Changelog
A dated record of what changed on this site: content additions, data updates, and substantive corrections, newest first. How the content itself is produced is documented in the editorial methodology.
- — The worldwide cyclotron directory is temporarily offline. The map, the full directory table, the visiting page, the registries page, and the census download are unpublished while the way the dataset is offered is reworked; their URLs carry a notice rather than a dead end. The dataset itself continues to be maintained and verified, census-derived figures elsewhere on the site still compute from it, and the amateur builds census, Design Guide, Library, legal survey, and calculators are unaffected. Research access in the meantime: [email protected].
- — The registries directory now publishes each source's scholarly layer — citation, coverage, what it gives, and caveats — and no longer describes how each source is operationally worked. Negative findings publish as a count rather than a named list, so the coverage picture stays honest without mapping the dead ends, and the directory is now a reference page only: the bulk registries export is retired (its old URL redirects here). The sources behind every census record remain fully documented and citable; the census, design-rule and bibliography datasets remain downloadable under CC BY 4.0.
- — The regulator links come back verified. The morning's review had exposed every state's "agency radiation-control program" link as either malformed or a mislabeled copy of the rule text — only six states carried a real agency page — so the legal workstream re-verified the lot against a strict criterion (an agency program or registration page on the agency's own site, never another copy of the regulations). Every state page now names its actual agency — including the reorganizations that had moved programs out from under the names the rules are filed under — and 48 of 51 link the program page with a same-day verification date; Massachusetts, New Hampshire, and New Jersey show their identified agency without a link, stated plainly as pending because those domains block automated verification. Wyoming's entry says what is true: no machine-registration program exists there, and the linked Radiological Services page is informational. The RF matching calculator also completed its referee pass (eight findings, all fixed): the tank formula now demands the unloaded Q, the power formula states its peak-voltage convention, beam loading gets its real scaling instead of a dismissal, component stress is named as the safety step the page does not compute, the tapped-coil shortcut warns that the displayed inductances do not carry over, and the bandwidth row says which bandwidth it is — the L-network match's, not the loaded tank's.
- — Corrections from a second independent review of the whole site (its dispositions are in the repo's review ledger; every numbered finding verified real before fixing). The physics-prose ones: the builds summary no longer narrows its own census — the energy range now names Rutgers 9-inch (~184 keV, computed) and Iowa State (1.5 MeV, stated maximum) with evidence types labeled; the experiments figure gains Iowa State and a hollow-marker convention separating stated-capability figures from reported operating energies, and its caption's half-the-lithium-threshold claim now names Iowa State as the exception it always was; the how-it-works phase discussion no longer calls a 2% frequency mismatch "the whole phase budget" — mismatch and accumulated phase are different quantities, and the sentence now does the 2πNδ arithmetic it teaches. Two Design Guide editorial notes carried the same absolute ("below every (p,n) threshold", "every channel closed by >600 keV") that indium-115's 287 keV threshold refutes; both corrected with dated brackets, mirrored upstream, and "every" joins the audit pattern that should have caught them. The patents page separates what it had blurred: copying a patent document (allowed, with the 37 CFR 1.71(d) copyright caveat), practicing claims (infringement includes making and using, not just selling — enforcement rarity is not permission), and term (from filing date with adjustments and maintenance-fee lapses, so check the record, not the grant year) — all verified against USPTO sources this week. Eleven state pages' regulator links were malformed rule-citations in disguise, one 404ing live; they now render as the verification gaps they are (the legal workstream has the queue). The visit page groups its 33 public-tour machines into 27 actual destinations and says plainly that a site tour does not promise sight of any particular machine. Smaller: the starter page's Sources count is computed now, its odds language states the observed fraction with its bias caveat, UCRL-476 is eighteen pages not forty, result labels on five more calculators became proper row headers, and the news filter learned that a Microsoft consultancy named Cyclotron is not accelerator news.
- — A page for researchers and librarians, at /about/cite/: what the databases are with current counts (computed at build, like every count on the site), the evidence standard in one place — per-record sources and last-verified dates, verbatim quotes on every design rule, live-fetched nuclear data, monthly Internet Archive submission of cited sources, corrections as dated notes — the scope statement with the deliberate exclusions spelled out (cyclotrons only, documented coverage, builder privacy, the builds-census boundary, no copyrighted hosting), suggested citation formats under the site's organizational name, and the link-stability commitments a subject guide relies on. Built as the stable landing page for editorial-link and outreach requests.
- — The Locations page now says what it is: the heading and title read "Worldwide Cyclotron Directory", and the intro states the scope in one breath — every machine class the dataset holds (enumerated from the data, so a new category appears in the sentence without an edit) at every lifecycle stage, from under construction through operating to decommissioned and museum pieces — with the registries link now flagged as the place to read about coverage gaps. Prompted by an outside assistant's account of searching for a worldwide cyclotron list and not finding this one: the page ranked its own name over its purpose. Internal links from the methodology and starter pages now use the descriptive name, the AI-assistant map entry matches, and the registries line there catches up to the real source count (86, not the 79 it had aged from).
- — The census gains its 27th entry, recovered by the historical campaign's 1958 mining: the Howard College Cyclotron, a faculty-led in-house build begun in 1948 at the Birmingham college that became Samford University — a 16-inch, 12-ton machine designed for 4 MeV protons, with air-cooled coils, iron-ore-block shielding, and most materials donated by local industry. Its builder filled in the 1958 world-census data sheet himself, ten years in, with the machine still under construction and the performance page blank; no record of completion or disposal has been found, and every figure is labeled the design number it is (verified against the hosted ORNL-2644 scan's page images, printed pp. 119–120). The map's 1958 batch adds 29 more institutional machines (1,418 total), and the starter-path page's census claim — which had silently aged from 17 entries to 26 — now computes its numbers from the dataset like everything else.
- — The world map gains its historical layer: 112 machines from the 1966–2004 Cyclotrons-conference world lists, matched and audited by the database workstream, bring the census of record to 1,389 machines in 92 countries. Most of the newcomers are decommissioned or of unknown fate — the gray layer nearly doubles (173 decommissioned, 105 unknown) — which is the point: the map now shows where cyclotrons have been, not only where they are. One judgment call is recorded rather than buried: the Pomona College teaching machine (a 2 MeV Hughes Aircraft build donated in 1960) joins the map but not the amateur builds census, because the census admits documented builds and a donated factory machine is not one — the same boundary that keeps hospital machines out.
- — Legal corrections from the peer review, researched and re-verified by the legal workstream against primary sources, merged site-wide. The substantive one: the survey no longer treats "discharging into a medium" as an extraction clause — only South Carolina's rule says "external to the accelerating device", Texas names no medium at all, New Jersey names cyclotrons outright, and Massachusetts and Pennsylvania treat internal-target ion implanters as accelerators — so the closed-beam reading is now presented as an arguable position, with the survey's expectations re-anchored to practice: routine registration either way. The NRC section states the 10 CFR 30.4 purpose condition — incidental beam-only activation is outside the byproduct definition, production for use is not — and the FDA paragraph now names the real reporting-table category and the abbreviated-report obligations that actually apply to analytical and industrial accelerators. Every state page gains links to its governing rule text and its regulator with a per-state verification date (nine states' regulator links are marked as open verification gaps rather than papered over), the survey table links every citation, Arizona points at ADHS rather than the agency dissolved into it in 2018, and Idaho and Wyoming's no-program labels say plainly that they are medium-confidence.
- — Safety-tier corrections from the peer review's third round, plus the search fix. The shielding widget no longer converts half-value layers to material thickness at all — its options are now HVL counts, exact by definition, because no single lead HVL is conservative across the band (just below the 88 keV K edge the NIST-derived HVL is 0.32 mm, not the 0.27 the widget assumed — retrieved this month). The experiments page's "no neutrons" summaries now say no direct (p,n) neutrons and carry the photoneutron exception in place (Be photoneutron channels open near 1.57 MeV, NNDC). The shielding page's census claim names its real exception: Iowa State's 1.5 MeV machine, about 20% under the lithium threshold with no factor-of-two margin. Search no longer erases a query typed while the index loads. The referee re-pass then corrected the activation staging a second time — 55Mn(p,n)55Fe at 1.03 MeV is the first structural channel, since manganese is in practically every steel; the threshold table gains manganese and chromium rows and the all-clear boundary moves to ~1 MeV — and hardened a dozen more statements: capacitors hold charge for hours or days without working bleeders, coil-guarding thresholds are not touch-safety limits, home machines answer to the general-public RF tier, and neutron work needs neutron instrumentation, which a GM tube is not. Legal findings from the same review are with the legal workstream.
- — Vacuum calculator and vacuum-budget page corrections, from the second independent peer review (ten findings) plus the site's own referee passes (twenty more) — all dispositioned. The calculator's real defects: orbit mode silently applied its hydrogen loss tables when air was selected (the gas is now locked to H₂ in orbit mode, stated on the control); mode-hidden inputs stayed visible and editable while being ignored (a shared-style override — fixed at the component, and inactive controls are disabled); and the turn sum could hand the cross-section and the geometry different energies when the turn count didn't divide evenly (the model now uses one energy per step, with a final partial turn at its midpoint energy — every published worked value recomputed against an independent implementation, and all displayed numbers stand). The pressure input is now labeled true gas pressure with the O'Hanlon H₂ gauge-sensitivity cited; invalid input clears the chart with the table instead of leaving a stale curve; sub-microvolt typos can no longer hang the page in a billion-turn loop. On the budget page, the worked example now carries one case end to end — 0.3 sccm gives 2.9 × 10⁻⁵ torr total in the dee and 73% proton survival, replacing three inconsistent roundings — and a paraphrase of ORNL-1196's gauge-placement advice that had drifted from the source while wearing quotation marks now quotes the report verbatim, network paths corrected. A dozen further claims were scoped to what their sources support, including the LN₂ cold-trap safety caveat the pump section had been missing.
- — The peer review's remaining findings are closed, all eight now dispositioned. The energy calculator explains invalid input inline (positive-ion domain stated, with the charge-magnitude convention for negative ions) and clears stale warnings; its metadata describes what the form actually takes; a compact energy-frequency-rigidity summary sits directly after the inputs (politely announced to assistive technology, and the mobile fix at the same time); result labels are proper row headers; energies are labeled per ion; the radius input is an orbit radius, since the calculation never required extraction; and the references link to real targets — NIST for CODATA, the library for Livingston & Blewett. Site-wide wording: the census now claims documented coverage rather than exhaustive coverage on the homepage, the builds page, and the AI-assistant map, and the Learn careers card drops a stale typed count.
- — Energy-calculator physics corrections, from an independent peer review of the site (its reports and browser evidence now live in the repo). The real defect: the results table computed speed classically as p/m and could display a superluminal β labeled "final speed" beside an exact γ — speed and β are now physical (p/γm), and the table shows the classical cyclotron frequency and the relativistic revolution frequency f/γ side by side, labeled, with the h = 1 assumption stated. The subtler one: the 1% relativity warning claimed a machine-feasibility meaning, and both the calculator and the how-it-works page quoted a flat 1–2% frequency-slip tolerance. The warning is now an approximation gauge only, and the phase section derives the tolerance from turn count — a constant mismatch δ slips phase by 2πNδ, so a 100-turn machine holds its ±90° budget only below about 0.25% — with the Kleeven & Zaremba CERN-school treatment added to the sources. Also fixed: a floating-point cancellation that reported zero relativistic energy at tiny momentum (stable hypot/γ+1 forms now, validated against the reviewer's independent reference to 1e-9 on all recorded cases), the relativistic force balance shown beside the classical one, rigidity defined with |q|, and honest input-domain wording. The calculator now carries its own review date; the referee re-pass ran and its findings are dispositioned in the review's own tracking format.
- — The safety page gains the hazard that arrives first: pump-exhaust oil mist. An oil-sealed rotary-vane pump aerosolizes oil out its exhaust — "looks like smoke", per the hobbyist write-up behind the library's documented $45 exhaust filter — and it scales with pump-hours, not beam: a build's vacuum system runs for months or years before any other hazard on the page exists, usually in an enclosed basement. The section cites the NIOSH/OSHA mineral-oil-mist exposure limits (5 mg/m³, retrieved this month) and the documented filter, and the hydrogen section now notes the pump exhaust is part of the gas path. The addition triggered a full-page external referee pass, and all twenty findings were fixed, including one real internal contradiction: the activation staging said copper and steel activate "above ~4–5 MeV" while the page's own threshold table lists ⁶⁵Cu(p,n) at 2.17 MeV — the staging now runs 1.65 MeV (⁵⁷Fe, verified against NNDC this round) → 2.17 → 4–5, and the all-clear boundary moved down to ~1.5 MeV. Five other safety absolutes were softened to what the sources support, the fusor neutron-shielding number is now labeled the community rule of thumb it is, and the registration rule is re-anchored to before-the-machine-can-make-X-rays rather than before first beam. Dispositions in the page's notes file.
- — Erratum round for the hosted Mark I record, from the builders' side: the simulation-models note is now v0.2 (13 pages) with a dated revision section listing four corrections — builder-measured coil geometry, an outer-boundary defect in the 3-D mesh fixed and backed by a pole-edge mesh-convergence check, the dee remodelled with its real open mouth instead of a closed block, and a deflected-lid capacitance accounting; the magnet note picks up the measured coil form and a field-polarity sentence (field up through the gap, protons clockwise seen from the top lid). All four model packages are rebuilt to match (the electrostatic set grows the open-mouth cases; the Elmer set the edge-refined cases). The other six notes are byte-unchanged.
- — Correction, per the builders: the Mullins Cyclotron is a basement build, not a garage build. Fixed in the census summary, the amateur-collection page, and the study path.
