A site whose pitch is accountability keeps its errata in public. This page lists every correction to a number or label the site had already published — newest first, each with what was shown, what it became, the date and the changelog entry it was logged in. Method changes that move a published figure (the publication-dated gauge rebuild of September 1, 2026; the v1.2 and v1.3 constructions of September 2; nowcast v1.2 and forecast v2.1) are listed alongside data corrections (the frozen Manheim used-vehicle months of July and August 2026, a hero coverage figure that read 39% instead of 43% for ten minutes on September 4, a hard-coded source count, an official CPI YoY shown as 4.47% instead of 4.25% in July, canonicals that pointed at the old domain) and check corrections (the status page's amber rows of September 7). New data replacing old data is not a correction and is not listed. Every entry is versioned and dated in the engine changelog; historical readings are stored vintage-true and never rewritten.
10 corrections, 5 label or copy fixes, 6 method changes, 1 known issue, 1 check or process fix, 6 published numbers. Each entry records what was shown, what it became, where on the site, and the dated changelog or build-log entry it was logged in.
| Date | Kind | Where | What was shown | What it became |
|---|---|---|---|---|
| October 5, 2026 | correction | the Daily Brief archive (/brief/<date>) and its feed: twenty wire paragraphs from August 11 to October 1, 2026 | Fifteen headlines said the reading "holds" on days it moved (by 0.01 to 0.13 points); three paragraphs ended in leftover code characters and three in a stray quotation mark; one printed a raw data field name ("dod_pp 0"); the September 26 paragraph said the Nowflation Forecast of 3.581% was "above" the Street consensus of 3.7%; the August 30 paragraph called the in-house nowcast of 3.226% the Street's (the Street consensus was 3.4%); and the September 13 paragraph named the in-house nowcast by an internal name. | Each paragraph is corrected and carries a dated note under it saying what changed; the rest of each paragraph is as published. New paragraphs are checked for these errors before they are stored. |
| October 4, 2026 | correction | /analysis: the May, June, July and August 2026 print-day reports | The four reports, drafted by the Nowflation engine and published under the machine label, were written in the first person. The August report said "I called 3.4%" (the Nowflation Forecast locked the evening before was 3.34%, fourth of five) and that Nowflation "beat the field by a wide margin over time" (on the live record its average miss is 0.248 points against the Street's 0.086). They cited track records "across 86" and "88" prints that match no graded record, quoted CPI Gauge readings from methods since retired (1.79% on July 14, where the current method reads 3.11%) with "gap signal" statistics and component contributions built on them, showed the June forecast at the frozen 3.895% row (the receipt says 3.935%, fourth of five, not third), listed July's 3.421% as an ordinary call (its receipt was written after the release), called the Cleveland Fed's nowcast a Nowflation Forecast and used internal names. | Each report reads in the third person under a dated correction note listing what changed. Gauge readings are restated on the current method and the statistics built on the old readings are removed; the scoreboard rows, ranks and receipt notes follow the receipts; the record claims are removed; the headlines no longer say the Gauge "called" the prints. The print figures, forecasts as of each report's date and fuel forwards are unchanged. |
| October 4, 2026 | correction | three articles: /articles/the-fed-hiked-into-an-oil-shock, /articles/fed-cannot-hike-its-way-out-of-an-oil-shortage and /articles/new-york-wealth-migration | The September 23 article twice called the September 16 hike "yesterday's". The oil-shortage article said August's 0.3% monthly core CPI matched the consensus estimate (the consensus was 0.2%; the 2.4% annual rate did match), embedded a live chart of the 30-year mortgage rate that its publisher does not allow to be embedded (the chart showed "not available as an embed"), and linked a CFPB report at an address that no longer works. The New York migration article's summary said $9.9 billion left "last year" (the IRS figures cover 2022 to 2023) and gave the 99,404 millionaire returns of 2024 as the base of that outflow. | The dates, the core CPI sentence, the CFPB link and the summary are corrected, the mortgage chart is a link to the mortgage page, and each article carries a dated correction note and an updated date. An article that embeds a series the site may not redistribute now gets a link to the series page instead of the chart. |
