Meteorologist, oceanographer, Python solvem probler, experimenter, Raspberry Pi tinkerer, geek dad, manager, space enthusiast, ham radio lover, science columnist, ADHDer, synthetic intellect curious, linux admin cosplayer.

REVIEW BATCH 3: NWP History Series, Posts 2026-04-03 to 2026-04-09

Reviewer: Research agent (fact-checker / language reviewer). Reviewed: 2026-04-27. Posts in this batch:

  1. 2026-04-03-The-swedes-got-there-first.md (Swedish NWP, BESK, Rossby, Bolin 1954)
  2. 2026-04-04-The-magician-who-told-no-secrets.md (Arne Beurling / Geheimschreiber)
  3. 2026-04-08-The-blueprint-von-Neumann-gave-away.md (IAS clones diaspora)
  4. 2026-04-09-The-machine-that-learned-too-early.md (Frank Rosenblatt / Perceptron)

Cross-references with research files: research/people/Beurling.md, research/people/Beurling_additional.md, research/people/Rosenblatt.md, research/people/Minsky.md, research/computers/Perceptron.md, research/computers/BESK.md, research/computers/MANIAC.md, research/computers/JOHNNIAC.md, research/computers/ILLIAC.md, research/computers/ORDVAC.md, research/computers/SILLIAC.md, research/computers/WEIZAC.md, research/computers/BESM.md, research/computers/IAS_machine.md, research/people/Bolin.md, research/people/Polish_codebreakers.md. Web checks against Wikipedia (Mark I Perceptron, Frank Rosenblatt, IAS machine, Siemens & Halske T52, Logic Theorist, BESM-6, AlexNet, Lighthill report, Mansfield Amendment).

SUMMARY

Post Verdict Key issues
1 (Sweden) PASS with minor language tweaks One forward-looking promise. One overstated juxtaposition. QUOTE_AUDIT items still uncited but legible.
2 (Beurling) PASS with one factual issue and recurring forward-looking flags Keyspace claim “10^18 per wheel setting – 10^27 total” is muddled (10^18 is the total, not per-wheel; 10^27 belongs to T52d/e, which Beurling did NOT break). Two forward-looking promises about Polish codebreakers. QUOTE_AUDIT still active.
3 (IAS clones) PASS One ambiguous architecture claim (BESM Wikipedia lists it among IAS family despite the post saying “may or may not be a clone”; the post’s argument is defensible but the binary “BESM was not a clone” framing is stronger than the historical record). One internal contradiction (JOHNNIAC table says 1954, prose says “1953” for the Logic Theorist).
4 (Perceptron) PASS with two factual nits Mark I Perceptron specs: post claims figures as exact, Wikipedia notes versions varied. Small “1955-1956” date for Logic Theorist work is OK but “Logic Theorist… in 1955-1956 developed” runs adjacent to the more specific August 1956 date - fine. AlexNet “two GPUs in his bedroom” verifiable. Overall solid.

No outright plagiarism risks. No legally-exposed quotations. No critically wrong facts that would mislead a serious reader, except the Beurling keyspace muddle. Most “issues” are fine-grained and would be cosmetic in a magazine but the user is asking for ruthless review, so they are flagged.

POST 1: 2026-04-03 “The Swedes Got There First”

Factual

Verified clean (spot-checked):