- — The hosted Mark I record is now cross-linked where it answers a page's question: the Design Guide's reference-machine block points at the notes, so its editorial notes are checkable at document depth rather than stopping at the census entry; the census entry links the notes and models beside the build log; the starter path's cost discussion cites the as-built bill of materials as the census's one complete itemized record; and the beam-measurement, vacuum-chamber, and RF/dee pages cite the matching subsystem notes in their reading lists.
- — The site hosts its first contributed document series and its first downloadable simulation models: Notes on the Mullins Mark I Cyclotron — eight technical notes (116 pages, CC BY 4.0) on the census's Mullins Cyclotron, written up from the machine's own records and hosted with the builders' permission. The series covers the as-built electromagnet, the chamber and vacuum system, the RF chain and ion source, instrumentation and the field-sweep resonance-mapping method, an operator's manual, the seven-year performance record, an itemized as-built bill of materials against the 2017 budget, and the machine's simulation models. Alongside them, four downloadable model packages: the builder's original FEMM design models, an as-built FEMM rerun compared against the measured field, an electrostatic dee model, and a 3-D Elmer H-frame model — each a zip with a README stating what the model is, how to run it, and results against measurement, under the same license. They join the machine's section on the amateur & teaching machines collection, and the modeling-tools page links the packages as a worked example beside the solvers it surveys. Every file went through a publication review before hosting: full-text privacy scan, embedded-image metadata check, license and author lines verified on every title page and README.
- — The library gains a second cross-cutting collection: Amateur & Teaching Machines (/library/amateur/) — the 42 documents the small-machine community wrote about its own machines, regrouped by machine across eight programs, from the Niell home builds of 1993 through the Rutgers/Maryland line, Houghton College, the MIT 2 MeV project, the Cyclotron Kids, COLUMBUS, the Kwolik St. Cloud build, and the Mullins Cyclotron. Each section links its builds-census entries and each document its library annotation, with Internet Archive badges wherever a snapshot is verified; the Rutgers/Maryland section defers to its dedicated reading guide rather than repeating it. Groupings live in a dataset with build-time guards (every id must resolve, no document may belong to two machines), and every count on the page is computed.
- — The scan re-read campaign lands: all 96 queued unverifiable details from the audit campaigns were settled against 300–600 dpi page images, and every result is now on the cards. 85 items were confirmed — the quotes extend verbatim to carry figures that OCR text layers had dropped, restoring more than a dozen wrongly-removed claims (the 10 kW per-dee heat load, the observed ten-to-one resonant-charging step-up, the 1%-per-minute servo slew, the ±0.2 mrad alignment tolerance, the 500 µA carbon-septum survival, Forringer’s 70% beam-to-aperture ratio with its meaning corrected, among others). Nine items were refuted and corrected with dated brackets: the sub-$1000 300 keV machine was a stated goal and budget cap, never a claimed achievement; Stanford’s census field was 12.3 kG as-run (12.5 was its design sheet — the design-versus-operating lesson inside the census itself); Rochester’s “3–4%” drop-off was a misread decimal 3.4%; a dee-capacitance comparison read 1600 µµF as 1000; and one earlier audit fix of ours was itself wrong — the 1972 manual sets its public dose tier “lower by a factor of ten”, not ~30×, and the card now says what the page says. One attribution error is fixed (a shielding datum credited to Moyer belongs to the Nevis report’s own authors), and four source typos are documented in place, including a printed heating constant missing its minus sign and an equation whose printed denominator disagrees with the manual’s own worked example. The census was checked against the refutations and needed no changes.
- — Wave 18: the Design Guide grows by 242 rules (now 1,878 across 166 sources) and the library by 13 documents (283) — the Rutgers/Maryland 12-inch corpus, fully mined the same day its reading guide shipped. Every new rule went through the external audit before this announcement: 242 audited (78 FAIL / 146 REVISE / 18 PASS — the usual ~92% non-PASS of a fresh wave), and all 224 findings are dispositioned. As usual most findings dissolved on source contact: 43 quotes extended verbatim (the contested facts were on the page, six of them read off the 9-inch writeup’s page images, since that document is a scan with no text layer), and one finding was declined outright with page-image evidence — the referee suspected “270° spiral pole-tips” was a text-layer fusion of “2 70°”; the page prints 270 degrees. The genuine catches are recorded on the cards: a note that had inverted the phase-resolution proportionality, a wrong Child-Langmuir gloss, a dee-voltage factor-of-two explained incorrectly, a dimensionally invalid spiral-edge relation corrected to tan ξ = r·dθ/dr, and a dose-to-source-strength conversion route rebuilt with the worked arithmetic. Source-internal problems are documented in place with dated brackets rather than silently fixed: a “100 ms” ring-up delay impossible at the paper’s own 20 Hz, a diffusion-pump speed printed in liters per MINUTE, and a harmonic-mode explanation whose printed fields sit at an even, not odd, multiple — flagged as unresolved. The census picked up the same discipline: the 12-inch’s deuteron pair (7.15 MHz / 0.96 T) is now labeled as recorded settings ~2% off mutual consistency, not a matched pair.
- — The library gains a reading guide to the Rutgers/Maryland 12-inch corpus (/library/rutgers-umd/): the seventeen papers the best-documented amateur-origin cyclotron’s builders wrote across two decades — construction and commissioning memos, the Cyclotrons 2013 and CAARI papers, and the Maryland-era neutron reports — each linked to the authors’ own hosted copies on their NPL collection page, with Internet Archive fallbacks rendered wherever a snapshot is verified. Deliberately link-through: the documents are their authors’ copyright and their collection stays the canonical home; the archive-coverage cycle now watches these links alongside the census and library citations. Titles, authors and dates were transcribed from each PDF. In the same pass the census entries pick up the builder’s newly restarted project history at timkoeth.com — which also supplies the Maryland 19-inch’s first stated design energy, a 5 MeV target, labeled as such — and the builds page’s program section now records the lineage in the record’s own words: COLUMBUS’s ion source “built after the pattern of Tim Koeth”, IUAC designing with the 12-inch’s builder advising, and students elsewhere using the documentation as a model.
- — Site-wide numbers sweep. Every count in page prose was checked against the datasets; six pages carried typed rule or census counts that had gone silently stale and now compute them at build time: beam dynamics said 209 rules where the domain holds 245, beam quality repeated the 209 twice, beam extraction’s provenance line said 84 against 89, the RF count on choosing-your-machine said 333 against 366, the industry page’s census figures (1,250 machines, 804 with manufacturer) had drifted to 1,264 and 818 — and the ion-source page managed to disagree with itself, 114 rules in the introduction and 154 in the closing list, against an actual 189. The AI-assistant map (llms.txt) was refreshed the same way: the builds line still described a 19-machine, four-country census (26 machines across six countries including India and South Korea), the map lines said 1,259 machines (1,264), and the classics line now states the checkable facts — 56 hosted PDFs across 26 pages, a queue of 10. Verified clean in the same pass: census and map agree entry for entry, Iowa State’s 1.5 MeV and the Mullins ≈164 keV figures are consistent on every page that states them, and the cited-numbers audit reports no new mismatches.
- — Library dedup: the Antokhin permanent-magnet PET-cyclotron paper existed twice (lib-082 and lib-083 — the phantom 2004 year came from an archive copy missing the RuPAC 2006 page stamps; the registrar diffed the two copies and confirmed one document). The duplicate entry is removed, the design-guide anchors moved to the surviving entry, and the library stands at 270 documents. The same investigation resolved a mis-citation on the safety page: its “~560 gauss at the zero setting” permanent-magnet warning, loosely credited to a 2010 conference, is in fact the measured residual of Antokhin’s own flux-engine magnet (RuPAC 2006, p. 2) — now cited to it on both the safety page and the new permanent-magnet section, which reads the number both ways: easy disassembly by the builders’ lights, never magnetically safe by ours.
- — The magnet-design page gains its permanent-magnet section, built on the wave-16 PM corpus. It carries the trade in one place: what deleting the coils buys (LBNL’s 1 T, 30 cm SmCo machine, judged portable; Antokhin’s 900 kg, 1.4 T PET-cyclotron magnet with its rotatable flux-engine off switch) and what it costs (the current knob — the Cyclotrino team’s documented reason for choosing an electromagnet). The section covers rare-earth-versus-alnico and the Br/2 open-assembly bound, the volume-scales-with-energy sizing relation, the Halbach crown-and-barrel circuit with the iron pole as precision element, correction without a current knob (oversize-and-cut-back shimming, trim coils, movable shunts), the −0.07%/°C temperature drift and its NiFe compensation, and assembly forces — cross-linked to the safety page’s never-off warning. The record’s state is stated plainly: the amateur-scale prototype gap is Houghton’s 2002 thesis at 0.493 T, and none of the permanent-magnet cyclotrons in the library records a demonstrated beam. Externally refereed before publication; six PM sources join the page’s source list.
- — Design Guide 1,545 → 1,636 rules and the Library to 271 entries, in two waves mined together. Wave 16 is the permanent-magnet magnet-design corpus the guide lacked: the LBNL cyclotron mass spectrometer pair (Clark/Halbach, Cyclotrons ’95; Young, PAC ’93) — crown-and-barrel PM topology, the pole as precision element, oversize-and-cut-back shimming — plus the 2025 NSRRC hybrid PM/EM dipole with its trim-coil, movable-shunt and NiFe thermal-compensation calibration data. Wave 17 documents the two newest census machines at rule depth: the Iowa State trio (Burns’ beam-technology experiments, Mueller’s phase-integral calculations, Moses’ trajectory integration — a complete 1961–63 model-versus-machine commissioning loop) and the IUAC document set (the magnet tender’s procurement-grade specifications from steel chemistry to hipot acceptance, the IPAC2026 LLRF paper, the annual report and a partner-institute report). All 91 new rules went through the external referee before this announcement and every finding was dispositioned: 89 quotes were extended verbatim after scan checks — the audited facts were on the cited pages — while the editorial layer took real corrections (441 keV called a resonance rather than repeating the source’s “threshold”, the NiFe drift reduction stated as 37% rather than “roughly half”, safety scoping on hipot testing and bucking-field assembly, machine-specific figures no longer offered as universal budgets). Five source-internal errors are now surfaced on the cards themselves, dataset-errata style, the largest being a lost minus sign in Moses’ printed field formula — as printed the field would rise a kilogauss at the pole edge, contradicting the paper’s own figure and its n = 0.2 resonance radius; the correction was verified against the page image and by recomputing the resonance location.
- — Five more builds join the census — a recovered historical layer, every claim re-verified against its primary source: Carl Svoboda’s 1955 Oakland machine (a Bettmann press photograph is the whole surviving record), Reinier Beeuwkes III’s electron cyclotron, which won the 1958 Westinghouse Science Talent Search (official winners booklet + Popular Electronics, and the census’s one electron machine), Seoul National University’s 1959 graduate build — Korea’s first domestically built accelerator per the official biography of Kwon Young-dae, whose extraction claim is reported as the biography’s, without a surviving technical record — Albert Swank Jr.’s circa-1967 teenage build (distinct from his 2005 relocation project; one builder, two machines), and James Kwolik’s unfinished 9.25-inch St. Cloud machine, documented by his own 2015 account in this site’s library. A new Fermilab-thread paragraph records Ron Walker’s and Chris Olsen’s personal builds as builders’ history — no machine of theirs has a recoverable record, so none gets a row. Four leads (Greenberg, Tan, Guanajuato, Zarco) were vetted and declined, with reasons in the page’s design notes. The census stands at 26 entries; the map at 1,264 machines; the programs-intro beam split and the extraction page’s census count are now computed rather than typed.
- — The census gains its first national-laboratory build and its first Indian machine: the IUAC Tabletop Cyclotron, under development at the Inter-University Accelerator Centre in New Delhi since 2024, with Timothy Koeth of the Maryland 19-inch advising remotely — a documented design lineage from the Rutgers/Maryland educational machines. The record is procurement-grade and honest about what it is: a magnet tender specifying 1.2 T on 305 mm poles, two fabricated chambers and a Dee–dummy-dee RF system in the annual report, RF bench-tested to 200 W at 18.2 MHz in an IPAC2026 paper — and no beam demonstrated, with two unreconciled design energies (1 and 1.2 MeV) listed as such and the director’s stated milestone, beam extraction, quoted rather than assumed. A new “accelerator lab” builder type joins the comparison table, and the census count moves to 21 (map: 1,259 machines).
- — The census gains its earliest university build and its highest documented beam: the Iowa State University Undergraduate Cyclotron (1954–at least 1963). Begun by the Undergraduate Physics Club in fall 1954 and first run in spring 1957, the 2.5-ton, 10-inch machine accelerated protons to 1.5 MeV at 2 µA — above the Berkeley 11-inch’s 1932 milestone figure — and served laboratories, NSF undergraduate research and two master’s theses, including N-13 production from carbon. Four Iowa Academy of Science papers document it (McGuire, Burns, Mueller 1961; Moses 1963, all on UNI ScholarWorks); its fate after 1963 is undocumented. The addition moved the documented amateur record’s top from 800 keV to 1.5 MeV, and six pages that stated the old bound — experiments-by-energy, choosing-your-machine’s band four, beam measurement, should-you-build-one, beam quality, and the history chart (whose above-1-MeV guard fired exactly as designed) — were corrected with it. A matching map record joined the world census (now 1,258 machines).
