Atlas charts get clearer and easier to scanImproved
Every figure, recomputed moves from Home to Methods. Landscape gains summary charts and topic sparklines. Lab vectors clear exact dots. Playback widens rank cards and preloads three real Paths. Archive keeps First seen timestamps on one readable line.
Discussion readings now cover every commentImproved
Every readable stored comment now enters a lossless, chunked reading—even in the largest threads. New comments trigger a complete resummary. Migration runs newest-first and stays there as new threads settle, improving current discussions before history.
The section index marks where you are againFix
Opening the section list showed every section looking the same, with nothing marking the one you were already on — everywhere except Landscape, which happened to load the file that styling had ended up in. The same slip left the Start page's definition terms unstyled and dropped a column width from three of the tables. All of it is back. The cause was a build step that decides which styles each page needs: given one rule covering several pages, it was handing the whole rule to just one of them.
Charts you can reach by keyboard, and a clearer What's newFix
The small trend and comparison graphics beside each figure could not be reached by keyboard at all — on the Methods page that is ninety-eight of them. Tab now enters the group once and the arrow keys move between them, so each one can be reached without adding ninety-eight stops to the page. This page reads more easily too: dates are written out, each entry carries a coloured tag saying what kind of change it was, and the type is smaller. Separately, a build that cannot read one of its stored days now stops rather than publishing a story trajectory with that day missing.
Every page names tr3.fyi, and now every page links itFix
The footer link to the parent site appeared on some pages and not others, so a reader who arrived on a year archive or an error page found the site named but no route to it. Every published page carries it now, as its own entry at the end of the footer's reference links.
A phone layout that stays put, and links that lead somewhereFix
On a phone the homepage scrolled sideways: a table wider than the screen pushed the whole page instead of scrolling itself. It scrolls on its own now. The prompt asking you to apply a waiting update is gone — updates arrive at your next visit — and the notice that appears when a page needs reloading fits the header instead of growing a second row. The footer had begun printing the copyright and policy links more than once per page; each states them once again. On the year archives, five footer links named pages that no longer exist.
One copyright holder across every page and licenceImproved
Footers and licence files across the fleet named three different holders, so the chain of title read differently depending on which page you landed on. Every notice now reads © 2026 TR3.FYI.
tr3.fyi said twenty years of Hacker News; the record is nineteenFix
Hacker News left 50 stories in 2006 against 22,869 in 2007, so the archive counts whole years from 2007 — as the Atlas has always said. The parent page counted from the earliest story and got twenty. It now reads 2007–2026.
945 stranded discussion readings, and how long the log really runsFix
The Atlas changed how it reads a discussion on 12 August, and which model does it a day later. The 945 read in between were marked finished under the old model, where the job that upgrades old readings could not see them. It can now. The parent page said the front page had been logged for twenty years: stories and comments go back to 2006, but the ranking has only been logged since 19 June 2026, every few minutes, not every fifteen. The page now dates the log. Discussion pages also asked for their stylesheet by name, not version, so a returning visitor could meet a year-old layout.
A page for every discussion the Atlas has readNew
The Atlas has read 55,396 Hacker News discussions but could only show the few thousand from the last eight weeks. Each finished reading now has its own page, covering what the discussion was about, its themes, and how many comments were actually read, with the model named. Those pages were findable by search engines and by nobody else, so there is now an index of every year the corpus covers — how many discussions were read in each, and each one linked with its score and comment count.
The homepage now leads with which figures no longer reproduceNew
Every figure the Atlas prints is recomputed from the corpus on each publish. That receipt was two clicks away; it now sits under the thesis, naming the figures that no longer reproduce rather than quietly correcting them. All 96, with the method, stay on the methods page.
Discussion readings beside the story, and whole threads capturedImproved
Where the Atlas has read a story's discussion, that reading now appears under its rank, score and comment figures — in snapshot playback, in the figure and in the observatory — naming the model and how many comments it read. Readings are now taken a day after a story appears, once it holds essentially every comment it will get, and twenty-seven written too early will be read again. The crawler no longer stops at 500 comments a thread, so the largest discussions are held whole, and 55,582 comments posted under polls are now linked to them. Pages also load about a tenth less.
Story names become links, and a panel that waits instead of guessingFix
The paragraph above the figure names the top-ranked story and the most-discussed one, and quoted them as plain text, the one place telling you which discussion mattered with no way to open it. Both are links now, as is the label on a selected point in the chart. The interpretation panel used to wait before drawing anything and reported that no interpretation existed when the request was merely slow. It now draws first and fills in when the interpretation arrives, saying so meanwhile, and Methods reads the interpretation at the moment it describes it.