| October 2, 2026 | correction | the per-print tables on the homepage and /scoreboard, the post-mortems, /cpi-release-dates, the /cpi-preview result pages, /fed-watch, /oil-vs-cpi, /i-bonds and /cola | The per-print tables, the post-mortems and the Nowflation call column on /cpi-release-dates showed the June 2026 Nowflation Forecast at 3.90% (the scoreboard row frozen at grading, 3.895%, error +0.395pp) and the July 2026 figure of 3.42% as ordinary live calls ("July 2026: a hit"). The receipts and the leaderboard graded June on its receipt (3.935%, error +0.435pp), and the July receipt was written after the print. The closest-call rates counted prints where the in-house model's backtest was closest, so the Street read 46% and the Cleveland Fed 27%. /fed-watch and /oil-vs-cpi called the in-house model's 3.46% headline and 2.25% core the Nowflation Forecast when the published call was 3.59% and 2.45%, and /i-bonds and /cola projected from the in-house figure as "the site's published call". | June 2026 is graded on its receipt, with the frozen row shown beside it, and July 2026 is marked as written after the print and is not counted as a live call. Closest-call rates count only the calls on the record before each print (on the same data: the Street 61%, the Cleveland Fed 39%). /fed-watch, /oil-vs-cpi, /i-bonds and /cola use the Nowflation Forecast and give the in-house model's figure beside it. No receipt or scoreboard row was changed. |
| September 30, 2026 | correction | /cart and the item price trackers (/eggs, /gas, /ground-beef, /chocolate, /coffee and the rest), with their titles, descriptions and crawler text | /cart priced a "Soft drinks 12-pack" at $0.60, which is the BLS price of one 12-ounce can ($0.603 for August 2026), and listed frozen orange juice concentrate as "Orange juice" with its July price under an August 2026 heading. Eleven grocery tracker titles said "Today (September 2026)" over August BLS shelf prices, because the advertised-price index was stored for September while the month was still under way; the "same weeks last year" change beside that index was August's (eggs -19.1% where September's own change was -10.9%, ground beef +7.8% where it was +3.2%), and the index was called "USDA's" although Nowflation builds it from USDA reports, for whole categories. The /gas year-over-year tile was blank and credited to the BLS, and its 24 daily history rows all read "September 2026". /chocolate said chocolate prices rose over a chocolate chip cookie price, the "Upstream" sentences used raw series names and called potato chips upstream of potatoes, and coffee's "2019 average" covered only October to December. | A 12-pack is twelve cans ($7.24 at August 2026 prices), the juice line is named frozen concentrate, and a cart is priced in the latest month every item has, with any item that holds it back named. Titles follow the BLS shelf price ("Egg Prices (latest: August 2026, BLS)"); the advertised-price index stores only months that have ended, any row stored for a month under way is removed at the connector's next run, and the same-weeks change is quoted only for its own month. /gas takes its year-over-year change and history from the EIA weekly average for regular (+43.2% for the week of September 28, 2026), dated by week. The priced item, units and roles are named, and a year's average needs all twelve months. No published reading, receipt or scoreboard row was involved. |
| September 30, 2026 | correction | /thanksgiving, /heating-season and /farm-to-shelf, and their crawler text | /thanksgiving gave food-at-home CPI as +2.1%, the seasonally adjusted series (the BLS publishes +2.2% for August 2026, not seasonally adjusted), compared an August basket with November 2019, the holiday-feature low, for "+60.4% vs 2019", and said the dinner was priced every month since 2018 although no total exists for December 2018 to August 2019, when the BLS published no ground coffee price. /heating-season showed index 99.5 and $920 to search engines and 99.4 and $921 on the page, and listed fuel oil among the CPI's energy services. /farm-to-shelf showed futures quoted in cents as dollars (wheat as $795.793), called the farmer's share exact while it divided the newest farm month by the newest shelf month, and its crawler text carried a broken sentence. | Food at home is the not seasonally adjusted series. The 2019 comparison uses the same season of 2019: for August 2026 prices, September 2019, the nearest 2019 month with every line priced (+49.2%, $75.42 against $50.55). The months with no total are named. The heating index and bill come from one unrounded calculation on the page and in the crawler text, and fuel oil is described as a household fuel the CPI counts with energy commodities. Futures carry the exchange's own units (cents per bushel, cents per pound), and the farmer's share uses the latest month both prices cover. No published reading, receipt or scoreboard row was involved. |
| September 30, 2026 | correction | /oil (reserves in the ground), its crawler text and the oil reserves series pages | The reserves section said the world held 264 billion barrels of proved oil reserves and had 8.7 years left at current production, with the United States at 7.5 billion barrels, Guyana first on the shortest-runway list at 0.0 years, and the figures credited to the Energy Institute / BP Statistical Review for both 2021 and 2020. The source table is the U.S. Energy Information Administration's, published by Our World in Data in cubic meters, and the figures had been read as barrels. | The figures are converted to barrels (1 cubic meter = 6.2898 barrels) and credited to the EIA, for 2020, the latest year the table covers every country (its 2021 figures leave out the United States). On September 29 data: world proved reserves about 1,662 billion barrels, the United States 47.1 billion, and a world runway of about 62 years at May 2026 production. Countries with no proved reserves in the table are no longer listed, and each runway uses the country's latest confirmed production month. The stored reserve rows were corrected in the database; no published reading, receipt or scoreboard row was involved. |
| September 30, 2026 | correction | /commodities, the /commodity pages, /energy, /heating-season, /fx, /oil, /oil-vs-cpi and the soybeans series page | Futures histories included Sunday-evening sessions, bars dated on US exchange holidays and closes of 0: cotton, aluminum, platinum and palladium showed a one-year low of 0, and WTI's one-year high of $115.58 was a partial Sunday session. The Day column could show the move since the same day's settlement, and contract rolls showed as price moves. /energy gave natural gas storage as 3.3% (its card) and 2.1% (its chart, the same calendar-week match /heating-season used) above its five-year average for the week of September 18 (the EIA reported 2.9%), and its rig cards stopped at February 2026. The dollar-index chart never loaded and the /fx dollar tile showed a dash. The series named Soybeans was soybean meal and fed the agricultural composites. /oil-vs-cpi printed full-history correlations under a "last 25y" label, called a month-to-date average "WTI now", and named 2009 as the only modern deflation; /oil named Trading Economics as its rig source and listed Syria from 2013. | Only daily settlements are stored and charted (WTI's one-year high on September 29 data: $112.95 on April 7). Day is measured from a settlement. A Day, Week or Month change that spans the day the WTI, RBOB gasoline or heating oil series switched to its next contract is marked "roll", with that day taken from the exchange's expiry calendar; Brent and natural gas changes are not marked, because the day their series switch could not be confirmed. The storage comparison uses the EIA method, each of the prior five years interpolated to the same date (3,351 against about 3,255 Bcf). Rig counts come from Baker Hughes, weekly for the US and Canada. The dollar index chart and tile load. The soybeans series is the soybean futures contract. The oil-vs-CPI panels show statistics for the 25 years they chart, the WTI figure is labeled as the month-to-date average it is, and the page lists 2009 and 2015. No published reading, receipt or scoreboard row was involved. |