- — Census and registries re-synced from the database workstream after its second external review round. The headline correction: the Weill Cornell CBIC record was wrong, not merely source-less — the operator’s own facility page shows an ACSI TR-19/9 (19.2 MeV protons, 9.5 MeV deuterons), not the Siemens Eclipse the record asserted from an anonymised IAEA row; the record is now sourced to the operator and its note preserves the history. Central High School Philadelphia’s tours field corrected to none (the linked tours excluded the cyclotron), so the visit page no longer lists it. Housekeeping across 95 records: 84 FDA source links now open the DECRS landing page rather than downloading the 2.4 MB bulk file (dated Wayback snapshots kept), Japan Steel Works normalised to one label on nine records, and the PTCOG registry entry rewritten with its live URL. Record count unchanged at 1,257.
- — Erratum on dg-1031 (the MSUCP-12 dee-gap-field benchmark table): an upstream implementation-plus-scan verification proved the transcribed equations exact but found the printed Table 1 reliable only to about three decimals — the 1960 computation carried its intermediate roots to four decimals, up to 3.2×10⁻³ relative error at k/h = 2.0. The card now says to verify implementations against the equations rather than the printed table, and that matching the table to only ~10⁻³ is the signature of a correct implementation; a five-figure match would mean reproducing the paper’s rounding. Dated bracketed note per the dataset-errata protocol; the verbatim quote is unchanged.
- — New page: The Vacuum Chamber and Its Seals (/learn/vacuum-chamber/) — the tenth subsystem deep dive, over the chamber and seals Design Guide domains that waves 13–15 filled out. The vessel half treats the chamber as the thing living inside the magnet gap: the gap stack-up, lid deflection (with the record’s one predicted-versus-measured pair, from the Mullins build log — 0.030 in designed, 0.033 in measured), magnetic force on ferromagnetic covers, material choice, port budgets, in-vacuum clearances and conditioning. The joints half carries the gland arithmetic, elastomer data, the grease question, the double-gasket guard vacuum, weld and virtual-leak discipline, and motion feedthroughs as their own failure class. Eight census chambers serve as worked examples, from El Cerrito’s 1947 rebuild to the 2019 printed chamber. The chamber and seals domains also gained their shared anatomy figure, rendered on both Design Guide pages. Externally refereed before publication; the vacuum-budget page now hands the vessel and its joints here.
- — All seven calculators are now shareable by URL: the address bar tracks every input that differs from its default, so a specific calculation — a coil geometry, a vacuum budget, an energy point — can be copied into a forum thread or an email and opens exactly as configured. A link with no parameters behaves as before, and pages without JavaScript are unchanged.
- — Design Guide 1,515 → 1,545 rules and the Library to 261 entries: the COLUMBUS JACoW series, mined as one machine’s decade-long narrative. Four open CC-BY papers join the library and the census entry’s sources — the 2013 design snapshot (whose parameter table now anchors a documented design-versus-operating-point contrast three layers deep for this machine), the 2013 ion-source companion, the 2016 simulation-and-detection paper whose predicted slow molecular-ion bundle matched a measured impostor peak, and the 2019 study that leak-tested 3D-printed 316L vacuum hardware against wrought stainless, cleaned and uncleaned, with printed numbers. All 30 new rules went through the external referee before this announcement and every finding was dispositioned: most dissolved into verified verbatim quote extensions — the two design conditions, the free-of-charge chamber, the 45° support threshold, the leak-rate table, and the simulation’s architecture were all on the cited pages just beyond the quotes — while genuine overreach in the editorial layer (a universal frequency-coverage claim, a “no helium leak detector required” promise, a leak-freedom inference from a reached pressure, a universal Viton-dominance hierarchy) was cut or scoped. Two published rules gained errata from upstream re-reads: the extraction turn-separation formula is now scoped nonrelativistic, and a queued scan re-read delivered the volts-per-turn comparison values it was waiting on. The study path’s COLUMBUS entry now points at the whole open series.
- — The study path’s working-machine set grows from three to four: the COLUMBUS school cyclotron in Coburg joins Rutgers/Maryland, Houghton and Mullins. Its record adds a fourth documentation style — a published book — and the one thing the other three do not cover: a decade of running a cyclotron as a teaching instrument. The entry states the access caveat plainly (the book is German-language and commercial; the freely readable layer is its two open conference papers) and reports the builders’ no-extraction rationale as theirs, along with their own later revisiting of it.
- — Search is now federated across the site’s datasets. A query on /search/ returns counted groups — design rules, library documents, census builds, mapped machines, and pages — and each dataset item is its own result landing at its own anchor: a rule on its card, a document on its library entry, a build on its census entry, a machine on the world map, which now accepts ?id= deep links and flies to the record. Search was previously page-granular, so the design rules could surface as at most 24 domain-page hits and no mapped machine could surface at all. The group counts are computed per query by the search engine, and each rule, document, build and machine exists in the index exactly once.
- — New page: How to Learn to Build a Cyclotron (/learn/study-path/) — the self-study curriculum the field’s small literature makes possible. Three tracks: a reading sequence (the site’s introductory pages, then Wouters’ UCRL-476, then Livingston & Blewett chapters 5–6, then the Design Guide by subsystem while designing), the three census machines whose records are deep enough to study end to end — the Rutgers/Maryland lineage with its video course, Houghton’s chain of student theses, and the Mullins build log’s photographic record — and the practice layer, from safety and state regulators through the adjacent communities. Externally refereed before publication; the reading order on So You Want to Build One now points here for the external half, as does the resources page for readers who want a sequence rather than a map.
- — The numeric-fidelity audit was rebaselined over the vacuum deep dive, which shipped after its last baseline, and its 25 open flags were each read against the sources. One was a real error of exactly the class the audit exists to catch: the vacuum page rendered the gas-density constant as 3.3 × 10²² per torr where the cited rule and the physics give 3.22 × 10²² at 300 K — corrected, with the temperature the constant assumes now stated. Two flags exposed facts that were in the sources just beyond the quotes, which are now extended verbatim: the fusor thesis’s chamber surveys detected X-rays from 18 kV (its hazard line is 20 kV — an earlier audit had trimmed the 18 as unsupported, and its page citation has been corrected to the pages the quote actually sits on), and the MIT thesis’s ~300 m expected flight distance now sits inside its own rule’s quote. The remaining 22 flags are benign — machine names, report numbers, superscript renderings, and a worked table that labels its own calculator provenance — and are baselined as reviewed.
- — The reference machine now has a published parameter block, and every Design Guide note has been reconciled to it. An external review noticed that the notes described one machine with a pole gap of 3 cm in one place, 5 cm in another, and 1.2 or 1.5 inches elsewhere — and paired three different usable radii with energies from 115 to 164 keV. Every calculation was internally correct; the inputs were not one machine. /design-guide/ now opens with the machine’s parameters in full, sourced entirely from its builds-census entry and public build log, and labelled by what each figure IS — the field it operates at against the field it has reached, the energy computed rather than measured, the dee voltage read from an uncalibrated pickup coil. Twenty-three notes were corrected against it, including two that were simply wrong: a hydrogen flow quoted at 2 sccm was a controller setpoint read as a flow, on a controller whose full scale is 1 sccm, which inverted the comparison the note drew; and the gap correction moved four separate ampere-turn, field and fringing results. The block also explains the tension the review found: the machine collects beam further out than its flat field extends, so the energy the uniform-field formula predicts sits below what the machine demonstrates.
- — A counting-layer pass over the pages whose prose restates their own data, after an external review found five documents listed as “in the queue” on the Construction Classics page that the same page already hosted. The collection’s size line was five entries and ten PDFs stale (it read twenty-one and forty-six; it is twenty-six and fifty-six) and is now computed, and the RF Power cluster — titled “three reports” while listing four report numbers and hosting four PDFs — is corrected along with its size total. Four build-time guards now hold these together: a document cannot be both hosted and queued, and a cluster’s title, size string and report-number list must all agree on how many PDFs it hosts. AECU-1951 is dated 1951 rather than 1950 on the Classics page, matching the Library and the archive’s own record — the report number reads like a year. The locations map legend is generated from the same category list the table uses, so it no longer shows three markers for the six the map draws (teaching and proton therapy have their own; industrial was missing). The Design Guide’s level chips now say why they add to 1,512 rather than 1,515: three rules bind lesson design rather than machine design and sit off the axis entirely. The legal page’s jurisdiction counts are counted from the survey rather than typed, which fixes a group of 35 described as “states” that includes the District of Columbia; its US-only scope is now stated at the top rather than in the last paragraph; and the careers page links the state-by-state survey to the survey, with its family table no longer showing the medical track’s two lanes as two families.
- — Design Guide 1,374 → 1,515 rules and the Library to 257 entries, from two teaching machines and the only published survey of amateur builds. The COLUMBUS teaching cyclotron arrives twice — the builders’ own Springer book (in German, quoted with translations) and their 2022 conference retrospective — and brings the guide its first two new domains in months: controls & instrumentation (gauge chains, mass-flow feeds, Hall-probe field readout) and pedagogy (what a real accelerator adds over the textbook treatment, and the workshop formats that carry it). The Cressman thesis contributes the survey chapter that exists nowhere else: El Cerrito’s 1947 high-school machine, the Cyclotrino, Niell’s 1995 build, both Rutgers machines, and Knox College’s — including the failure that stopped Knox outright, unsecured source wires moved by the field until they shorted the dees, a Lorentz-force lesson the 1931 first cyclotron had already engineered against. All 141 new rules went through the external referee before this announcement, and every finding was dispositioned: 43 quotes were extended after checking the scans, three source-internal errors are surfaced on their cards rather than repeated (a printed Q-value that is actually the deuteron binding energy among them), and two vendor-table misprints arrive pre-flagged do-not-copy by the upstream index.
- — 323 further Design Guide rules corrected, completing the follow-on external audit of the 782 rules outside the level-1/2-and-safety campaign. Every finding from the 782-rule audit now has a written disposition, and with it every published rule has been through external audit and correction. This wave (the REVISE-severity tail) resolved differently from any before it: a third of the findings — 106 rules — dissolved into verified verbatim quote extensions, the “unsupported” procedure details, mechanisms, and construction specifics turning out to be in the sources just beyond the excerpts, from a complete 1950s five-step RF tuning procedure to the full pack-rolling craft of the target-makers’ proceedings. The rest are the familiar classes: one machine’s numbers relabeled as that machine’s, methods transferred with their validation conditions attached, and editorial mechanisms downgraded to hypotheses. Five citations gained a page; a handful of scan-illegible figures are queued for page-image re-reads upstream.
- — 428 Design Guide rules corrected — the first wave of the follow-on external audit covering the 782 rules outside the completed level-1/2-and-safety campaign (levels 3–5 and level-exempt, audited by the same qualified referee and brief). Every FAIL-severity finding is now dispositioned; the REVISE tail follows in a second wave. The defect classes repeat the first campaign’s — machine-specific figures universalized, invented scaled numbers, absolutes now scoped — with this wave’s distinctive find being how much of the “unsupported” material was actually in the sources: 85 verbatim quotes were extended after checking the original scans, in each case recovering on-page facts just beyond the excerpt. A handful of source-internal contradictions (a printed current that fails its own arithmetic, a materials table that transposes its own examples) are now surfaced on the cards rather than silently repeated. Verbatim quotes were never altered except by verified extension; two citations gained a page (the extension text sits on an adjacent page of the same scan); roughly thirty report-attributed figures await page-image re-reads upstream.
- — 313 further Design Guide rules corrected, completing the 592-rule external audit campaign begun with the level-1 and safety wave earlier today. Every finding the audit raised now has a written disposition. The corrections follow the same classes as the first wave — machine-specific figures attributed instead of universalized, era-labeled dosimetry and limits, and scoped absolutes — plus a run of failed-arithmetic repairs (worked numbers recomputed against their own formulas, a flutter-convention error that would have understated focusing, a field-index sign error, and the historical ALAP expanded correctly against modern ALARA). Verbatim quotes untouched throughout; 62 report-attributed figures the scans’ text layers could not verify are queued for page-image re-reads upstream.
- — 120 Design Guide rules corrected in the first wave of a systematic external audit of the guide’s most-exposed rules — every level‑1 rule and every safety-axis finding from a 592-rule campaign (levels 1–2 plus the safety and shielding domains, audited by the external referee under an upgraded brief). The defects were almost entirely in the layers this site writes, not in the quoted sources: machine-specific figures stated as universal limits, 1972 dosimetry carried as current practice (now era-labeled with modern values beside them), safety absolutes now scoped with their exceptions named, and two worked numbers that failed recomputation. Seventeen verbatim quotes were extended after checking the original scans — in each case the “unsupported” fact was in the source just beyond the excerpt. Every finding has a written disposition; verbatim quotes were never altered except by verified extension; corrections are queued upstream to the corpus.
- — Three Design Guide corrections, found by the rule audit’s own calibration battery: dg-057’s formula used L for both coil inductance and magnet length (the source’s symbol) — disambiguated; dg-840’s editorial note generalized one report’s worked example into “amateur systems routinely lose 40–60%” — the numbers are now attributed to the report and the invented population claim is gone; dg-1044’s note stated a neutron-shielding material choice as “already decided” — it now keeps the conservative warning and defers sizing to the actual source term and survey. Verbatim quotes untouched; dispositions recorded; corrections sent upstream.
- — A title and description pass over the section landing pages, so each page’s title says what a searcher would type rather than a bare section label: Cyclotron Design Guide, Cyclotron Calculators, Cyclotron Library, Amateur Cyclotron Builds, Cyclotron Locations: World Map & Directory, and Is It Legal to Build a Cyclotron? — which has been the legal page’s own heading since launch. The 22 per-domain rule pages now lead with the machine too (“Cyclotron magnet design rules”). Page headings, navigation, and content are unchanged.
- — New page: Cyclotrons You Can Visit (/locations/visit/) — the census’s public-tour machines listed by country, working facilities and preserved early machines alike, each linked to its facility’s own tour page, which stays the authority on hours and booking. The verified virtual-tour list moved here from /locations/, which keeps a pointer card and the map’s ★ public-tours filter. Every count on the page is computed from the dataset at build time.