The discussion readings now follow whole threadsImproved
Each discussion's reading was made from the twenty-five most-replied-to comments, with nothing to show which answered which. It now reads whole threads, a comment and the replies beneath it, and usually covers the whole discussion where before it saw a tenth. The model reading them has changed: on 24 real discussions it names a third more specific things per sentence and hedges a fifth as often, saying what people argued where the old one said what was discussed. Older discussions are being re-read most-discussed first, taking several days, so a page may still show its earlier one.
What a reading can report, and when it was takenFix
The panel that reads the front page called its discussions live while the reading itself was fixed at the last release, so the date it was read is now stated beside the title. Time on the list and times captured both need a story seen more than once, and were reading Unavailable on every figure built from a single moment. They are left out until a story is seen twice. A correlation of 0.46 was said to account for 22% of the variation, where the figure shown squares to 21, and the share is now worked out from the number a reader can see. On phones the header is one row again.
Outside attention, thread shape, and one way around the siteImproved
Methods now shows outside attention on the subjects it covers: Wikipedia pageviews over 30 days and the share of monitored news mentioning them. Archive reads today's largest discussions against a captured baseline of 260 threads, each live discussion quotes the opening comment that drew the most replies, and every subject with a share of titles now carries a share of comments beside it. Agents take 6.49% of this year's titles and 2.60% of its comments. The header shows where you are and one control that opens every section, and the verification table lists all 96 published figures.
A verification receipt for every figure, and two new measurementsNew
Methods now names every published figure whose recomputation no longer matches the page. Figures are recalculated from the retained corpus on every publish, 82 of 96 reproduce at the precision they are printed, and the 14 that do not are listed with their basis, each set against a twenty-year share line or its peers. Archive carries two measurements taken across the whole record, how long a story stays on the front page and how often a resubmitted link later lands. Type, spacing and line spacing now follow one scale, and a section link that cannot open loads that section's own page.
A value that could not be measured no longer looks like one that wasFix
A story seen only once has no change to report and no capture history, so four of its six boxes read “Unavailable”. That word was set in the same large type as the measured numbers, was wider than its box, and broke mid-word. It is now smaller and italic, on one line.
Snapshot playback restored, and small screens stop breakingFix
Snapshot playback had shown an empty workspace since 5 August, when a cleanup removed a function it needed. It works again, with all 49 captured days and every exact tick replaying as before. Every page title and description now names Hacker News. Landscape no longer scrolls sideways on phones, headings beside a button or chip wrap to their own line, and a repeated Unavailable stops running outside its panel. Labels on the survival chart step apart when they would touch, and the published reliability figure now counts only resources the site itself asks for.
What changed since yesterday, and a way onward from every sectionImproved
The paper now opens with a line comparing today's front page with the capture a day earlier, the median score and the total discussion, both shown, and it disappears when there is no capture a day back. Six sections that ended with no route out now close with a link to the section that follows, in the order the section list already uses. Every section sits on the same 1120px sheet, the header lost a fifth of its height and lines up with the page's own gutters, and going back returns each section to the position you left it. Typing 1 through 9 opens a section and / opens the finder.
One mark across the fleetImproved
Every page, stylesheet, script, feed, and legal document now carries the same notice: TR3.FYI. The previous operator name is gone from everything these sites serve, including the machine-readable publisher data search engines read.
A one-question survey and charts that read in any color visionImproved
The reader survey is now a single question — what the Atlas should explain, fix, or investigate next — still anonymous, with no name or email field. The four story lists use colors checked by computation to stay distinct for red-green and blue-yellow color vision alike, with axis numbers on round values and a legend that matches the marks in both themes. Capture details read one line each, which returns the paper's opening finding to the first phone screen, and a waiting update takes its own row instead of crowding the header. Glossary pop-ups no longer let wide pages drift sideways.
The interface thins out around the evidenceImproved
Most of the trim is gone: the story dossier and its pin panel, the bottom route bar, section numbers and labels, the pulsing dot and two footer rows. The header keeps the name, a Sections control naming the current page, and the capture time, and every route stays in the index. Topic correlations now stop at the last complete month, because a month the crawl was two days into lifted every topic's share at once and read as false co-movement, and this release carries the corrected numbers. Padding and gaps that had drifted across 28 sizes now sit on one scale.
Repairs to the private console, and a clearer reader surveyFix
Removing the Search Console panels also deleted three helper functions the rest of the private operator console still used, so every report failed to draw. They are back, a test now checks that nothing the page calls has gone missing, and a report that cannot be shown says whether the request failed or the drawing did. The sign-in no longer carries the retired Signal Bridge name. Elsewhere, the dot beside Hacker News Atlas in the site switcher matches the Atlas's own accent instead of the old orange, and the reader survey now asks whether the measurements were clear.
A Start here page, one section index, and a new markImproved
A new Start here page explains what Hacker News is, how a story reaches its front page, and what rank, attention, discussion and time each count. The header now names Paper, Observatory and Lab, with an All sections control opening the full index, and a new mark — an A whose crossbar is a measurement bar — carries across every link preview. Topic co-movement now stops at the last complete month instead of the one still being collected, and monthly topic shares carry an interval, so a thinly sampled month reads as uncertain. Other fixes keep the capture time readable in the header.
Landscape arrives and the old interface is retiredImproved
A new Landscape workspace connects 32 annual topic series, 14 reviewed discussion themes, four activity cohorts and four discussion surfaces to complete tables, with its chart and tables held inside named scrollers on phones. About, changes, feedback, privacy and terms now belong to Atlas, and former Explore, Research, Time Machine, Evidence and Methodology addresses permanently redirect. Navigation gained one clear close path on phones and room for all 15 sections on tablets, and Atlas now refuses future, duplicate, oversized or incomplete capture data.
- from feedback
Every dashboard receives only the interface it usesImproved
Owned map, ranked-measure, drill-through, ledger, and live-interaction layers now ship only where rendered. HN sheds 3,828 CSS and 6,252 JS bytes raw; each sister sheds 1,219 CSS and 7,524 JS bytes. The U.S. index still collapses at its intended laptop width.
One name, keyboard-reachable evidence, and lighter pagesImproved
Atlas is now the sole product name, with Hacker News as its subject, and each route ships only its own workspace under its own heading. Every chart in the Visualization Lab replaces hundreds of invisible controls with one exact-value inspector that arrow keys walk through, and the Observatory, Archive and Playback gain the same treatment, the largest comparison keeping all 1,432 exact observations. Methods traces capture through interpretation across six sources, ten formulas and twenty-one fields. Retired files and styles left the release, and the footer is one compact directory.
Phone layouts, quieter fine print, and keyboard-reachable tablesFix
On touch screens a rule meant to keep buttons easy to tap was inflating the little help icon into a large grey circle beside headings. It is back to its intended size, and small chart labels no longer shrink to half their size on phones. On iPhones the top bar sits below the status bar rather than behind it, panning charts fade at the edge, and research pages stay inside narrow screens. Method notes and caveats are small and muted again, wide tables and charts that scroll sideways can now be focused and scrolled by keyboard, and an available update waits until the reader applies it.
A live-first dashboard, one set of names, and two community chartsImproved
The dashboard now opens on the live briefing, leading with current signals and front-page movement and keeping the nineteen-year highlights one route away. The desktop bar and phone menu settle on Briefing, Explore, Research and Evidence, Explore's question rail is grouped into four themes, and Research opens with five labelled routes and category filters. Two community charts join the Explorer: half of 2007's newcomers posted again the next year against about a fifth since 2017, and a question draws 3.2 times the replies a demo does. Captions now lead with what each chart found.
One shared scale for type, spacing and shapesImproved
Card and panel titles had been rendering larger than the section headings above them, and headings now step down in a visible order at every screen size. Text sizes, weights, line height and letter spacing come from one named scale, rounded corners from seven defined shapes, and the gaps between and within sections from a small set of steps. Chart labels are set per canvas so they read the same on every chart, fine print sits at one size, and twenty-one rules for elements that no longer exist stopped being downloaded.
What reposting is worth, and a second corpus that shows the same shiftImproved
A new study compares links posted more than once with links generally: those that missed first time reach 100 points on a later try about 9% of the time, against roughly 5% overall, and it explains why that gap is weaker evidence than it looks. The residency study now tests whether the first hour of discussion signals a final score. It does, but weakly. The evidence page charts the skills named in HN's Who is hiring? threads, where AI and ML rose 8.6 points while React fell 9.8. Phones get a menu drawer holding all ten destinations, and chart labels read the same on every figure.
Live briefing readings, three new studies, and a corrected corpus countImproved
An accuracy sweep found the site had understated the archive, which holds about 49 million public items, 42.8 million comments and 6.2 million stories, not the 18 million stated. The briefing now reads the front page as it moves, giving a hot thread's comments per hour, what each busy discussion turns on, and how today's volume compares with recent weeks. New pages follow 19 once-prominent title terms from peak to near-silence and measure front-page residency, about three hours for the middle story. A new grid shows submissions by hour and weekday across 5.9 million stories.