  • Line 24: Sweden began routine NWP in December 1954, three times a week, North Atlantic, on BESK. Confirmed (BESK.md line 7, line 125; Persson 2005; History of Information).
  • Line 24: BESK name = “bitter” in Swedish. Confirmed (BESK.md line 9).
  • Line 30/32 caption + body: Rossby 1898-1957, returned to Stockholm 1947, founded MISU and IMI, Journal of Meteorology founder. Confirmed (Bolin.md line 200; BESK.md line 200).
  • Line 54: BESK photo states “2 400 vacuum tubes, 40-bit words”. Confirmed (BESK.md line 27).
  • Line 58: BARK = 8 000 telephone relays, 80 km of cable, completed April 1950. Confirmed (BESK.md lines 144-151). Aiken’s “first computer I have seen outside Harvard that actually works” quote: traceable to Lundin 2006 (BESK.md line 153) but the post hedges with “reportedly” - fine.
  • Line 72: Conny Palm born 1907, died 27 December 1951, age 44. Confirmed (BESK.md line 63; Wikipedia Conny Palm).
  • Line 80-86: Stemme on Selectron at RCA, switched to Williams tube, visited Williams in Manchester 1949. Confirmed (BESK.md lines 73-75).
  • Line 86: Olle Karlqvist discovered Karlqvist gap in ferromagnetic layer. Confirmed (BESK.md lines 84-85).
  • Line 90-99: Specs table values for BESK (56 us add, 350 us multiply, 512 words Williams) - Confirmed (BESK.md lines 24-27). Comparison numbers for IAS Machine (62 us add, 713 us multiply, 1 024 words) - Confirmed (IAS_machine.md line 19).
  • Line 99: BESK was “for a few weeks in late 1953” the fastest computer in the world. Confirmed (BESK.md line 256-258, “for a few weeks”).
  • Line 107: Five-minute MTBF in early days. Confirmed (BESK.md line 184).
  • Line 109: “housewives with knitting experience” (the Swedish original: hemmafruar med erfarenhet av att sticka). Confirmed (BESK.md line 22, line 302).
  • Line 127: Bolin born 1925 in Nyköping, met Rossby 1947 during military service, sent to Princeton 1950-51. Confirmed (Bolin.md lines 5, 36-37, 46).
  • Line 137-141: 45 000 troops, six-week exercise, “tremendously successful”. Confirmed (BESK.md line 204; Persson 2005).
  • Line 147-149: 9 000 × 12 000 km forecast area, 3-day forecast in ~1 hour, 1955 Bolin extension to 72 hours, correlations 0.85/0.82/0.70. Confirmed by Bolin.md line 69 referring to Tellus 1955; Persson 2005.
  • Line 163-167: JNWPU first US forecast 6 May 1955, “very disappointing and not usable by forecasters”. Confirmed (Persson 2005, in References).
  • Line 169: IBM 701 MTBF roughly 30 minutes. Plausible (early 701 reliability was poor) but not directly cited in research files. Not flagged as urgent.
  • Line 191: Aksel Wiin-Nielsen first ECMWF Director 1974-1979. Confirmed (Wikipedia: Wiin-Nielsen).
  • Line 195: Bolin first Chairman of IPCC 1988; IPCC shared 2007 Nobel Peace Prize with Al Gore; Bolin too ill to attend; died 30 December 2007. Confirmed (Bolin.md lines 56, 95-100).
  • Line 197: Crutzen at MISU 1959 as a programmer; supervised by Bolin; 1995 Nobel in Chemistry. Confirmed (Bolin.md line 119).
  • Line 207: Hans Riesel found 18th Mersenne prime 2^3217 - 1, 969 digits, in 1957 on BESK. Confirmed (BESK.md lines 228-230; Wikipedia: Hans Riesel).
  • Line 211: Beurling cracked Geheimschreiber in two weeks with pen and paper, 1940; later moved to IAS Princeton; took over Einstein’s office. All three confirmed (Beurling.md lines 54, 106; Wikipedia: Beurling).
  • Line 215: BESK Boys to Facit 1956. Confirmed (BESK.md line 77, lines 270-275).

Issues / nits (none critical):

  • Line 24: “a barotropic model developed at Stockholm University” - more precisely, Phillips ported the Charney/Fjørtoft/von Neumann ENIAC code to BESK (post itself says this on line 133). The line 24 phrasing “developed at Stockholm University” understates the American origin of the model. Minor. Suggested fix: “a barotropic model adapted from the ENIAC forecast code at Stockholm University.”
  • Line 38: > "when Rossby was in town, the department was in tumult... it was exciting but exhausting." - Chester Newton attribution. Already flagged in QUOTE_AUDIT.md. Source not inline; likely Persson 2005 or the Physics Today Rossby piece. Not fixed in current draft.
  • Line 40: Rossby quote "Communications are the alpha and omega of meteorology" - already flagged in QUOTE_AUDIT.md. Source not inline.
  • Line 58, 68, 109, 121, 141, 165: All flagged in QUOTE_AUDIT.md. