Changelog

A chain reads as a page on a phone. The pull-up sheet is gone from the chains view below laptop width. A phone now reads top to bottom: what the chain is, "Full evidence" and "On the globe" once, the route in one strip you can tap along, its makers in a line, one row of tabs pinned under the top bar, and the reading at full width — no grip, no second scroll, no 3 × 2 tab grid taking a quarter of the screen. A tablet shows the diagram with the reading under it; a laptop keeps the panel beside the diagram. The figures are one row of numbers instead of boxes, and money's answer is one word with the reasoning above it. (#178) Two more ways to shade the globe. The legend's picker now offers Historical trend (each country's production change between the two years USGS prints) and Risk heat-map (how many export restrictions the OECD inventory records on the resource, per country), beside Production volumes and Reserves. Trade flows and environmental impact zones are listed as not yet available: no sourced figures for them are on file. (#177) Every chain, read six ways. The panel beside a chain now has six tabs: Overview, Markets, Research, Money, People and Policy. Each says what the chain means from that side, in sentences built from the records, then the few figures that carry it, the records behind it, and the page that holds all of it. Markets names the listed makers by ticker and what a stop would reach; Research, how far along each way around it is; Money, whether money is the constraint at all; People, the work it needs and where to train for it; Policy, the rules that touch it and the restrictions on its raw material. A tab is a link, so a reading can be shared, and it stays chosen as you walk the chain. (#176) A fuller description of each chain. The overview now leads with why the thing binds, the grade that actually ships, the four tests behind its severity score in words, and what has happened to it. (#176) Producer sites on the globe. Pick a resource on the world view and every mine, smelter and refinery the USGS Minerals Yearbook lists for it appears where it is: about a thousand locations in 37 countries, placed from GeoNames. A solid dot is the facility or its town; an open ring means only the province is known. Click one and the globe flies in and the panel lists who operates what there, at what capacity, with both sources. (#175) A chain, laid over the Earth. "On the globe" on any chain draws it across the world: one line from where each input is made to where the step that needs it is made, with a light running the way material moves, and numbered labels that match the list beside it. EUV scanners: optics from Germany and tin from China, Peru and Indonesia into the Netherlands, then on to foundries and packaging in Taiwan. (#175) What you can do, by reader. Every chain now ends with one row each for equity investors, commodity traders, industry teams, job seekers and learners, each built from the records and linking to the page that serves them. (#174; folded into the six readings in #176, each naming the reader it serves most) The globe says where. Picking a resource labels its largest producers in place ("Australia · 32%") and turns to face them; a clicked country is reached by a flight that lands with a ring. (#174) Clearer headings, no tiny text on the map. The map is titled "Global Resource Map – Location, Production, and Supply-Chain Connections"; on the chain diagram the column headings are larger and bold and the "Made by" headings lighter; no label on the map is below 12px; the sidebar's Explore heading says what is inside. (#177) The chain is a route, not a hub. The diagram now shows every step two levels up and down — for leading-edge foundries, quartz and polysilicon into wafers, scanners, helium, neon and photoresist, into the foundries, into packaging — left to right in the order material moves. Makers sit inside the bottleneck they build. (#175) The chain reads in plain words. A maker opens beside the diagram instead of leaving the map, and the panel says in sentences what the chain tells us. (#173) One slip still finds it. A mistyped name now finds what was meant in the globe's country finder ("Liechtenstien"), the ⌘K page list ("expsoure") and the filter boxes on Exposure, Data series and your desk ("wolfsped"), where one wrong letter used to return nothing; the matched part of each name is marked. Site search already forgave typos and now also forgives one in the word still being typed ("photres") and in a word the site has only as the start of longer ones ("canno" finds Canon). Filter boxes now match the start of each typed word, in any order, rather than one run of letters anywhere. (#179) No line runs through the makers. The old diagram drew the bottleneck's connector straight through the cards of its part suppliers. (#175)

Answer Ask in one tap. When an answer asks you something back, or the next step is obvious, two to four replies appear under it; tapping one sends it as your next question. (#167) The front page says what it is, then shows it. One line, one sentence, one button; the three worst constraints right now; five doors as rows; the three newest checked events; the counts; a line for anyone who can fix a row; how live the sweep is. On a phone it was twenty screens (18,686px) — thirteen problem cards, two chip clouds, two event lists, the latest rule — every piece true and nothing said first. What left it is still on the site: the chips are the filters on Bottlenecks, the latest rule tops Policy, unchecked finds lead News, and the problem cards' copy stays for the role views. One row of buttons under an answer. The suggested replies and the questions built from the records Ask read now sit in a single row, replies first, nothing offered twice, at most five — instead of two rows under the same answer. A short reply is a question. "No" or "OK" used to vanish from the Ask box without being sent; two characters now send, and anything not sent stays in the box. (#167)