| September 30, 2026 | correction | /series/manheim_muvvi (its title, description, chart and latest value) | The Manheim Used Vehicle Value Index series page showed a September 2026 full-month value of 206.2, which was Manheim's mid-month read for September (data through the 15th). Cox had not published a full-month September figure. The mid-month read had been filed in the full-month series before the connector gave it its own series, and the cleanup that removes such rows ran only on the days the connector polled. | The cleanup runs every day, so the full-month series ends with August 2026 until Cox publishes September. It also removes any full-month value stored before the 5th of the following month, since Cox publishes the full-month index on the fifth business day: August's mid-month 207.4, stored on September 1, and a 206.0 from Manheim's old index page (which stopped updating in December 2025) stored for July and August. The mid-month read stays in its own series, where Nowflation Realtime uses it and labels it as the mid-month read. Input ages on /realtime are measured from the observation in use and never shown below zero. No published reading, receipt or scoreboard row was involved. |
| September 30, 2026 | label or copy | the homepage fuel panel, the daily brief and its archive, the Outlook's fuel driver and the daily wire | The two-week pump-price forward was labeled as implied by RBOB futures, but since the v1.4 inputs fix it has been built from the EIA's New York Harbor spot gasoline price, which the EIA posts weekly (on September 30 the latest price was from September 22), and the homepage printed its formula with internal field names. The Outlook computed a second forward from the raw front-month RBOB contract (+13.4 cents beside the fuel panel's +18.5 cents when the audit checked). | The fuel panel and the brief name the price the forward is built from and the date of its latest observation, and the fuel panel states the method in words. The Outlook's fuel driver uses the fuel panel's forward and names the same price; the daily wire names it too. The forward's value itself is unchanged. |
| September 29, 2026 | correction | /gauge, /cpi/june-2026, the category pages (/food, /shelter, /energy, /core, /services), the CPI Gauge's live share, pace and day, week and month changes on every page that shows them, Realtime's week change, /nowflation-vs-truflation, the homepage and the print pages | From September 26 the /gauge page loaded blank for visitors (a render error; search engines still read its text), and /cpi/june-2026 had loaded blank since at least mid-September. After the September 28 switch the category pages kept the v1.4 construction (on September 29: food 2.48%, energy 20.26%, core 2.28%, services 2.13%, shelter 2.00%), their contributions did not add up to the reading, and each stepped at the January weight change. The CPI Gauge's live share still counted Kelley Blue Book's July figure as live after the method had set it aside for the BLS August print (43.5%), /gauge listed v1.4 component figures (new vehicles 1.58%), and the day, week and month changes and the seasonally adjusted pace compared v1.5 readings with v1.4 ones (+0.158pp on the week and +0.388pp on the month on September 29, most of it the switch). Realtime's week change was labeled the published change while it compared with a v2.2 reconstruction, /nowflation-vs-truflation labeled v2.2 reconstructions for September 20 to 27 (3.16%) as live readings, and the homepage said inputs covering 3.8% of the basket were past their refresh window. The June 2026 print page graded the 3.895% scoreboard row in its tables under a headline that quoted the 3.935% receipt. | Both pages render. The category pages are built on the v1.5 December chain with exact contributions that add up (on September 29: food 2.62%, energy 22.61%, core 2.12%, services 2.15%, shelter 2.02%). The live share counts only sources that price a component that day (39.7%), /gauge shows each component as v1.5 takes it with its exact share of the headline, the pace is measured on v1.5 at both ends, and every day, week and month change is like for like, with the move in the published readings stated beside it while the window spans the switch. Realtime shows both its like-for-like and its published change, the Truflation comparison reads Realtime as published (3.03% to 3.04% for September 20 to 27), and a source held because the BLS has a newer month is labeled held, not stale. Print pages grade the receipt locked before the release in every table, with the frozen scoreboard row named. No published reading, receipt or scoreboard row was changed. |