- — A discoverability pass, prompted by an external review that found the site underselling its own scale. The homepage now opens with a computed strip of headline counts (machines mapped, countries, design rules, library documents, amateur builds, calculators) and a starting line with two doors: the Design Guide filtered to level 1 — the rules that govern any machine’s feasibility — and the public-tour list. The Design Guide intro gained the same level-1 entry point, and the builds census moved “What the census teaches” above the comparison table, so the findings meet the reader before the database does.
- — The careers section’s two main pages merged into one guide. The role-card hub and the “Without a PhD” routes page turned out to be near-parallel cuts of the same territory — each entry route led to the roles that route serves — so /careers/ now carries seven career families ordered by the lowest credential a documented posting accepted: craft trades, operations, field service, the medical track, engineering, radiation safety, and — once, last — the PhD path. The ordering makes the section’s thesis structural rather than asserted. Every verified title, quote, and review disposition from both pages carries over; /careers/without-a-phd/ redirects here.
- — The careers section moved to the top level: /careers/ (hub), /careers/without-a-phd/, /careers/university-programs/, /careers/schools/, /careers/self-study/, with a Careers entry in the site navigation. The six pages shipped under /learn/ over the previous two days, but the section serves a different reader than the Learn course — students and career-changers rather than builders — and had grown to section scale. Moving on day one kept the cost near zero: the old URLs had no inbound links yet, and all five now 301 to their new homes. The Learn landing keeps a single pointer card.
- — The careers section is complete: /learn/careers/ is its hub, carrying the ten-role taxonomy as uniform cards — what each job is in employers’ own posting language (dated sightings, August 2026), the credential floor, the exact verified titles to type into a search box, and where those roles actually post — plus a short high-school subsection (courses, the cloud-chamber-to-ham-licence ladder) and routing to the four section pages shipped earlier today. Six of the ten roles have no public career explainer anywhere that this site’s research could find; the cards for those lean entirely on primary sources. Externally reviewed before publication, dispositions in the page’s design notes.
- — The careers section grows by three pages, drafted in parallel from the committed research dossiers and reviewed sequentially. /learn/university-programs/ is a verified directory of university accelerator programs on three independent axes (whether students operate a machine, named coursework in live catalogs, formal versus customary lab pipelines), rendered from a new dataset with per-row verified dates and a build guard — and a “programs end” section built from the verification pass’s rot register, IUCF first. /learn/accelerator-schools/ catalogs the non-university training layer (USPAS with its open materials archive, CAS, JUAS, DOE SULI and CCI, lab internships, international placements, medical-side training) with application windows stated only as seasonal patterns, because a stale deadline harms a reader in a way a stale citation does not. /learn/self-study/ sequences the site’s own pages, hosted classics, and the free accelerator-physics canon into three reading levels, each ending in a self-test. Counts on all three pages are computed from their datasets at build time.
- — New Learn page: Accelerator Careers Without a PhD (/learn/careers-without-a-phd/) — the first piece of the chartered careers section. Five documented routes into accelerator work (operator, engineering BS/MS, the medical track, vendor field service, the craft trades), each grounded in fetched primary sources: the HEPAP 2014/2015 and Snowmass 2022 workforce reports quoted verbatim, certification-body rules (NMTCB, JRCNMT, the UC operator series), DOE traineeship and apprenticeship program pages, and employer postings cited as dated sightings because requisitions rot in weeks. No salary figures by policy; the medical section’s “where most of the cyclotrons are” claim is computed from the locations census at build time. Externally peer-reviewed before publication; the editor’s thirteen findings are dispositioned in the page’s design notes.
- — Design Guide rules now point back at the pages that discuss them. A rule card whose rule is cited by a Learn deep dive carries a “Cited in:” line linking there — 151 rules at present, computed at build time from the Learn pages’ own citations, so the index maintains itself as deep dives are written and asserts nothing editorial: it says a page references the rule, which is checkable by clicking. The reader’s path this closes: search and permalinks land people on the evidence layer, and until now a rule gave no hint that a narrative synthesis spends three paragraphs on it. The same scan now fails the build if a Learn page cites a dg-id that matches no published rule — a rotted citation used to ship as a silently dead anchor. The companion idea from the same external review, a standing agreement/conflict layer between rules, is declined with reasons recorded: the reviewer withdrew it themselves after reading the audit history, and where rules genuinely tension, a deep dive or figure resolves it under referee review, one case at a time.
- — Every Design Guide rule now carries an applicability level, 1–5. The level answers one question — how early and how universally does this rule bind a cyclotron design? — from fundamentals that govern any machine’s feasibility (level 1, 31 rules) to rules for large or professional machines only (level 5, 29 rules). It ranks breadth, not weight: a level is never permission to skip a rule whose trigger a machine has, and safety rules are never skippable on level alone. The point is approachability — “magnet, level 1” is a shortlist of 15 rules instead of an undifferentiated 333, and levels 1–2 together cut the domain roughly in half — without this site asserting which rules will doom a project, which is a safety claim the external review that prompted the feature had proposed and the audit history argues against. Levels were assigned by the corpus maintainer under a written rubric with a hand-reviewed anchor set, a two-pass calibration against predeclared agreement thresholds (88% exact, 100% within one level), and a post-run audit that demoted two level-1 assignments before delivery; the method is on the methodology page, and a benchmark gate governs how future rule batches get ranked. On the site: a level badge on every card, level chips beside the domain chips on the filterable view (?level= deep links, composable with ?domain= and ?q=), a by-level line with filter links on each per-domain page, and a level field in the CC BY data export. Some domains start at level 2 by structure — no measurement rule gates feasibility — which is description, not omission.
- — “How a cyclotron works”: the phase-stability figure moves to where its physics lives. The Veksler–McMillan mechanism and its diagram sat in the classical timing section, where the diagram’s caption had to assert that a synchrocyclotron’s revolution period lengthens with energy — a fact the page only explains two sections later, in the relativistic limit. Both now sit beside the synchrocyclotron itself, so the sequence reads: γ makes the particle slip, the synchrocyclotron sweeps the frequency, and phase stability is why the sweep holds the bunch together. The classical section keeps what belongs to it — the ±90° phase budget and why builders chase dee voltage. Figures renumbered to stay sequential (weak focusing is now Figure 2, phase stability Figure 3); section anchors unchanged. From the same external review as the additions below.
- — Three additions out of an external review. The builds census gains a short synthesis section, “What the census teaches”: how long first beam takes on the machines that date both ends, that a scavenged magnet saves money and months without deciding the outcome, that documented energies are keV rather than MeV once the famous larger claims are examined, and that the commonest recorded ending is silence — each figure computed from the entries at build time, with the sample’s censoring stated first. “How a cyclotron works” gains a checkpoint paragraph after the dees section, marking the point where the complete classical machine is on the table before phase, focusing, and the relativistic limit complicate it. And the locations page now leads with its most human feature: the visiting section is retitled “Cyclotrons you can visit”, and the page introduction says how many machines run public tours and points there.
- — The builds census and the world map now agree, and a check keeps them agreeing. The census holds 19 documented amateur machines; the map dataset carried 17, because two builders were each recorded once for two successive machines. Fred Niell built two cyclotrons. The 1994 science-fair winner ran on residual air with a fixed 1–2.5 MHz oscillator and could isolate no ion species; the machine he built the following school year replaced that oscillator with a signal generator feeding a tube amplifier across 0.5–16 MHz, which is what let him map resonance peaks for hydrogen and helium and match prediction to about 2%. The Rutgers 9-inch prototype, which produced its first proton beam during Hurricane Floyd in September 1999, had been folded into the 12-inch machine’s record. Both now have their own entries on the map, written from the builders’ own accounts rather than the secondary ones. The 9-inch is listed under the United States with no marker, because the project moved into the Rutgers physics building only in 2001 and no source says where it ran before that. Two errors came out with the conflation: the surviving Niell record described a four-inch proton machine, a description that comes from a conference paper which also dates his grand prize five years late, and the builder’s own write-up gives no diameter and says the beam was nitrogen out of residual air. Census entries now name their map record, the site build fails if that record has been renamed or dropped upstream, and the other direction — a machine on the map with no census entry — is reported by a script rather than by memory. Directory total 1,257 machines.
- — Note erratum, <a href="/design-guide/#dg-581">dg-581</a>, second correction to the same editorial note. The 2026-08-22 fix removed the “non-issue” absolute but left the exception clause imprecise: it said the neutron channels “open far lower” when the beam sees lithium, beryllium, boron or a deuterated material. Checked against NNDC (retrieved 2026-08-23), every one of those (p,n) thresholds is above 1 MeV as well — ⁷Li 1.88, ⁹Be 2.06, ¹¹B 3.02, deuterium 3.34 MeV — so they are closed to the reference machine’s protons too. The genuinely thresholdless neutron route is a deuteron beam: D–D and Be(d,n) are exoenergetic, which is what <a href="/design-guide/#dg-1047">dg-1047</a> has said all along. The note now states which side of that line each case sits on. Raised by the corpus maintainer’s upstream sweep of the same failure class; the safety page’s neighbouring threshold claim (¹¹⁵In(p,n) at 287 keV) was re-verified and stands. The same sweep tightened one sentence on <a href="/learn/choosing-your-machine/">choosing your machine</a>: “deuterium anywhere in the machine” is now “deuterium in the beam”, which is the side of the machine the thresholdless channels live on — deuterium sitting in a target is closed to sub-MeV protons like everything else.
- — Builds census: the Dawson College entry is corrected and completed from a first-party account. A project co-author, replying to this site’s inquiry through the college foundation, reports that the project — BICEPs, the “Bi-Institutional Cyclotron for Experimental Physics”, run by the roughly ten-student “Engineering a Mini Cyclotron Crew” — began in fall 2023 and wound down in 2025 as its members moved on to university, so it is no longer listed as in progress. The entry gains its design point (300 mm dees, 1 T, protons and H₂⁺), its published research — a betatron-motion and radiation-shielding study presented at the Canadian Undergraduate Physics Conference in 2024 and published in the college research journal in 2025, now cited and submitted for preservation — and its outcome: no construction milestones were published, and the recovered laboratory electromagnet now sits unused in a McGill University physics lab. The map record moves with it.
- — Locations: the map dataset sync brings two upstream changes. The Takanashi home cyclotron now appears in the table — under Japan only, with no marker, city or coordinates, because the builder has published extensively about the machine but not about where they live, and the database’s privacy rule declines to assert a precision it is not entitled to. And the Oyama KOSEN palm-sized cyclotron is recategorised from operating to under construction: across the amateur and teaching slice, the database’s working discriminator is a confirmed beam, and the AxeLatoon collaboration itself still lists first unambiguous ion-beam observation as pending. The change reverses a judgement this site had argued for, and the upstream record documents both sides.
- — The registry directory carries archived copies too, and the preservation check now runs monthly. Registries rot faster than papers do — government portals reorganise, agencies rename, national lists move behind logins — so the sixty registry entries that have a verified snapshot now show it beside the live link, the same dated fallback the census and library gained. The check itself is now a standing monthly cycle rather than something remembered occasionally: measure every cited source, report anything that has stopped answering, submit whatever has no snapshot, and verify those captures on the following month’s run, which is the margin the archive’s indexing delay needs. It refuses to publish a result that would drop a snapshot recorded the month before, on the principle that the archive does not lose captures but a bad network run can look exactly as if it had. Facility links in the map dataset are deliberately still outside the cycle: those are universities, hospitals and national laboratories, which are durable and replaceable in a way that an amateur builder’s personal site is not.
- — Every citation on the census and the library now carries its archived copy. The sources this site leans on hardest are the least durable ones — personal build logs, a university department page last touched in the Netscape era, maker-faire listings, a defunct WordPress blog — and until now a reader meeting a dead link had nowhere to go. Each source that has a verified Internet Archive snapshot now shows it beside the live link, dated, so a stale fallback is visible as stale. Where there is no snapshot, nothing is shown: an archive link that 404s is worse than none. Building it turned up a measurement error in how this site had been checking its own preservation. The Wayback Machine stores the address a link finally resolves to, not the address as written, so any citation that redirects — a trailing slash, http to https, a publisher appending .html — read as unarchived when the copy was sitting there under the other form. Checking both forms, 21 of 192 cited sources, eleven percent, turn out to have been counted as gaps that were not. Of the genuine remainder, eight cannot be captured at all: paywalled publishers, government portals that refuse crawlers, and one database query endpoint. The rest are preserved. Also fixed: two sources added yesterday were cited by unstable addresses — a conference paper by a file hash that no archive held, and a thesis by a repository API path that moves between software versions — and now use the published DOI-backed and catalogue forms instead.
- — Builds census: the Knox 2001 row is filled, and two dates on COLUMBUS were wrong. Knox had been in the census with no design at all — neither of its two sources described the machine. Two more do: the college physics department’s own projects page gives the design point (1.5 MeV protons on the NMR lab’s 2 T magnet, using that lab’s RF sources and amplifiers), and §1.2.5 of a 2006 Houghton College thesis surveys the machine in detail and states why it never ran — the magnetic field moved the ion-source supply wires until they shorted the dees. Its outcome is therefore no longer undocumented but stalled, which is a different claim: the record says there was no beam, rather than failing to say. For COLUMBUS, the builders’ own 2022 paper dates first beam to April 2014 and the project to 2012; the census had 2013 and 2010, neither sourced, and a conference year is not a beam year. That paper also documents the machine as it now stands — a 200 mm chamber, two selectable frequencies, a base pressure an order of magnitude below the 2013 figure — and it revises a rationale this page had reported: the team who gave keeping the beam inside as their reason not to extract began building an extraction system in 2021. Two claims here fell with it. The page said Maryland was the only machine in the census designed to extract; Knox built an extractor around a deflection plate, and one dee on The Cyclotron Project carries an unused extraction slit, so extraction hardware turns out to be designed more often than tested. Japanese entries gained sources rather than numbers: two AxeLatoon papers, and four programme posts that place the home cyclotron and the home linac in front of students in late 2021 — presented, the record is careful to say, not run. Toyota’s vacuum work is now in the programmes section, cooking pot and bathroom caulk included. Raised by an external note; three of its claims did not survive checking and are recorded with the rest.
- — Dataset erratum, <a href="/design-guide/#dg-460">dg-460</a>: the rule had repeated its source’s conclusion that ~2×10⁻³ torr is adequate for a fast machine. The vacuum deep dive traced that figure to a 5 km mean free path read from the thesis’s own plot, which implies a cross-section four orders of magnitude below the measured proton capture value (ORNL-6086); at that pressure the real capture mean free path is shorter than one turn, and the thesis machine reported no beam. The criterion (mean free path ten times the flight path) stands; the number is withdrawn with a dated note, the quote untouched. Raised by the owner. The beam-quality page’s citation of the same rule now carries the caveat.
- — New reference page: <a href="/learn/vacuum/">The Vacuum Budget of a Cyclotron</a>. The pressure target derived from charge-exchange survival along the spiral rather than a rule of thumb; the ion source as the dominant gas load; effective pumping speed, conductance, and the pressure inside the dee; diffusion, turbo and scroll trade-offs for a small machine; voltage holding, leaks, and the operating records; and a worked budget at the reference point. The charge-exchange cross-sections come from ORNL-6086 (Barnett 1990), read from the OSTI scan, and they replace the calculator’s earlier four-point table, whose 100 keV and 300 keV entries were high by roughly 2× and 3×. One result worth stating: at 1 kV on the dee a proton beam survives 10⁻⁵ torr at 90% but an H₂⁺ beam at 71%, because the molecular ion’s destruction cross-section never falls within a tabletop machine’s energy range. The <a href="/calculators/vacuum/">vacuum calculator</a> gained an along-the-orbit mode that integrates σ(E) turn by turn for H⁺ and H₂⁺, validated against an independent computation; the single-σ form remains as simple mode. The census does not record vacuum as the reason any build stopped, and the page says so; what it records is vacuum as the open item on builds that have not reached beam, with the professional operating records putting it at the top of the maintenance ledger. Glossary O-ring entry corrected to the realistic 10⁻⁶ torr elastomer floor (it said 10⁻⁵ Pa).
- — Classics: method versus tooling, across the shelf. After the TID-5215 note, every classics reading guide was checked for the same pattern — a construction-era practice endorsed as the method without saying that the tooling has moved. Eleven pages gained a sentence or a paragraph: the model-magnet sequence in NYO-780 and the analogue machine in ORNL-3540 (a field solver and an orbit code now, dg-080, dg-142, dg-817); the fractional-scale RF models in TID-454 and UCRL-64 and the grid-dip oscillator in UCRL-3187 (simulation plus a network analyser, dg-306, dg-236); the tetrode retrofit in the RF-power cluster (solid-state at tabletop powers); the film-and-densitometer energy measurement in the targets cluster (a silicon detector, or radiochromic film); the regenerator hand procedure in the extraction cluster; the diffusion-pump faults in the operations records; the carbon-disc neutron detector in AECD-2149; and UCRL-476’s claim that its magnet and vacuum content “does not age”, now scoped to the formulas and thresholds. The collection index carries the general rule. Raised by the owner.
- — Classics, TID-5215: a paragraph on reading the model-magnet chapter today. The page praised a 1949 method — build a scale model before the full-size iron — without saying that a builder now gets the same picture from a field code first (FEMM, for nothing, in an afternoon). Added the modern sequence from dg-080 and the point that a simulation and a scale model fail in the same places (dg-811), which is why the chapter’s substance still matters. Raised by the owner.
- — Design Guide: the remaining twenty-nine audit findings, read against the new label. With the editorial note now labelled as extrapolation, each finding was asked one question: does the note carry a <em>method</em> to a tabletop machine, which is what the label promises, or a <em>number</em> or a <em>claim</em> the source never made? Twenty-six were the latter and are corrected with dated notes, quotes untouched; three were declined as method extrapolation the label already covers, with the reason recorded. Seven of the corrections were errors in the rule text itself, which no label touches: a worked pole-profile example whose geometry did not check, a claim that two points fit square-root scaling when they differ from it by a third, a rule that said stored energy was enough to pit electrodes where its own quote reports no pitting, a welding rule that made one laboratory’s practice universal, and a neutralization procedure the source never gave and that could exceed tube ratings. One fidelity finding was settled by reading the hosted scan rather than arguing: two of the three facts the rule was accused of inventing are on the cited page, just outside the quote, and are now quoted; the third was not, and is withdrawn.
- — Design Guide: the editorial note is now labelled as what it is. Every rule card has three layers that carry different weight — the quote is the source verbatim, the rule is this site’s paraphrase, and the note beneath is this site’s extrapolation to a tabletop-class machine, which the source never addressed. The note was headed “Tabletop:” and sat under the same disclosure as the quote, and the methodology page said every rule was a source extract — true of the rule and the quote, never of the note. It now reads “Editorial note, tabletop extrapolation” on all 1,374 cards, with a hover explanation; both Design Guide introductions name the three layers; and the status vocabulary says which layer it does not cover. The change was prompted by the first monthly audit, which found two fidelity findings in fifty rules and nineteen for overreach: the extraction is sound, and the layer this site writes itself is where the errors live. Labelling it is not a substitute for auditing it, and the audit continues.
- — Design Guide: ten corrections from the first monthly back-catalogue audit. Fifty rules drawn at random were scored by an external referee on fidelity, physics, overreach and safety conservatism: 12 failed, 28 were marked for revision, 10 passed. Fidelity to source was sound throughout — two findings in fifty. The damage sits almost entirely in the applicability notes, the editorial layer this site writes around each extract, and the ten high-severity or safety findings are corrected today with dated notes and the source quotes untouched. Among them: a note that called a logged service-mode jumper equivalent to a two-key radiation-safety-officer interlock bypass (it is not, and the rule now says what the second key actually buys); a note that presented an Army manual’s self-described unreliable X-ray source-term assumption as a bounding figure, which inverted the source’s meaning; a third recurrence of the “below nuclear thresholds” absolute already corrected twice this month, this time in a rule about activation measurements; one measurement of “no H₂⁺ observed” turned into a proton-purity recipe; and an RF power rule whose formula, run with its own numbers, gave either 81 kW or 323 kW depending on an unstated voltage convention rather than the 96 kW it quoted. The remaining twenty-eight revision findings await an editorial decision on whether they are defects or an artefact of auditing notes that are extrapolation by design; the dispositions file records each one.
- — Changed pages are now announced to search engines directly. The site implements IndexNow, an open protocol supported by Bing, Yandex, Seznam and Naver: after each deploy the URLs whose content actually changed are submitted, and appear in those indexes in minutes rather than whenever a crawler next visits. Only genuinely changed pages are sent — the daily rebuild that refreshes the news feeds submits nothing, because a site that re-announces every page every morning teaches search engines to disregard it.
- — Build now verifies its own canonical tags. The site answers on three hostnames — the apex, <code>www</code>, and the Cloudflare <code>pages.dev</code> deployment domain — and the canonical link is the only thing telling a search engine those are the same page rather than three copies of it. All 147 pages were checked and every one is correct, including as served over <code>www</code>. Because that guarantee is load-bearing and invisible, the postbuild check (renamed <code>scripts/check-indexing.mjs</code>) now fails the build if any page loses its canonical or points it anywhere but its own apex URL. A redirect from the alternate hosts would be better still; neither is currently indexed, and neither can be redirected from inside this repository — the reasons are recorded in the project spec.
- — Sitemap: a submitted-noindex error fixed, and honest change dates added. The aggregated <a href="/news/">news list</a> is deliberately noindex, and it was also being submitted in the sitemap — a pairing Search Console reports as the error “Submitted URL marked ‘noindex’”, and one that had been there since launch. It is now excluded. The sitemap also shipped 146 URLs with no <code>lastmod</code> on any of them, which leaves a crawler nothing to prioritise by. Each URL now carries the date of the newest commit among the files that actually produce it — its page source, plus the dataset behind it for the data-driven sections — so a corpus sync moves the dates of the pages it changed and leaves the rest alone. Build time was rejected as the value: a daily rebuild would stamp every page as changed every day, which is both false and self-defeating. Where a page has no git history the field is omitted rather than invented. A postbuild check (<code>scripts/check-sitemap.mjs</code>) now fails the build if a noindex page ever reappears in the sitemap.
- — Page descriptions now say what a page lets you do, not only what it is about. A comparison of cyclotron teaching sites, written by a language model working from web search, sent readers to an external HTML5 simulator “for intuition” and placed this site as the one to graduate to later — while /learn/how-a-cyclotron-works/ has carried a live animated cyclotron, with the magnetic field and dee voltage on sliders, since it was written. The review was describing the pre-cutover site: the search index still serves the old WordPress pages, including four “design tools” that were empty placeholder pages and never existed. Nothing on the site was wrong, but its own metadata never mentioned that the seven calculators run in the browser and redraw a scale diagram as you type, that the state-law page is a tile map, or that the Design Guide filters client-side. Those descriptions are rewritten. The theory page’s opening paragraph now points at its own working model instead of sending the reader to the calculators for interactivity.
- — Editorial methodology, corrected, and the corpus counts made self-updating. The page that describes how this site works still said the Library was an annotated index of 201 documents, and that no documents are hosted here. Neither had been true for weeks. The Library indexes 254 documents, 51 of which are public-domain reports the site hosts itself, each with a written rights rationale on its own page. An external reader found it, and the cause is unglamorous: counts typed into prose fall behind the datasets they describe. Every corpus count in site prose is now computed from the published data — the rule total on the home page, the 404 page and the Design Guide description, and the library, hosted-document and source-document totals on the methodology page. That last one was reported as 86 source documents and is 136. The methodology page also now describes the standing external rule audit and the dated-erratum convention, and carries a Provenance strip of its own so its own freshness is visible.
- — Design Guide: the n = 0.2 rules reconciled. Three rules stated the Walkinshaw coupling resonance in absolute terms their sources do not support. dg-136 said the beam “is lost” if n = 0.2 falls inside the orbit region, where the thesis it quotes says the value “must not occur” there, which is a design rule and not a prediction. dg-152 said a beam crossing the point “blows up axially”. Elsewhere the same guide advises crossing the resonance in as few turns as possible, so it contradicted itself, which is what an external review pointed out. All three now describe the resonance as what it is: a driven coupling whose severity depends on the field imperfections exciting it and on how many turns the ions spend within its width, which is why a few kilovolts on the dees is the named worst case. dg-145 is faithful to its source and its text is unchanged, but its applicability note read as an instruction to shim toward n = 1 without mentioning the contour at 0.2; the cross-references are now there. Each correction carries a dated note and leaves its source quote untouched. The beam-dynamics deep dive gains a paragraph stating the hierarchy outright: 0 < n < 1 is the stability condition, the resonance is a rate rather than a threshold, and clearing it is necessary rather than sufficient.
- — Safety: two absolutes removed from the Design Guide, and one from the Safety page. Two rule notes said “there is no activation to survey” and “No activation at tabletop energies”. The site’s own safety and shielding pages explain why statements of that shape are unsafe: 7Li and 11B have (p,α) channels that run well below 100 keV, and the (p,γ) captures have no threshold at all. That is the correction the guide already carried at dg-1041, made in August, and these two notes had not caught up with it. Both are now scoped to what they can support, the activation of ordinary copper, steel and aluminium structure, and both point at the target-side exception. The Safety page’s own summary sentence, that below ~2 MeV activation is a non-issue, is replaced for the same reason: it collapsed an isotope dependence the page had just spent three paragraphs explaining.
- — Locations: visiting moves up. The page was map, then the full 1,254-row table, then the virtual tours — which put the tours some seventy thousand pixels down, below a table nobody scrolls through. Now a “Visiting and virtual tours” card sits directly under the map: the in-person side in one line (31 machines offer public tours and 67 more by arrangement, with a link that applies the public-tours filter and shows them on the map) and the eight verified online tours beneath. The A–Z table becomes a collapsed “Full directory” at the bottom, open by default for readers without JavaScript, for whom it is the directory, and folded for everyone else, for whom the map and filters are. A jump line under the intro names the page’s three parts.
- — Library and Design Guide navigation. The Library’s 254 entries are now grouped into four sections by era and kind — classic hands-on (68), timeless reference (36), transitional (33), modern (117) — each with a paragraph saying what the class is and why it scores as it does, and an on-this-page rail. The search box, relevance filter and topic chips still work across every section; a section with no match collapses to a one-line note rather than vanishing. On the Design Guide, the directory of the 22 per-subsystem pages moves above the schematic and becomes a panel of labelled tiles (“Browse by subsystem”) so it reads as navigation, while the filter bar is labelled as what it is: a filter for the all-in-one page. The filter chips had been styled more prominently than the links, which is backwards. Domain labels and descriptions now live in one shared module used by both views.
- — History: the beam in air. Donald Cooksey’s colour photograph of 5 October 1939 — a 16.5 MeV deuteron beam leaving the 60-inch’s target chamber and ionising five feet of open air — joins the 1933–1939 section, with Cooksey’s own notation on the print quoted (resonance 17 April, proton beam 17 May, deuteron beam 7 June). Sourced from the National Archives record marked unrestricted; the same photograph exists under two other NARA records whose digital files carry a University of California commercial-use reservation, which is recorded in the page’s design notes as a trap for future image work.
- — History: who thought of it first, and a style pass. The 1929–1931 section now records that Lawrence was not the first to conceive a circular resonance accelerator — Craddock’s history lists Gabor (1924), Flegler (1926) and Steenbeck (1927), and Leo Szilard filed a German patent application for one on 5 January 1929, about six months before Lawrence; it was never granted, never published, and abandoned, and the one early attempt elsewhere, Thibaud’s in Paris, followed Lawrence’s publication and failed. The verdict is quoted from Craddock rather than asserted: credit lies in showing that it works. The page’s prose was then read against STYLE.md, which it had not been: sixty-nine em-dashes became eight, sentence lengths were varied, and the set-piece constructions the guide rations were removed.
- — History: the original first-cyclotron chamber. The page now opens on a photograph of the actual brass-and-sealing-wax chamber Livingston built in 1930–31, held by the Smithsonian National Museum of American History (object EM.N-09621). It is the one image on the site that is not public domain — the Smithsonian marks the photograph “usage conditions apply” even though the catalogue record is CC0 — and it is used under the museum’s non-commercial educational terms, with that stated in the caption and recorded in the page’s design notes as a single written exception, on the same model as the library’s one non-bit-exact document. The 1958 replica photograph stays, now captioned as a replica of the chamber above it.
- — History of the Cyclotron, rebuilt as the front door to the site’s own primary sources. The page gains its first photographs — the 1930–31 Livingston chamber (a 1958 replica, in its presentation case), sheet 1 of Lawrence’s 1934 patent, Livingston and Lawrence at the 27-inch in 1934, the Crocker 60-inch in August 1939 with its external beam line, and the 184-inch’s dee tank being lowered between the poles — plus the 1952 cutaway drawing of the Oak Ridge 86-inch installation taken from ORNL-1196 in the library. Every image is public domain and its caption says why: US government photographs via the National Archives, a US patent drawing, and a figure from a hosted public-domain report. The Livingston-style energy chart is corrected (the 27-inch is now plotted at its 1933 first operation rather than at the year of its best energy, and the caption states that the energy is the best documented figure for each machine, not the first-beam figure, and that the two medical entries sit at nominal years) and gains a second series: the amateur census drawn on the same axes, which makes the page’s closing argument visible — the frontier climbs four orders of magnitude and the amateur line never leaves the band the 4-inch and 11-inch occupied in 1931–32. New sections cover the destruction of Japan’s cyclotrons in November 1945, H⁻ stripping extraction as the development that made the compact commercial machine possible, the 1958 Oak Ridge world directory (140 machines) against the map today, the amateur line 1947–present, and a list of preserved early machines computed from the world map dataset. Sources are now linked where freely available, the text links to the library documents it draws on, and the page carries a Provenance strip.
- — Builds census gains four programme sections — the Koeth lineage at Rutgers and Maryland, Houghton College, COLUMBUS at Coburg, and Japan’s AxeLatoon — because six of the nineteen entries are machines belonging to sustained programmes rather than standalone builds, and four of those six reached beam against three of the thirteen that stand alone. Each section covers how the programme keeps going rather than restating specifications: the annual student project, the thesis record, institutional migration, and a collaboration that names generational turnover as its structural problem. An external peer review of the page produced seventeen findings, all dispositioned. The most consequential: the page asserted a builder’s safety reasoning in its own voice, that keeping the beam inside the chamber means no radiation leaves — which the page’s own Rutgers entry disproves, since an internal target there produces neutrons with no beam leaving either. That claim is now attributed rather than endorsed. Three further findings were one fault: the table paired design fields with bench-measured frequencies, so the two columns did not satisfy the cyclotron relation. RF values now carry the same tier markers the field values do, including a new marker for a bench measurement of a subsystem, and the table states that the two columns are not always a matched pair.
- — Builds census: two Japanese entries, and the programme behind them. Takaoki Takanashi, a theoretical particle physicist with no experimental training, built a tabletop cyclotron in his apartment living room and exhibited it at Maker Faire Tokyo in 2019 — buying a lathe and a milling machine, sourcing RF gear secondhand, and being turned away by machine shops that doubted an individual would build such a thing. No field, frequency, dee voltage, pressure, beam energy or current has been published for it, so its outcome is recorded as undocumented, the same standard applied to the 1947 El Cerrito machine. The second entry is the Oyama KOSEN cyclotron, operating since January 2023: twelve first-year students simulated the field and particle orbits in CST Studio before designing the chamber, and their detector now registers particles as a current, though the team is explicit that first unambiguous ion-beam observation is still pending. Both sit inside AxeLatoon, a programme at eight Japanese technical colleges in which students build cyclotrons mentored by accelerator scientists from KEK, RIKEN and several universities — and which chose the cyclotron as its target machine precisely because a collaboration member had already built one at home. The census introduction now describes the programme, including the sites that are assembling, the one that stalled when its members hit job-hunting, and the one whose machine is a linac and therefore out of scope. AxeLatoon is also in contact with the COLUMBUS team in Coburg, with a Japanese translation of their book in preparation.
- — Correction: the Rutgers 12-inch cyclotron produces D–D fusion neutrons, and an entry published earlier today wrongly called Houghton’s 2022 deuteron run the census’s only documented nuclear reaction. The Rutgers machine has run deuterons for longer and at higher rate — tuned to 7.150 MHz at 0.96 T average with spiral azimuthally-varying-field pole tips, driving a deuterated internal target on a radial feedthrough at a nominal 100 keV incident energy, for about 500,000 neutrons per second during a five-second continuous run, measured with a ³He detector calibrated against a NIST-traceable ²⁵²Cf source. Its builders have also pulsed the machine down to a 10 µs beam-on-target window to measure neutron diffusion time in a polyethylene moderator, and attempted the ¹⁹F(p,αγ)¹⁶O reaction with fluorine internal targets. The entry, its specs and the comparison table now carry the deuteron operating point alongside the proton design figures, and the table’s introduction says two machines have driven nuclear reactions rather than one.
- — Builds census, corrections from the source re-read: the Cyclotron Kids’ 1.6 T is labelled a design target, since lab safety rules for minors meant the magnet was never energised, and the project is redated to c. 2006 when the students conceived it (construction began 2009). Niell I’s 2.2–2.3 MHz peak is identified as roughly a tenth-order harmonic of the nitrogen fundamental rather than the fundamental. Houghton’s entry gains its 2022 deuteron run and the detected D(d,n)³He neutron signal — the only documented nuclear reaction in the census — along with its beam currents and the note that its 170 mm chamber replaced a smaller brass one. The Knox College entry is trimmed to what its two linked sources actually support: the claims that it was fixed-frequency, an honours thesis, and completed in 2001 appear nowhere in them. The Rutgers 12-inch gains an explicit extraction line — the 2009 electrostatic deflection channel turns beam onto an internal phosphor screen and serves as a velocity filter — and both Rutgers entries now cite the two Cyclotrons2013 papers. El Cerrito’s Physics Today mention is attributed to the later account that makes it.
- — Builds census correction: the COLUMBUS cyclotron (Coburg) is still running. Hochschule Coburg’s applied vacuum technology laboratory describes it as fully functional in October 2023, used for undergraduate measurements and two to three school workshops a year, having been commissioned about ten years earlier — so the entry moves to 2010–present and to routinely operated, and joins the operating-today group. The 2016 trade-press citation, which had become unreachable, is replaced by three live sources plus the design reference its builders published: Prechtl and Wolf, Das Lehr-Zyklotron COLUMBUS (Springer essentials, 2020), cited rather than hosted. No measured beam energy appears in any published source, so the machine’s energy stays absent from the table and its 24–48 keV remains a design figure. Separately, the comparison table’s extraction note now names the Rutgers 12-inch’s electrostatic deflection channel — real extraction hardware, built 2009, but aimed at an internal phosphor screen and used as a velocity filter to measure beam energy, per the machine’s Cyclotrons2013 paper — rather than leaving the impression that no builder has attempted the geometry.
- — Builds census: a comparison table (/builds/#compare) puts all seventeen machines on one grid — start year, builders, magnet provenance, pole diameter, field, RF frequency, vacuum, particle, energy, beam current, years to first beam, and whether the machine produced beam and is running. Every cell was re-harvested from each entry’s linked sources with a citation per cell (hover a value for its page). Editorial rule: energy and current appear only for machines that produced beam, and only as the sources report them — an energy computed from field and radius is marked as such, a design field is marked as design. The harvest also corrected the entries it read: El Cerrito moves from operated to outcome unknown (the 1947 article reports the machine operating, not a beam); Houghton’s start is 2001, first beam 2007, and its currents are nanoamp-scale with a ≈100 nA peak rather than picoamp; Niell II’s field and energy are marked calculated, not measured; Rutgers 12-inch gains its measured 15.06 MHz resonance; COLUMBUS’s 24–48 keV badge is withdrawn as a design figure; the Mullins first beam is May 2018 per its log. The cards are grouped by documented outcome with a jump nav, and the timeline palette now separates beam from no-beam outcomes by hue.
- — Builds census: the Mullins Cyclotron entry is updated from its build log’s August 2026 status note — still operational, over 170 logged runs with the most recent in 2025, typical operation 1–2 nA at 580–590 mT and 8.9–9 MHz for about 150 keV protons at roughly 6 × 10⁻⁶ torr, and a hydrogen delivery system around a used 1 sccm mass-flow controller as the one significant upgrade. Status moves from beam-demonstrated to operated and the timeline bar now runs to the present; the ≈164 keV best-demonstrated figure is unchanged.
- — Figures on the per-domain Design Guide pages — where a figure is a map into the rules, and nowhere else. Three new: a vacuum pressure ladder with fifteen rules’ pressures ticked on a log axis (pump regimes, breakdown regime, gauge coverage, computed mean free paths); a safety onset ladder on electrode voltage and proton energy, every tick cited to a rule or to the live-fetched nuclear data on the experiments page, and no value offered as “safe below”; and the weak-focusing tune diagram for beam dynamics, which makes the hierarchy explicit that an external review found the rules stating inconsistently — 0 < n < 1 is the stability condition, n = 0.2 is a coupling resonance whose severity depends on imperfections and crossing rate, and a small many-turn machine should not linger there. One new iron-circuit map serves the magnet and coils pages with the numbers the rules agree on. Four figures the subsystem deep dives already carried (RF coupling, ion-source anatomy, extraction geometry, the suppressed Faraday cup) are now shared components, so each exists once. Every region of every figure is a link that opens the guide filtered to the rules on that part. Practice-list domains (fabrication, materials, seals, chamber, modeling, project management) deliberately get no figure.
- — Legal: the 51-jurisdiction survey now has one page per state and for the District of Columbia (/legal/states/tx/ and so on). Each page presents that jurisdiction’s record from the survey — accelerator definition, how it reads against a small closed-beam sub-MeV machine, registration or licensing mechanism, incidental-radiation exemption wording, citation, and notes — followed by the same three-step guidance and the same not-legal-advice notice the overview carries. The map tiles and the table’s first column on /legal/states/ link to these pages. Content is unchanged from the August 2026 verification; nothing was added beyond the existing survey data.
- — Design Guide published per domain. The guide’s 1,374 rules are now also served as 22 static pages, one per domain tag (/design-guide/rf/, /design-guide/magnet/, and so on), each with a one-sentence scope note, a related-domains line computed from tag co-occurrence, and its own provenance strip. Rule anchors (dg-NNNN) resolve on both the domain page and the all-in-one guide, whose filterable view and deep links (?domain=rf,dee) are unchanged; the all-in-one page gains a domain directory and is excluded from site search so results land on the smaller pages. The rule card is now a single shared component, so the two views cannot drift. Internal links across the site that pointed at single-domain filters now point at the static pages, and five stale rule counts in that link text were corrected.
- — Site plumbing from an external review: a descriptive homepage title, breadcrumb structured data on every page, dataset structured data for the four downloads on /data/, the aggregated news list marked noindex so it cannot compete with original pages, and the homepage’s project-assessment column now leading with Should You Build a Cyclotron? and including Choosing Your Machine.
- — New page: Choosing Your Machine (/learn/choosing-your-machine/), the decision guide between "should you build one" and the subsystem pages. It takes the two choices that shape everything downstream — purpose and energy band — and ends each of the four sub-MeV bands in a project brief: the design point (field, radius, frequency, computed from CODATA constants and validated against the energy calculator), the radiation footprint with its onsets, the closest documented builds from the census, and what to read before spending money. The pattern it argues from the record: the demonstrated amateur beams cluster below 200 keV, the one machine above that mark took eleven years, and the census’s megavolt entries are all intentions — so the honest move is picking a band on purpose rather than discovering it in year three. It also gives the copy-or-originate decision its own treatment, anchored on AECU-1951’s documented case of a university copying a working machine from borrowed drawings. Cross-linked from the learn landing, the starter path’s target-energy decision point, and the experiments-by-energy scorecard.
- — Design Guide 1,291 → 1,374 rules from the three documents hosted earlier today, which closes a question this site had been chasing for weeks. TID-5215’s third chapter reports scale models of every magnet type in a wartime production programme, and its fourth chapter checks them against the full-size machines that were then built: the real magnets came in slightly BETTER than the models predicted, on field uniformity, stray field and efficiency alike. That is the rare case where somebody published both halves of the comparison, and it is what makes building a small model before committing iron a defensible use of a builder’s time rather than an article of faith. The report we had been hunting for that answer has never been digitised; the answer was in a different volume all along. The dee-coupling deep dive also gains something it lacked: a documented disagreement. Oak Ridge argued for a separate oscillator driving an amplifier, because a self-excited driver makes harmonic amplitude and phase interdependent on a multi-frequency machine; Berkeley ran the resonator as the frequency-determining element on the 88-inch and bought back the alternative’s advantages one at a time. Both machines worked, and the page now presents both positions instead of averaging them — the small-machine recommendation is unchanged, but it is a recommendation with a counterargument behind it rather than a consensus.
- — Five more documents hosted, taking the collection to 55 across 26 pages. All five came out of a library holding that four earlier searches had reported empty, because the catalogue had been queried in a metadata format that does not carry report numbers. Two are new pages. TID-5215, Magnets and Magnetic Measuring Techniques (Berkeley, 1949), is a 226-page volume about magnets and nothing else — design requirements, measuring instruments, model magnets, and tests on full-scale magnets — and its third chapter answers a question this site had been chasing through undigitised reports for weeks: how much a small model magnet can tell you about the full-size one. AECU-1951 is the University of Washington 60-inch, and it is a construction record, the scarcest class here: buildings, magnet, vacuum and gas supply as one job. Its most useful passage is not a number but a decision — the machine follows Berkeley’s Crocker cyclotron, from a complete set of plans supplied by E. O. Lawrence, which is a documented case of a physics department copying a machine known to work rather than originating a design. Half its concrete building is buried, which is the cheapest shielding there is. The other three joined existing clusters: the Sparker and a hooded arc ion source join the Rochester 27-inch, which now covers both halves of one machine’s life rather than only what came after first beam, and a self-excited oscillator study joins RF power engineering. Two of the five exceed the host’s 25 MB per-file limit and are served whole and unrecompressed from files.cyclotroninfo.com. One warning is on AECU-1951’s page rather than buried: its OCR is the worst in the collection, and its text layer is a search aid only.
- — Design Guide 1,272 → 1,291 rules from two short Rochester reports recovered from a library holding that four previous searches had reported empty. The reason they were missed is worth stating: the catalogue was queried in a metadata format that does not carry report numbers. Requested in the format that does, the same collection produced five documents previously written off as lost. The find of the pair is a free diagnostic for a failure every builder eventually meets — RF drive goes up and dee voltage does not, because a low-voltage electron discharge absorbs everything the amplifier delivers. Ring the dee with the power off, watch the decay on a scope, and look for a kink: the envelope falls smoothly, drops steeply through the loading band, then decays slowly again below it. That locates the band on your own machine instead of inferring it from someone else’s, and answers whether your operating voltage sits inside it. It also refines a 1946 figure this site already published — the discharge was known to exist below about 500 volts, and this gives the band an observable top edge as well. The dee-coupling deep dive gains both that test and a third cure to go with the two it already carried: Rochester rejected dee biasing and a tickler oscillator as too awkward on a variable-frequency machine and built a spark gap that kicks the dee straight through the band, on the insight that the kick need only clear the top of it, not deliver operating power.
- — Design Guide 1,157 → 1,272 rules, library 245 → 249, from four volumes of conference proceedings that had been freely available for decades without ever surfacing in a search. The targets domain goes from 12 rules to 127 — the thinnest substantive subject on this site becomes one of its deepest, in a single pass. The International Nuclear Target Development Society published its proceedings as national-laboratory reports, which is why they never answered to their own subject name; the society was also called something else before 1977, so the founding volume, which holds the best survey paper, is invisible to a search for the name it uses today. What the volumes add is the half of target-making that evaporation recipes cannot cover: rolling, sputtering, electrodeposition, chemical vapour deposition, cooling, and Corwin’s calculation of how much beam a foil takes before it fails — which turns out to be usable at sub-MeV energies, so it now sets a target current limit that nothing here previously bounded. All four are cited rather than hosted: they are international conferences, and the pre-1978 argument that clears an unmarked US report does not clear its foreign-authored papers, because the URAA restored US copyright in foreign works that had lost it to a missing notice. One volume additionally reprints a journal paper under permission granted to Argonne. Every quotation is held to 25 words and attributed to its individual paper and author. The classics page on targets, which said the thick-and-durable half remained uncovered, now points to where it is covered.
- — Correction — the shielding deep dive and this changelog said Moyer measured a paraffin transition increase of 100% following CONCRETE. The report says lead: “Paraffin yields a transition increase of 60% following Fe, and of 100% following Pb with similar geometry.” The Design Guide rule was right and the error was introduced when the rule was paraphrased into page prose. It matters more than a slipped word, because lead is exactly the material someone is most likely to wrap paraffin around, so the warning had been pointing at the wrong pairing. Found while verifying figures against page images before hosting the document.
- — Four documents recovered from the Defense Technical Information Center’s mirror are now hosted, taking the collection to 50 documents across 24 pages. Two are AEC reports declassified in 1947–48 and circulating under military accession numbers, which is a route into 1940s laboratory literature this collection had not used. AECD-2149 (Moyer, Hildebrand, Knable, Parmley & York) gets its own page: six pages, and the first source here whose shielding numbers were MEASURED rather than derived — half-value thicknesses for six materials taken with the cyclotron as the neutron source, with two detectors that disagree, which is the useful part. Its dose map is stated per unit beam current, so a reader can scale it. AECD-1850 joins the “knowing before building” cluster, and completes a pair that was already half here: MacKenzie proposed the “three quarter wave” RF system in September 1947, and Anderson reported the half-scale model tests of that same system three months later — the collection now holds a proposal and its verification. A new operations-records page carries two accounts of machines being run rather than built: the Naval Research Laboratory accounting for every hour of six months (60.1% operating, with power supplies and vacuum together consuming twelve times more time than the RF), and a Harvard quarterly that opens on four days lost to a cold-trap refrigerator and low diffusion-pump oil. Rights: the two AEC reports on the declassified-and-issued basis used for the MDDC memoranda; the NRL report as a work of a United States Government laboratory, approved for public release and sale; and the Harvard quarterly on pre-1978 publication without a copyright notice, evidenced by the Defense Documentation Center’s own 1964 sale stamp rather than by inference. Every figure quoted from the two typescript scans was read from the page image, because their text layers render 831.3 as “851.3” and 1964 as “196*+”.
- — Design Guide 1,114 → 1,157 rules, library 241 → 245, from four documents recovered from the Defense Technical Information Center’s mirror at the Internet Archive. Two of them turned out to be AEC declassified reports circulating under military accession numbers, which is a route into 1940s laboratory literature that this collection had not used before. All four are cited rather than hosted for now, each linked to the public copy. The find is Moyer et al., Character of the Radiation Field and Shielding at the 184-Inch Cyclotron (AECD-2149, 1947): the shielding domain had methodology but no MEASURED attenuation behind it, and this is concrete and water measured with the machine itself as the source. B. J. Moyer later gave his name to the Moyer model, the standard analytic method for accelerator shield design. One of its results is now on the shielding deep dive because it inverts the obvious intuition: adding a hydrogenous layer AFTER a dense one raised the reading behind the shield rather than lowering it, by about 60% for paraffin following iron and 100% following lead — so layer order is a design variable and a composite shield cannot be sized by summing its layers. Also added: a 1947 report on measuring a radio-frequency system at three-quarter scale before building it, and two accelerator operations records, one of which keeps an availability ledger showing power supplies and vacuum causing roughly six times the downtime of the RF. Shielding is now 50 rules, from one at the start of August.
- — Design Guide grown from 1,032 to 1,114 rules and the library from 238 to 241 entries, by mining the three documents acquired today. The additions land almost entirely in the two domains that were thinnest: — shielding goes from 12 rules to 40 and safety from 67 to 125, so shielding stops being a stub. Fifty-six of the eighty-two come from the Army radiation-protection manual (AD-755510) and are attributed to their individual chapter authors rather than to its editor; its 1972 exposure limits are carried as historical figures, with the method rather than the numbers as the transferable part. The rest come from two documents this site cites but does not host, because the rights determinations went against hosting while leaving the citations perfectly sound: the Columbia Nevis synchrocyclotron modification report (21 rules, quoted briefly and linked to INSPIRE’s copy) and a six-page Oak Ridge health-physics overview (5 rules, linked to OSTI). The classics page for AD-755510 now carries its rule count and a representative sample instead of saying the document is unmined. One limit is now stated on the shielding deep dive, because it decides whether the manual helps you at all: its X-ray shielding curves run from 1.0 to 40 MeV, so a machine with a dee at a few kilovolts sits off the bottom of every chart in that chapter. What it offers instead is a way to bound the source — assume the ion current drives a reverse electron current of about a fifth of it through roughly a third of the terminal voltage — and the report is candid that those are unreliable estimates.
- — Construction Classics: Accelerator Radiation Protection (AD-755510 / TR-73-7, US Army Natick Laboratories, 1972) — 196 pages, and the first document in the collection hosted with pages removed. It closes the last documented gap in the reference collection: almost every shielding document here was written for a machine far larger than anything an amateur builds, while this one is scoped to accelerators below 40–50 MeV and names cyclotrons in its keywords. It also exists for an unusual reason, stated in its foreword — a 1969 request to help draft an Army regulation on accelerator radiation protection found no reference that combined the literature, so Natick’s staff wrote one. Eight separately authored chapters: X-ray production and shielding, neutron production and shielding, induced radioactivity, exposure limits, pulsed-beam measurement, access limitations and interlocks, waste effluents. Chapter VI is the least duplicated content in the collection — an instrument that reads correctly in a steady field under-reads in a pulsed one, and the chapter derives the true-versus-measured counting rate for paralyzable and non-paralyzable counters. Rights: a US Army in-house work, unclassified, approved for public release with distribution unlimited, no copyright notice in 196 pages — but the acknowledgments record that permission for McGraw-Hill and ASME material was obtained by the Army, and that permission does not transfer to a re-publisher. Locating the borrowed material required examining all 196 page images, because the only evidence is superscript reference numbers in figure captions, which the scan’s text layer drops. Five figures were identified, all standalone plates; this copy replaces those five pages with notices naming the figure, its source, and where to obtain the unabridged original, which stays freely available from the DTIC copy at the Internet Archive. No body text is affected and the pagination is unchanged. Three further McGraw-Hill citations were checked and are not reproductions.
- — New deep dive: Shielding a Small Cyclotron (/learn/shielding/). The learn path had eight deep dives and none on shielding, which was defensible when the Design Guide held a single shielding rule; it now holds twelve, and this page assembles them with the hosted sources behind them. Its argument is that shielding starts as a question about nuclear thresholds rather than materials: X-rays arrive early, from around 18–20 kV on electrodes, while the lowest practical proton-induced neutron door among light targets is ⁷Li(p,n) at 1.881 MeV — so a machine at a few hundred keV is an X-ray problem wearing a neutron problem’s reputation. Deuterium is the exception and changes the problem categorically. The method sections come from ORNL-3540’s chapter 11: build the estimate as a chain from dose limit through source term and attenuation to secondary buildup so a wrong answer can be localised; design from maximum-credible source terms before operating and verify by survey afterwards, rather than treating a measured spectrum as the design input; and treat penetrations as the real leak path, with duct transmission set by length-to-radius ratio and roughly 0.1 for a well-made 90° bend, multiplied element by element. A figure and the report’s own worked four-leg maze show that multiplication, with its scope stated: those factors are ORNL’s low-energy neutron example, not a general maze formula. The page is explicitly not a dose calculator and does not produce shield dimensions; every figure on it is a historical example quoted to show a method, the professional standard (NCRP 144) is named at the top as prerequisite reading, regulation and dose limits stay at /legal/, and hazard identification stays on the safety page. Nuclear thresholds were verified live against IAEA INIS and OSTI, NCRP 144 was confirmed current, and the draft went through two external review passes — a physics referee and an editorial one — whose findings are dispositioned in the page’s own record.
- — Correction — the residual-radiation page described Cr-51 and Mn-54 together as what distinguished stainless steel from plain iron. ORNL-3158 is more precise: stainless produces two long-lived isotopes, Cr-51 and Mn-54, where plain iron produces only the Mn-54, so the chromium is the distinguishing element and the nickel brings Co-58 with it. The correction sharpens the point rather than weakening it — the marker nuclide follows the alloying element, which is the actual argument for treating material choice as a radiological decision. Found while spot-checking citations for the new shielding page against the scan.
- — Design Guide grown from 925 to 1,032 rules and the library from 225 to 238 entries, by mining the thirteen reports hosted earlier today. Every hosted document in the collection again links to the rules extracted from it. The seven-report Oak Ridge seam was mined as one narrative rather than seven documents, and it produced what progress reports produce: resolve individual turns with a 0.020-inch tantalum wire probe, and expect a spurious rise in its reading with radius because the wire heats and emits thermionically; flatten the base field before testing shims; sequence RF design around measurement; and outsource castings and weldments at your peril. The Rochester trio yielded external-beamline architecture — condenser, then a beam-defining slit placed inside the shield wall because the slits intercept most of the beam, then the analyzer — plus the observation that a cyclotron needs no beam sweeper for time-of-flight work, since the beam arrives already bunched in RF phase. One correction to a page published this morning: the Oak Ridge crew aimed at ±0.01% field uniformity with a portable grinder, and the seam shows that after a further half-year of grinding they reached 0.05%, five times short. Both numbers are now on the page, because a single report would have printed only one of them. The commissioning-practice page now also notes that the 22-inch cyclotron it discusses was renamed the 44-inch in 1953, so a reader following that machine is not left with half its record.