News leads with what was just found. /events is now News: the newest sweep finds of the last 14 days first, each marked not yet checked, then the checked events grouped by month with their source named. (#153) Anyone can pull fresh news. "Update news now" on News searches the web for the bottlenecks searched longest ago — signed out, no AI — and says plainly when the last search is more than a day old. (#153) "On this page" for long pages. A list of the page's own sections beside the content, or a strip under the top bar on a phone, marking where you are. (#153) No more tenders or kitchen worktops in the news. A government scanner contract and "quartz surface products" no longer pass as news about high-purity quartz. (#153, #154) Ask no longer rewrites a correct answer. A figure written into a field name (net_pressure_90d) counts as evidence, saving a model call. The chat components now come from npm. (#150)

Ask has a tool behind every section — technologies and their readiness, research papers, policy rules and who asked for them, SEC filings, price series, reserves and Substrata's own notes — so it no longer says the site does not hold what the page it is on shows. (#145, #146) Update now reaches research too. On a bottleneck page it also searches the open research databases, with no AI. (#145) Ask is faster and finishes its answers. On the same twelve questions: 29 → 20 model calls, about 30% fewer tokens, 121 s → 54 s; an answer cut off at the length limit is completed once. (#145, #146) One article is one lead. A search engine's tracking parameter no longer turns the same page into a new lead every day. (#145) Ask answers in words, not tool calls. It no longer replies with a raw tool call, reads the site's own numbers, production tables and open roles, converts units correctly, answers in the language of the question, gives links you can copy, loosens a job search instead of returning nothing, and says whether a change went up or down in words. (#136, #137, #138)

The front page is live. Checked events first, then up to four recent sweep finds that report a change and are not yet filed, each labelled as not yet checked with a way to check it. (#134) Accepting an event publishes it at once. An event accepted at /review is on the site immediately, with its quote checked against the page at the moment of acceptance. (#135)

The world map loads only when you open it, from a manifest that records each dataset's source and licence; a failed load offers a retry instead of a stuck globe. (#126, #127)

/roadmap renders again. It had returned an error since the roadmap started coming from the fleet map. (#125)

The world view is a globe. The atlas opens on a globe you can turn, painted by the same USGS data as the flat map. (#120) Task-flow criteria, measured. Every tool page now answers first, has one header and one empty state, and the flows a reader actually runs are measured after each deploy — 20 of 20 task runs completed with no dead ends. (#117, #121) Quality scores for every dataset. /data/quality holds each dataset to six written criteria — completeness, correctness, provenance, freshness, link health, consistency — with pure checks in the build and network checks every six hours. Fixed what the first run found. (#116) Every producer capacity names its own unit. A capacity figure no longer inherits a unit from its neighbour. (#122) Touch targets and an early paste. Breadcrumbs and reader tabs are 44px on a phone; the careers sidebar sits above Ask; the X-ray keeps a portfolio you pasted before the page finished loading. (#119) Shin-Etsu Quartz and MetOx are back to unverified. Two producer rows had been marked sourced without a page that names the company and the material. (#118) An accepted lead says so. The desk no longer shows "not yet reviewed" on a lead that was accepted. (#115)

Check a pasted key with its vendor. Bring your own API key: Substrata asks the vendor which models the key can use and lets you pick from them, instead of guessing. (#114) Unreviewed leads expire. A sweep lead nobody reviews within 30 days leaves the queue at read time and is listed at /data/freshness/expired, so a stale queue means the expiry is broken, not that nobody looked. (#113) The country panel has numbers. Measured resources, producers, export restrictions, sanctions and ranked peers for every country the map draws. (#107) One canvas, one bar, one panel. The atlas became a 50m vector map painted by USGS data, with supply chains drawn as a flow. (#106) USGS world production and reserves per country, and a choropleth API the map paints from. (#105) One shell for every page. A grouped sidebar, an icon rail on tablets, four tabs and "More" on phones, and a ⌘K palette — signed in or not. (#103) No orphan role label, and the figures table lays out by the panel's width. (#112) One queue per engine. The research engines each have exactly one Postgres queue; the model picker in Ask is usable on a phone. (#111) The world is framed in what the sheet leaves visible, and the feedback launcher stays off the panel. (#110) Empty blocks are omitted; a resource states its absence once, not per block. (#109) The unreadable country diagram is gone; the map counts the countries it actually draws. (#108) Ask sits in the phone top bar, and "Sign in" replaced a dot. A structural test now fails on overlapping chrome. (#104)