Ten new charts, terms that explain themselves, and a phone layout put rightImproved
Ten new figures arrived across the almanac, among them author concentration, front-page life spans, source checkability, every topic as a small chart and the verification ledger drawn. Hover or use the arrow keys on a line chart for exact values. Words like retained, census, observed and breakout are now dotted-underlined on first use on each page, so hovering or tapping one gives a plain definition. The phone layout was repaired: margins are back at every size, tab names stay on one line, and the section list, once reachable three ways, now lives only in the side rail.
Refreshed almanac data and lighter first viewsImproved
The five almanacs carry refreshed data, charts and source context from their latest verified research release. Pages also arrive lighter, since each site now downloads only the styling that applies to it and form styling waits until a suggestion or contact form is first opened. Styles and scripts are stamped with a fingerprint, so browsers keep them between visits while a new release still takes effect immediately. The offline screen now lists only pages that almanac can actually reach.
Steady navigation and a consistent shape across every pageImproved
Navigation is now a fixed rail beside the page, becoming a row of tabs on a phone, with all five main sections visible and the current page marked. Headings, page furniture and the link for writing in are the same everywhere, so moving around no longer changes the site's shape. The site mark follows your light or dark setting, Explorer charts no longer stay blank when a page opens in a background tab, and a footer notice on every page states that this is an independent project. Search engines can now read the terms covering reuse of the published data.
Fresher Ask, Show and Jobs figures, and counting without identifiersImproved
The Ask HN, Show HN and Jobs pulse now takes in newer scheduled observations, while keeping the real capture boundaries, the visible gaps and the still-accruing 30-day milestone exactly as they were. Anonymous aggregate measurement also went live: the site records only fixed, low-detail categories of product use, with no cookie, no visitor identifier, no raw address and no referring page. Global Privacy Control and Do Not Track are honored, and the aggregate reporting stays private to the maintainer.
Current research releases were synchronizedImproved
The five almanacs now carry their latest verified data, research, accessibility, and source-context updates together; superseded catalog pages were retired and every public release passed the shared quality checks.
Data freshness became easier to verifyImproved
A unified status report now follows collection, processing, context, publication, and site readiness across every dashboard.
Live attention history became more resilientData
Frequent ranking observations now flow into the main research archive automatically, while the broader item history continues catching up on a recurring schedule.
A guided briefing that leads with verified findingsImproved
The homepage briefing now sequences the strongest verified findings, links each one to the evidence behind it, and keeps scope and uncertainty in view instead of repeating the lead insight. Research modules and the pages around them gained clearer structure, more reliable navigation and better assistive context across the standalone site. Compact navigation now stays within the width of small screens.
A lighter Attention Lab and new headline, careers and community evidenceImproved
The Attention Lab now opens on annual topic and community indexes and loads the larger captured-story archive only when Story Lifecycle is approached, while the dashboard leads with a short briefing that routes into the long-form research atlas. The lab also gained privacy-safe community cohorts, source-domain history, and separate official-jobs and community hiring trends with explicit geography and pay bands. Its publisher-headline comparison separates direct title changes from summary wording and shows its progress toward the threshold it must clear to count as a primary metric.
Faster dashboards and clearer research pagesImproved
Deeper analysis now loads when it is requested, dense maps render more efficiently, and shared forms and services carry stronger safeguards. Research pages draw on more focused evidence and state their interpretation and provenance more clearly. Each one now has its own visual preview, and the update signals they send to search engines are more accurate.
Topic cards and year selection on the chartsImproved
Selecting a topic now reveals more detail in place, while the surrounding view reorganizes smoothly around it. Year selection on the charts is easier to recognize and use across devices.
Research pages, deeper analysis tools and a more compact layoutImproved
New topic, year, product, source, state, agency, hazard, and airspace research pages make verified dashboard evidence easier to browse, cite, and discover, and page metadata, update feeds, and search indexes stay synchronized automatically. Readers can inspect evidence, compare views, ask guided questions, and save private notes. Briefing, exploration, search, saved views, and context take less vertical space, and dashboards adapt more naturally from phones and tablets to large desktops. Dashboard names, shared-link previews, and selection controls are clearer and more consistent.
Historical front pages became explorableNew
Past front pages can now be revisited and compared with current activity.
The Hacker News Atlas opensNew
The Hacker News Atlas became available, giving a long-range view of changing technology attention and notable stories. Rising and fading subjects sit on the homepage, so topic momentum is visible before opening the deeper analysis. Readers can also propose ideas, report corrections, or ask questions directly from the dashboard.
What’s new in Atlas.
Every meaningful reader-facing improvement is listed newest first. Older entries remain visible and searchable on the page.