- — Thirteen more public-domain reports hosted across three new pages, taking the collection to 46 documents. The find is a seam rather than a document: seven consecutive Oak Ridge division progress reports follow ONE small cyclotron from December 1951 to March 1955. A numbered footnote in the 1953 report — "formerly referred to as the 22-in cyclotron" — is what joins them, because the machine was renamed the 44-inch when its magnet was reworked, and by 1955 there was a proposal to convert it again to a 48-inch heavy-ion machine. Two further reports in the same run were already hosted on the commissioning page and are shown in sequence rather than duplicated. Progress reports print what final reports omit: a removable spacer that moves the dees back from the field centre so one machine runs at either 1.5 or 5 MeV, magnet pole faces ground with a portable grinder toward ±0.01% specifically to give shim designs a fair baseline, an RF circuit left undesigned because the resonator it feeds had not been measured yet, and — in the last report — the contractor’s brazing errors and the admission that electrical work progressed only as fast as funds allowed. A second new page collects three University of Rochester reports on what a physics department does with a 27-inch machine once it runs: external-beam optics, a spectrograph built around the cyclotron, and neutron time-of-flight. Two of those carry an explicit "publicly releasable" stamp. The third page pairs one question answered fourteen years apart — Berkeley testing a half-scale model of an RF system in 1947, and Michigan State computing extraction optics and analytic dee-field formulas in 1961. Ordering note: the Oak Ridge reports are sequenced by the period each covers, not by report number, because the two disagree.
- — ORNL-3540 mined: 36 more Design Guide rules, taking the guide to 925. The yield landed where that document is genuinely unusual rather than where a machine report normally pays out — eight of the thirty-six are shielding rules, in a guide that had exactly one before, and most of the rest are project estimating, scheduling, and trade-study discipline. Shielding is now a real domain here rather than a stub: how to build a shield estimate as an explicit chain from dose limit through source term and attenuation to secondary buildup, how to design mazes and penetrations by multiplying per-element transmissions, stepped labyrinth joints so ordinary construction tolerances are acceptable, and the instruction to assume the estimate is low and design the expansion room in. On the cost side: structure an estimate as basic cost plus explicit adders, and ground every line in several bids with the median taken rather than a single quote. This is the last hosted document to be mined — every one of the collection’s nineteen entries now links to the rules extracted from it.
- — Correction — two publication dates on the RF power engineering page were wrong and are now fixed. COO-535-543 was dated 1963, following its catalogue record; the report describes an amplifier tube change of May 1964 and its second accession number is UCOL-65-39, so it is dated 1965 here. ORNL-2403 was dated 10 January 1958; its own title page reads "Date Issued DEC 26 1957". Both errors came from trusting catalogue metadata over the scan, which is the opposite of this collection’s stated method, and both were caught during the extraction pass reading the documents closely. The cluster is now described as 1955–1965.
- — Design Guide grown from 709 to 889 rules, and the library from 206 to 224 entries, by mining the eighteen construction reports hosted since 12 August. Every hosted document that was previously marked "not yet mined" now carries the rules extracted from it, and each of those rules links back to the document and page it came from. The additions are concentrated where the collection was thinnest: beam extraction goes from 21 rules to 71, targetry and activation get their first substantial entries, and the largest single yield — 45 rules — comes from the seven-report TID-454 volume. Several pages were corrected rather than merely extended: the beam-extraction deep dive no longer says the mining pass added no extraction rules, because it added fifty, and two pages that called extraction the thinnest subject in this collection now say so about the open literature instead, which is where the claim is still true. Rule and library identifiers are permanent, so existing links are unaffected. Mining also resolved two figures this site had hedged: the RF-extraction gradient is 4.3 kV/cm against a stated ceiling below 5 kV/cm, and lead half-thickness for the measured 510–810 keV activation lines is about 0.6 cm.
- — Ten more public-domain reports hosted across four new pages, taking the collection to 33 documents. Beam extraction — the thinnest subject in the open literature — gains four Berkeley reports spanning 1948 to 1963 that answer the same question four ways: a sixteen-thyratron deflector pulser with a 0.1 µs rise time, regenerative extraction theory, RF-driven extraction, and a deflector supply designed to limit the energy of each spark rather than prevent it (the last being the direct companion to the already-hosted UCRL-10654). A second new page covers residual radiation and personnel dose, both measured rather than estimated: which nuclides a machine structure becomes, differing between plain iron and stainless steel, and five years of film-badge data from a 60-inch cyclotron crew — presented with an explicit scope note, because these are 730 MeV and 60-inch conditions and activation is a property of specific nuclides above specific thresholds, not of materials. A third covers the target plane: a complete bench recipe for self-supporting boron and magnesium films at tens of µg/cm², against a collection with almost nothing on targetry, plus two 1952 methods for measuring beam energy and spatial distribution — one of which found internal beam energy sitting up to ±10% away from the value calculated from field and radius. The fourth is UCRL-3187, a 20-inch three-dee cyclotron built around a magnet that was already available, where moving the ion source off centre roughly doubled the beam. Page references on all four pages are PDF pages, stated as such: an automated check of printed-page alignment disagreed with itself and was not trusted. One shortlisted document was declined rather than hosted — UCRL-8767 is catalogued as a thesis, and a thesis stays its author’s copyright even when it also carries a laboratory report number.
- — TID-454 hosted: seven Carnegie Institute of Technology technical reports on cyclotron component design, published by the Government Printing Office for the AEC in February 1952 — magnets, cloud-chamber coils, high fields, the vacuum system, a rotating condenser for frequency modulation, a three-quarter-scale model oscillator, and a frequency discriminator that times the beam deflector. Its first report is the only cost-minimised magnet design in the collection, optimising coil dimensions against the unit cost of steel, copper and ten-year electrical energy, and concluding that the coils should be built smaller than the optimum because total cost varies slowly near the minimum. The volume was written under the same Navy contract as the already-hosted NYO-780, so the two are now cross-linked as one project’s engineering and its outcome. No searchable copy existed: this scan was assembled from the holding library’s page images, given a text layer with open-source OCR, and had its page boxes corrected to the catalogued 28 cm sheet so the nomographs print to scale. Equations and figure callouts remain unreliable in the text layer, as disclosed on the page. At 33.7 MB it is served intact from files.cyclotroninfo.com. The collection now holds 23 documents across 14 pages.
- — Newsletter section opened at /newsletter/ with an Atom feed. A monthly companion to the site: one machine from the installation census, one design problem worked through the Design Guide’s rules, the month’s census changes computed from dataset snapshots, and an edited news selection, plus a rotating feature (design article, classics reading guide, literature review, or census-anchored machine profile). Issues are published on the site first; every fact in an issue traces to the site’s datasets or a cited source, and issues pass mechanical fact-tracing checks and a review pass before publication. Email delivery is planned via an external service — this site continues to store no addresses and make no third-party requests. The contact page’s privacy wording was updated to match.
- — ORNL-3540 hosted: the 1963 Oak Ridge proposal for the Mc² isochronous cyclotron, an 810 MeV eight-sector machine that was never built — 327 pages of engineering design worked to construction standard, including shielding, costs and schedule, sections almost no accelerator document contains. At 26.5 MB it exceeds the 25 MiB per-file cap that applies on every plan of the site’s hosting platform. Recompression was measured across five approaches and rejected: 71% of the file is 39 scanned photograph and drawing pages, so every route under the cap degraded exactly the pages worth preserving, and the naive whole-file pass made the file larger. The document is therefore served intact from files.cyclotroninfo.com, verified byte-for-byte identical to the OSTI original after upload. The collection now holds 22 documents across 13 pages.
- — Construction Classics wave 3: seven public-domain reports added, taking the collection to 21 hosted PDFs across 12 pages. UCRL-10654 (Smith & Grunder, 1963) gets its own page and closes this collection’s thinnest subject — electrostatic deflector design, with electrode materials ranked by measured spark damage, the VE relationship, a stated design derating, and the tungsten-versus-carbon septum trade. Two new themed cluster pages group the rest by the question a reader arrives with rather than by report number: RF power engineering (a ceramic-tetrode amplifier rebuild, a 1958 argument for square-wave rather than sine acceleration, and a three-phase three-dee system) and commissioning practice (two consecutive 1952 Oak Ridge quarterlies logging three cyclotrons side by side, plus a 1966 axial-ion-source retrofit study). Every document carries a written public-domain rationale; all seven are queued for extraction into the Design Guide.
- — News page rebuilt around this site’s subject. It is now split into "Cyclotrons and particle therapy" and "Wider accelerator news", because a single date-sorted list buried hospital cyclotron and isotope-production stories under collider coverage — on the previous build only about seven of forty items concerned cyclotrons at all. Four cyclotron-specific feeds added (facility/commissioning wording, proton therapy, radioisotope production) plus TRIUMF, a cyclotron laboratory; Physics World narrowed to its medical-physics feed. Institutional feeds are no longer exempt from topic filtering — CERN and Fermilab previously passed unfiltered and contributed items on telescopes, quantum computing and a solar eclipse — and they are now tested for accelerator substance rather than for naming their own laboratory. Also added: a 120-day recency window with graceful fallbacks, exclusion of newsroom calendar entries, and deduplication of syndicated stories republished under the same headline.
- — Builds census now links machines to what was learned on them: entries whose documentation has been mined into the Design Guide carry a "mined into this site" line with live counts — the Houghton College cyclotron (52 rules, 8 library entries) and the Rutgers 12-inch (41 rules, 8 library entries). The Library now supports ?q= deep links like the Design Guide.
- — Proton-therapy sweep: 93 centres added to the Locations census, taking it to 1,250 machines in 92 countries (Georgia new). Proton therapy grows from 20 to 113 entries — 58 AVF-isochronous, 32 synchrocyclotron and 3 superconducting machines; synchrotron-based centres remain out of scope. IBA now leads the census by attributed machines (177) ahead of GE HealthCare (166). PTCOG facility lists added to the registry directory as the source behind the sweep. About page now states explicitly that manufacturers and operators are invited to correct what the census records about their machines, and that such corrections outrank secondary sources.
- — Locations census expanded from 661 to 1,157 machines across 91 countries (previously 64) after a global verification sweep — 945 operating, with proton-therapy, industrial, and teaching machines now separately categorized, and manufacturer attribution on 723 records. Published alongside it: a new registry directory (/locations/registries/) annotating the 78 national and international registries, surveys, and regulatory filings the census is verified against, including what each source covers, where it misleads, and the countries where no usable registry exists at all.
- — Contact page added (/contact/), collecting the routes for corrections, build-census submissions, locations-dataset additions, document suggestions, topic requests, and questions, each with a prefilled subject line, plus what happens to a report and how contributors are credited. Site footer now carries links to About, Contact, Methodology, Changelog, and Data.
- — Content wave 4: four deep dives — Beam Quality, Beam Dynamics, Beam Extraction, and Experiments by Energy Band (/learn/...). Beam Dynamics is an interactive laboratory: two RK4-integrated live figures validated headlessly against the analytic tunes, a tune diagram, and a preset replaying the 184-inch’s 1947 n = 0.2 beam loss from the hosted MDDC memoranda. Experiments by Energy Band works count-rate arithmetic at nanoamp currents with every nuclear threshold and stopping figure fetched live from NNDC and NIST at writing time. All four passed internal adversarial review; Beam Quality also passed the external referee and editor passes, while the other three await an external referee re-run (blocked at review time) and will be noted here when it completes.
- — Content wave 3: four subsystem deep dives — Magnet Design, Ion Sources, Dee RF Coupling, and Beam Measurement (/learn/...) — each synthesizing its Design Guide domains with the hosted construction classics, peer-reviewed (adversarial referee + editor passes), with new figures: iron-circuit flux budget, shimming curve, PIG source cross-section, RF coupling schematics, Faraday cup and signal chain.
- — Corpus update: Design Guide grown from 634 to 709 rules, extracted from the ten hosted wave-2 classics (NYO-780, UCRL-64, MDDC-1045, the six 184-inch commissioning memos, ORNL-2644); each classics page now links the rules extracted from its document, and the Library entries for the five hosted collection items carry fuller assessments.
- — Editorial transparency: methodology page documenting how each section was produced and defining the status vocabulary; this changelog; Design Guide caveat now states the source-extract status explicitly; Library exportable as BibTeX.
- — Homepage reframed around what the site is for, with a beginner start-here path; standard educational-reference notice added to the Calculators and Design Guide.
- — Corpus update: Design Guide grown from 560 to 634 rules (new extraction domain); Library grown from 186 to 201 entries with 8 newly verified outbound links.
- — Citations upgraded: every design rule now shows its full source title linked into the Library; 75 verified outbound links to freely available source documents.
- — Review-driven fixes: skip-to-content link, corrected heading hierarchy, corrections policy on About, search added to navigation; News filtering rebuilt with a measured include/exclude list and fallback so the page never goes empty.
- — Preserved from the old site: patents research page, verified virtual facility tours on Locations, and an expanded communities directory.
- — Site search added: self-hosted Pagefind index across all content pages; no third-party services.
- — Glossary published: 92 cross-linked cyclotron terms.
- — Locations census updated to 530 machines after a verification pass against the IAEA facility list.
- — Graphics and interactivity wave: animated cyclotron and physics figures on Learn, interactive anatomy diagram on the Design Guide, hazard-zone figure and shielding widget on Safety, live geometry diagrams on all seven calculators, historical energy chart, amateur-builds timeline, and the world cyclotron map.
- — Added under Library: a verified external-resources directory and a free-modeling-software guide organized by design question.
- — Site rebuild launched with the full section set: Design Guide (560 rules at launch), seven Calculators with derivations, Legal survey (federal layer plus 51 US jurisdictions), Library (186 entries at launch), Builds (17-machine amateur census), Learn (theory, history, applications, safety, starter path), News, Locations, and About with a public contact address.
- — Rebuild started on a new static architecture, replacing the 2021 WordPress site.
Spotted an error, a dead link, or an out-of-date claim? Corrections are welcome at [email protected] — fixes are made in place and noted here.