payrolls · the verdict changed · held → broken · corrected Private Payrolls · the verdict changed · held → broken · corrected Private Payrolls · the verdict changed · held → broken · new data U-6 Unemployment Rate · the verdict changed · held → broken · corrected U-6 Unemployment Rate · the verdict changed · held → broken · corrected U-6 Unemployment Rate · the verdict changed · held → broken · new data Wage Growth · the verdict changed · held → broken · corrected Current housing inventory · the verdict changed · held → broken · new data Current housing inventory · the verdict changed · held → broken · new data Residential Construction Jobs · the verdict changed · held → broken · corrected Residential Construction Jobs · the verdict changed · held → broken · corrected Residential Construction Jobs · the verdict changed · held → broken · corrected Fed Balance Sheet · the verdict changed · held → broken · corrected

A corrected claim is one whose sentence was changed after it moved. A new data claim is one whose figure a later release moved and the record accepted, or that holds again. An open claim still fails its check at the latest re-check. The nightly check re-runs every claim on a copy of the data; the re-checks stored here are the nightly one, the monthly one, the one after every deploy and the owner's own. A correction is listed the day it is made; a claim that moved is listed with the re-check that found it.

Methodology · Validation · Changelog