25 dated entries since 2026-08-21: data corrections and site changes, each with its number and its change reference.
2012–2026 data · 30,000+ schools · 10,000+ cities · Free to use. An optional $4.99 report on some schools
Every change to the data or the site that a reader could notice, dated, with the number behind it and a reference number to quote if you write in about it. A correction is listed as a correction. Newest first.
City, name and school pages compare their numbers
· 7,861 pages · Ref. #402
City pages now say where a city's best school sits in its state's list and compare the city's median graduation rate and school size with the state's. Pages for schools that share a name compare the group with the U.S. median. A school page compares its graduation rate and enrollment with its state's median for the same edition. Each comparison uses only sourced figures from one edition, and is left out when the full list or a figure is missing.
Homepage: a placeholder and two untrue claims removed
· 10,795 cities · Ref. #390
The homepage showed a "More Featured Cities Coming" card beside one hardcoded city with an unexplained score; that block is gone. The hero gave a city count of 491 that nobody had measured; it now says 10,000+, since 10,795 cities had a ranked school on 2026-10-21. It said "100% free" while some school pages sell an optional $4.99 report, and now says exactly that.
One-school city pages leave the index
· 8,808 city pages · Ref. #385
A city page with one ranked school mostly repeats that school's own page. On 2026-10-21, 10,795 cities had a ranked school and 8,808 had exactly one. Those pages now carry noindex, follow: the page stays, its links still count, and Google stops indexing it. The edge HTML and the React page read the same rule from one constant, the one the city sitemap has used since 2026-10-17.
Structured data written raw, not entity-escaped
· Ref. #384
JSON-LD on the edge went through the HTML-escaping path, so an ampersand in a school name reached the structured data as "&" ("Tech & Sciences"). The JSON-LD is now written raw; the visible HTML keeps its escaping.
A value whose own record names no source is no longer stated as a fact in any title, description, answer, tile or structured data. On a trend chart it stays as a hollow point labelled "no source on file", never joined by a solid line, and it is left out of the summary tiles and the years-of-data count. The 2026-10-19 audit of 150 live school pages had found 3,142 such assertions on 111 pages. An official non-U.S. News original (NCES, EDFacts, Census, a state agency) counts as sourced, and a test pins every source string the site stores or writes.
A receipt never attributes an unsourced value to U.S. News
· 253 receipts on 66 pages · Ref. #372
On 66 of the 150 audited school pages, 253 receipts described a value with no provenance entry as a U.S. News figure "as our crawler recorded it". The fallback that claimed U.S. News for anything unrecognised is gone: a receipt names U.S. News only when the value's own record names a U.S. News source; otherwise it says "No source on file".
Six untrue claims removed
· six claims · Ref. #373
The same audit found six claims the site could not back, each now fixed at the source with a test. The site-wide school count was about 5,000 too high and now reads "30,000+" (30,244 live school pages that day). A footer "Updated" date that no data refresh moved was removed. "Unranked · Modeled" became "Unranked". The 7 hand-written editor notes are hidden because each is stale against the 2026-27 edition. A "top N%" figure now rounds up (0.44% reads top 0.5%, not 0.4%). The score card no longer says scores use only figures traceable to a published source; two of its inputs have no provenance entry.
State pages open with the state's story
· 51 states · Refs. #365, #367, #369, #370, #371
Each state page now opens with the state's standing among states by schools in the national top 500 and up to four rank-history lines: the state's top-ranked school and its published movers. A line uses only exact ranks from 2019 on, needs at least 3 points, and stops at a state method break. The previous page layout is kept as the rollback.
State tables are built before the first visitor
· 5 to 7 s cold, now pre-built · Ref. #368
Cold, a state's moves table and story took 5 to 7 seconds to build (Iowa: 6.95 s), longer than the edge waits, so the first visitor after each deploy and each index reload got a page without moves, standing or rank history. A warmer now builds all 51 states one at a time at startup and every 2 hours.
The hero ranks and every academic tile carry a Source button that shows the stored row the number came from, matched by edition and exact value, and a "Sources for these numbers" drawer covers the comparison medians. A link is offered only when the page it points to prints the number; a withheld value sends only its plain-language reason. Before this, /methodology stopped claiming per-number attribution the site did not yet have, and a premium preview that showed made-up alerts now lists only what it monitors.
Biggest moves tables published
· 13 of 13 archived values matched · Ref. #355
Each state's "Biggest moves" table in the top 100 went live after a person re-checked the method against Internet Archive captures of U.S. News pages: 13 of 13 readable values equal to ours, 0 mismatches, 12 sample rows without a capture. 48 states get a table; New York, Colorado and Vermont show a method-break note instead.
History description on every qualifying school page
· baseline 35.0 clicks/day · Ref. #353
The 50/50 test of a rank-history search description ended before its readout; every school page whose history qualifies now carries it. Titles are unchanged. The effect is read as clicks per day against the 35.0/day baseline of 2026-10-08 to 10-14, directional only, since there is no longer a control group.
City sitemap keeps cities with 2 or more ranked schools
· 10,767 to 1,990 cities · Ref. #361
#358 had grown the city sitemap from 704 to 10,767 cities; 8,808 of the 10,795 ranked cities had a single ranked school, and a one-school city page mostly repeats that school's page. The sitemap now lists the 1,990 cities with at least 2 ranked schools. The other city pages stay crawlable from the state hubs.