Reader-initiated news updates. You can ask for the news on a bottleneck to be refreshed; AI drafting happens only on your own key. (#102) Reader views. Five doors — for equity investors, policy, science, careers, learning — each collecting the existing screens, plus a freshness page that breaks the build on stale data and a shared page header. (#96) Dependency rows as graph edges. 26 sourced dependency rows, pinned home listings, and a minimum move on every series. (#94) Dated, sourced number series per bottleneck. Lead times, prices, capacity, output, backlogs and trade volumes at /data/series, as CSV, with desk alerts on a move. (#92) Open roles from public job boards, with the skills and training each bottleneck asks for. (#91) Portfolio X-ray and scenarios over a sourced dependency layer. (#88) Ask on any AI, fast, with one-click verification of a claim. (#89) The science pipeline: stages, live paper and grant feeds, and who is doing the work. (#87) AI-drafted events from sweep leads, reviewed in one click, accepted straight into git. (#85) Exposure screen, tickers from SEC and OpenFIGI, and SEC filings on the desk. (#84) A desk you control: filters, read/save/hide, rails and sweep settings. (#81) Search: one grouped index with instant results, typo tolerance and highlights. (#79) Company profiles that say what the company holds and why it matters. (#78) A context-aware assistant with tool calling over the corpus, streamed. (#77) Figure: every number links its source or explains its computation. (#76) A live news feed on your rails, with a way to refresh it. (#75) SUSS Karlsruhe Advanced Packaging center, accepted as an event at /review. (#90) Readers first. Background jobs draw from a capped, reserved slice of the free AI budget so a reader is never queued behind a cron. (#93) Ask plans the obvious lookups before the model and makes one call for a page question; it walks on from a free link that shows nothing in 9 seconds. (#98, #100) News stays fresh on a smaller footprint; producer sourcing runs every six hours. (#97, #101) No solo framing. Judgements are Substrata's and open to challenge. (#82) Every number on the public pages opens to its source, rule or estimate. (#80) A plan outranks the page-only shortcut, and half-converted citation markers are mended. (#99) Dependency rows are labelled "sourced", not "filing" — the rows include annual reports and company blogs. (#95) One drafting run at a time, and a draft that looks like an already-accepted event is flagged. (#86)

Retrieval quality in chat: word-boundary scoring, IDF weighting and a relevance floor. (#73) The assistant answers the question asked, and the header is a menu again. (#72)

The research sweep runs on a timer, into a queue rather than the corpus; the leads have a reader, and freshness stops being a guess. (#68, #69) Relief for the two worst science rows, and the prompt no longer claims it has no tools. (#67) A timestamp column is a Date. Asserting otherwise shipped green. (#70)

Five more materials with figures, and the actual places named. (#66) Numbers with a year and a source, and the difference between a law and a schedule. (#65) The assistant can look things up, without letting the web pretend to be research. (#64) The assistant knows what you are reading. A kind is one file; loops became entities. (#63) The web as a schema, and the seams for eight languages. (#61) Every entity page is modules on one shared order; the company page and one entity registry came first. (#58, #59, #60) The corpus is indexed, and every entity has a path to the KPI. (#62)

A dossier for every country, a resource directory and a derived graph — not a demo for one. (#51, #53) Equal Earth map, comments, AI fact-check. (#54) Readable atlas chains, a mobile menu, follow companies. (#55) Ask as a thread, grounded on ai-kit. (#52) Desk, world map, a visible changelog, dark/light/auto. (#48) Mark, one header, and a chain figure on the front page. (#47) An unknown model id or an anonymous fact-check no longer spends the budget. (#57) Menus close outside, Auto model, dictate, attach. (#56) The research is back in the chrome; two shells, one map, inquire on gaps. (#49, #50) Wafer producer coverage is distinguished from market claims. (#46)

A checkable research atlas and the Substrata companion. (#43) Capital, Learn, and one module that owns every internal URL. (#41) Every readiness score cited, nine events accepted, part of the directory sourced. (#40) The track record started, and the engine no longer mistakes blindness for absence. (#39) A grouped megamenu, notes, and an invitation to contribute. (#38) Sections for policy, markets and science, in plain language, with a test that fails the build on untrue copy. (#37) Coverage claims constrained, and production verification recorded. (#45) Research service readiness and account identity verified. (#44)

The loop in nine stages, an assessment per bottleneck, dated events, and Today as the front page. (#35) The board: bottlenecks as one scannable list, a page per bottleneck, the programme as a ladder. (#32) The substrate-of-recursion programme, a research engine, a JSON map and a correction intake. (#31) 48 of 92 producer rows sourced from the engine's candidates. (#33) MIT licence. (#34)

FleetCrown was renamed to Loki throughout. (#30)

CI caches Next's build output, and the auto-merge sweep holds a token so workflow PRs stop stalling. (#24, #25, #26)

Green PRs merge themselves through the fleet auto-merge sweep, and the sweep reconciles deployment so a merged commit is never stranded. (#20, #22) Docs synced with the actual stack; the fleet's one package manager, pnpm. (#17, #19)

TypeScript 6, ESLint 10, Node 24 in CI and on the box, a formatter, and sitekit 0.3.0 for a fix this repo could not make. (#11, #12, #13, #15, #16)