| September 28, 2026 | method change | Nowflation Realtime and the Nowflation CPI Gauge: every page that shows either reading | Nowflation Realtime used method v2.0 and the CPI Gauge method v1.4. Both weighted each component by its official weight multiplied by its cumulative price change since January 2018, which under-weighted motor fuel by about 13% and over-weighted tobacco by about 24%. A monthly market source (Kelley Blue Book, Manheim) counted as live for up to 62 days, so Kelley Blue Book's July 2026 figure was still priced into both readings after the BLS had published August, and the CPI Gauge compared a monthly source with the same calendar day a year earlier, which spans eleven months when the file is late. On September 27, 2026: Realtime 3.03%, CPI Gauge 3.37%. | From September 28, 2026 Realtime uses method v2.2 and the CPI Gauge method v1.5. Both weight components as the BLS does, with relative importances applied to price changes since December; a monthly source steps aside once the BLS has published a newer month; monthly sources compare the same month a year apart; Realtime also takes Manheim's mid-month used-vehicle reading when Cox posts it. On the switch date this raised Realtime by about 0.13pp and the CPI Gauge by about 0.18pp. Readings published before September 28 are unchanged; the CPI Gauge index is linked to its v1.4 level on September 27, so it does not jump. Neither method is fitted to CPI or to any other index. Over the 44 prints from December 2022 to August 2026, recomputed on today's data, the CPI Gauge's mean absolute miss on each release eve is 0.224pp under v1.5 against 0.227pp under v1.4, and its average bias moves from -0.07pp to +0.02pp. |
| September 26, 2026 | label or copy | the /cpi/<month> and /cpi-preview/<month> pages and their titles, the homepage track record, /scoreboard, /nowflation-vs-truflation, /compare, /gas, /heatcheck and the homepage summary text | Several graded pages called a number Nowflation's "sealed forecast" when no forecast had been sealed before that release: the March 2026 preview said "sealed forecast 3.50%" (a backtest value; receipts began in June 2026), the July 2026 page called a receipt written three weeks after the print "sealed", and the June 2026 page and the homepage showed the 3.90% scoreboard row where the June receipt says 3.935%. The scoreboard said every receipt was SHA-256 hashed and never rewritten; the June receipt has no hash and a note in it (not the forecast) was corrected an hour after it was written. /nowflation-vs-truflation said every print since 2019 was "graded the same way" and printed a blank series name in its ranking; /compare left the closest forecaster blank on several prints; /gas called $4.49 a gallon a "record high" from a history that starts in July 2026. | A graded page names a Nowflation forecast only when its receipt was locked before the release, at the receipt's value (June 3.935%, August 3.34%; March and July show that no call was sealed). The hashing statement says it starts with the August 2026 receipt and discloses the June correction. The Truflation comparison says the CPI Gauge history before September 2026 is recomputed and ranks only the series it shows; /compare names the closest shown forecaster; /gas says "the highest since this series began". The scoreboard's graded values and every receipt are unchanged. |
| September 15, 2026 | method change | the headline number: the homepage, its title and meta, the header pill, the daily email, share cards and the RSS headline | The headline number was the Nowflation CPI Gauge (formerly the Nowflation Gauge, method v1.4): 3.25% year over year on September 15, 2026 (3.247%). | The headline number is Nowflation Realtime, method v2.0, with both shelter lines on an equal blend of Zillow ZORI (all homes) and Apartment List: 2.84% year over year on September 15, 2026 (2.8367%), its first live reading, stored as published. Side by side on September 15, 2026: Nowflation Realtime 2.84%, Nowflation CPI Gauge 3.25%, official CPI 3.4% for August 2026. The CPI Gauge continues unchanged, method frozen at v1.4, on /gauge and next to the forecast and scoreboard. Realtime values before September 15, 2026 are a reconstruction under method v2.0. |