Every URL serves its own page to crawlers that do not run JavaScript
· 7 URLs and 5 posts · Ref. #358
A crawl audit found 7 sitemap URLs (/, /atlas, /explore, /blog, /contact, /privacy, /terms) serving the homepage's title and body in their raw HTML with no canonical, the 5 blog posts serving the homepage's main content, invented city URLs returning 200, and 49 of the 183 cities the state hubs link to missing from the sitemap. Each page now has its own title, canonical and body; an unknown city is a 404; the homepage HTML links all 51 state hubs.
A national top-100 page
· 247 impressions at position 11.8 · Ref. #359
"high school rankings" (247 impressions at an average position of 11.8) and "best high schools" landed on the homepage, a dashboard. /rankings now lists the current edition's national top 100 as real links with national and state ranks, plus all 51 state hubs, in the HTML a crawler reads.
Shorter pages, comparison visuals, the Atlas
· state page 28.9 to 6.5 phone screens · Refs. #344, #345, #346, #347, #348, #349
School pages lost about 1.7 phone screens, state pages went from 28.9 to 6.5 screens with no layout shift, and city pages about 1.1 screens, without hiding a number. Added: a tile map of every state on state pages, the /atlas state board with per-state card views, and a school's top section as a card with a top-500 ribbon and two comparison bars.
Homes near a school hand off to Zillow's own search
· 2 clicks in 90 days · Ref. #341
The hero's "View homes nearby on Zillow" button linked the whole city and was clicked twice in 90 days. It is replaced by a block, after the school's Q&A, that opens Zillow centred on the school with distance (1, 2 or 5 miles), price and bedroom filters, and says that the box is not the attendance zone and that Zillow pays us nothing. Zillow is no longer listed as a data source, and the privacy page no longer calls these links affiliate links.
U.S. News scored College Readiness from IB exams only in New York, Colorado and Vermont for 2026-27, and in Maine since 2021-22, so ranks there are a different construct from earlier editions. No "up/down from", national percentile or cross-state comparison now crosses such a break, and the note on affected pages names the break. Measured on a mirror of the 2026-10-10 database: 1,699 school pages change, 0 of 1,500 control pages.
Lists, search, maps and compare never show a withheld value
· 19 of about 19,500 ranks trustworthy · Refs. #304, #311
School pages already left withheld values out; search, state lists, compare, map pins and the SEO index still read the stored values. Every list surface now strips withheld values before a row is chosen, filtered or sorted. The homepage rank showcase was withdrawn: only 19 of about 19,500 stored 2024-25 ranks were trustworthy, so no honest trend remained.
Model-filled values and College Readiness ranks are withheld
· about 431 schools kept real values · Ref. #293
Upstream marked, field by field, values that were not U.S. News measurements: model-imputed figures, values overwritten after extraction, and College Readiness Index ranks that had been stored as national ranks. They stay in the database and are no longer served as a school's number; a school page shows one caption when something was withheld. Deduplication was corrected at the same time so that two editions differing only in a withheld value are kept apart, which preserved real math, AP and state-rank values for about 431 schools.
Rank-movement claims and the insight pages withdrawn
· 96% of 2024 and 94% of 2025 ranked rows · Refs. #281, #283
The 2020 to 2025 extractor had stored U.S. News's College Readiness Index rank as the national rank for 96% of 2024 and 94% of 2025 ranked rows. Every comparison of a 2026 rank with those years was therefore wrong: the "down from #X" clause in school titles, descriptions and bodies, and the movers and AP-gap insight pages. A prior rank now counts only from the trusted extractor, the insight pages return 404, and every link to them is gone.
Search Console for the 3 months to 2026-10-05: 27.8K impressions, 79 clicks, a 0.3% click rate, with every SEO page serving crawlers the homepage's generic text under a swapped heading. One title builder now serves both the edge HTML and React, so the two cannot disagree; state, city and school pages carry crawlable ranked lists; 516 school-name hubs were added; /compare, /methodology, /about and /faq got their own edge titles; Title-Case city URLs 301 to the lowercase canonical.
Rating markup removed
· 3 Search Console errors · Ref. #250
Search Console flagged three Review-snippet errors: school pages published the U.S. News rank as an AggregateRating (best 1, worst 17,769, no count) and city pages a composite score as 5 stars. The site collects no reviews, so there is no honest form of either; both blocks were removed rather than repaired. The same change fixed sitemap lastmod dates (0 of 29,356 were valid) and gave city pages one lowercase canonical instead of two conflicting ones. Search Console kept listing the three against one old URL that now redirects; validation of the fix started 2026-10-21.
Current-edition metrics, not a carry-forward clone
· math proficiency 68 served, 62 real · Ref. #226
Record selection preferred the richest row, which for many schools was a 2025 row cloned from 2024, so a page could show a 2026 rank beside two-year-old metrics, and two different graduation rates at once. One production example: math proficiency served as 68, the 2026 value 62. The current edition's values now overlay the served row under four rules: never downgrade trust, never blank a field, one source edition, demographics move as a set or not at all.
The trend chart dropped every 2026 point behind a hardcoded 2025 cap; the rank on a school page was an older edition's; state lists sorted across mixed editions. Five pull requests fixed each, restated the data-coverage copy from one definition (about 180 copy sites, with a drift test) and rendered the per-field provenance that the FAQ had promised but nothing showed.