| September 15, 2026 | known issue | the Nowflation CPI Gauge: the new vehicles component and the headline (/gauge) | The CPI Gauge (method v1.4, frozen) computes the new-vehicle year-over-year change with a calendar-day year-ago lookup. When the Kelley Blue Book monthly file lands later than its usual lag, that lookup compares the latest month with the month eleven months earlier, not twelve. On September 15, 2026 it compared July 2026 with August 2025: 1.585% instead of the twelve-month 2.076%. | Not corrected: v1.4 is frozen, so the CPI Gauge keeps this lookup and its archive does not change. On September 15, 2026 the effect on the CPI Gauge headline was about -0.02pp (-0.019pp by an exact component bridge). Nowflation Realtime uses a month-keyed year-ago lookup and is not affected. |
| September 15, 2026 | label or copy | every page that names the Gauge; the homepage chart and cost-of-living section; /vs-bls, /lead-test, /methodology, /compare, /nowflation-vs-truflation, /guide, /gap, /housing | The v1.4 series was shown as the "Nowflation Gauge". The homepage and several pages also showed a "CPI-Tracker" variant and a "Cost-of-Living" variant next to it. | The same series, method frozen at v1.4, is renamed the "Nowflation CPI Gauge". No figure changed. The CPI-Tracker is no longer shown publicly (it remains a forecast input). The Cost-of-Living variant is shown only on /housing, labeled as the cost of buying a home at today's price and mortgage rate, an affordability measure and not an inflation rate. |
| September 8, 2026 | method change | the Gauge (every page); the Daily Brief archive, July 17 through September 8, 2026 | Gauge v1.3 (and v1.0 to v1.2 before it): hand-seeded weights with a 17.7% "everything else" residual riding core CPI, calendar-day year-ago bases for carried subindexes, shelter memory p 0.06 / K 36, an unchained January weight switch and several input lags set from the wrong reference date. Reading on September 8: 2.944%. Every Daily Brief from July 17 through September 8 was written under the method in force that morning and quotes that reading in its narrative (1.99% on July 17, for example), while the archived page title and Gauge figures were later recomputed under v1.4 (3.18% for the same date), so those pages showed two readings with no explanation. | v1.4, published September 8 at 23:40 ET: BLS relative-importance weights by year, the residual split into seven official subindexes (14 to 21 components), month-keyed year-ago bases, shelter memory p 0.045 / K 48, chain-linked weight switches, the October 2025 imputation rule and corrected lags. Reading on September 8: 2.944% to 3.152%. Print-eve MAE since 2019 (in-sample, 91 prints): 0.505pp to 0.298pp. Archived readings were recomputed under v1.4; the narratives were not touched. Every archived brief whose narrative was written against an earlier reading now carries a method note under its title giving the reading the narrative was written against and the v1.4 reading for that date, both from the payload. The title shows the v1.4 figure and the narrative stays as published. |
| September 7, 2026 | check or process | /status | Two source rows showed amber ("newly stale") and the header read 16/18 checks passing. | Every flagged series was current with its publisher; the checks were wrong (a FRED frequency tag, not the observed cadence; a transient USDA listing failure counted as unresolved). Checks made cadence-aware and unresolved-only; 18/18. The status header now separates critical from advisory checks. |
| September 4, 2026 | published number | dashboard hero | The hero’s live-coverage figure read "~39% of the basket priced from live sources" for about ten minutes after the 15:12 ET publish. | The figure reads the replay frames, which an out-of-schedule engine run had not rebuilt; frames rebuilt and republished at 15:23 ET showing ~43% (new vehicles had gone live that morning). The 8:40/18:00 jobs always rebuild the frames. |
| September 2, 2026 | method change | the Gauge (every page) | v1.2 (published the same morning) carried a calibrated map from the all-tenant rent index to the official rent scale and let carried-forward components drift at their trailing pace between prints — a forecast inside a measurement. Gauge 3.02%. | v1.3, published ~12:15 ET: the all-tenant index is used as measured, the calibrated map is a diagnostic only ("implied official rent in ~6 months"), and carried components hold their last print. Gauge 2.88%. Print-eve MAE since 2019: 0.58pp → 0.52pp. |
| September 2, 2026 | method change | the Gauge (every page) | v1.1 applied raw new-lease rent YoY to the whole shelter weight — it measured the marginal tenant and never converged to the print. Gauge 2.70%. | v1.2: shelter became the all-tenant rent index (new-lease rents tenure-weighted over every lease in force). Gauge 3.02% that morning; superseded by v1.3 the same day (above). |
| September 1, 2026 | published number | the Gauge, used-vehicles component | The Manheim connector had been reading a page frozen at "Mid-December 2025" and stamped its 206.0 onto every month, so the July and August 2026 readings carried a stale used-vehicle value. | The connector reads Cox’s monthly posts (month parsed from the heading) and the full history was backfilled from Cox’s xlsx. Those two months’ readings stay in the print-eve record as published; the input is corrected from 2026-09-01. |
| September 1, 2026 | published number | the Gauge (every page) | The daily history stepped monthly inputs in on their reference date, so the year-ago base ran one to two months ahead of the current level: the "YoY" was a 10-to-12-month window that shrank on the 1st of every month (2.31% → 2.16% overnight on 2026-09-01 with no new data). | v1.1: publication-date stepping (BLS subindexes on the release calendar, other monthly sources on their observed lag). Carried components now match their official subindex YoY to the basis point. Headline 2.16% → ~2.70%; gap to official −1.24pp → about −0.7pp. |
| September 1, 2026 | published number | /scoreboard, accountability board | Cleveland Fed, Kalshi and ensemble columns read the live benchmark tables, so a source’s post-print revision re-graded a print (June 2026 Cleveland was re-graded after their revision). | The board reads only the frozen scoreboard rows; live-graded and vintage-backtest records are reported separately and the rank uses the live MAE. |
| September 1, 2026 | published number | /scoreboard, backtest panel; /forecast | Nowcast v1.1 counted fuel seasonality twice (NSA fuel inside an SA aggregate, then the all-items NSA−SA delta), producing a spring-high / autumn-low error wave; the 12-month forecast passed two months of crude momentum forward at 85% on top of a pump forward that had already priced it (December 2026 central 4.57%). | Nowcast v1.2 (fuel on an SA basis): vintage-true MAE 0.164pp → 0.130pp over 88 prints. Forecast v2.1: December 2026 central 4.23%. |
| September 1, 2026 | label or copy | every page (canonical, Open Graph, sitemap) | Every canonical URL, share card and sitemap entry published since the July migration pointed at marketsbarometer.com (REALFLATION_BASE_URL in the engine .env). | nowflation.com everywhere; the first publish since July with the right domain. |
| July 7, 2026 | label or copy | homepage meta, /gap FAQ | The source count was hard-coded: 25 in the meta description, 23 in the gap FAQ; the live count was 24. | Every site-wide count comes from one published statistics file stamped into every page, never typed into copy. |
| July 5, 2026 | method change | /forecast, dashboard outlook | Forecast v1 extrapolated a six-month average of gauge momentum contaminated by the March 2026 fuel spike (and counted a partial month), over-forecasting mid-2027. | v2 component model: each of the 14 components forecast separately; the band is realized volatility × √horizon. |
| July 2, 2026 | published number | official CPI YoY on the dashboard | A gap in the series (no October 2025 CPI) shifted a positional year-ago lookup: official CPI YoY showed 4.47%. | All YoY/MoM math month-keyed; the figure was corrected to 4.25% and officials are quoted at one decimal on the NSA series, reconciled daily against the released actuals. |
A published figure that was wrong (a number), a construction change that moved a figure the site had already shown (a method change — versioned, announced, with the old readings kept as published), a label that misdescribed a number (copy), or a check that misreported the state of the data (process). New data replacing old data — a fresh print, a daily reprice, a source's own revision — is not a correction and is not listed; that is the ordinary life of the series, and the revision tracker covers it. Historical readings are stored vintage-true and never rewritten; the scoreboard grades frozen rows; every change to index or forecast math is versioned and dated in the engine changelog (this snapshot published October 17, 2026).
Reporting an error: the contact page says how — send the page URL and the series involved. Genuine errors get fixed and added here; silent corrections are against house rules. Related: the methodology page for the construction, the scoreboard for the graded record, the status page for the live data checks.
Zillow data: Data Provided by Zillow Group.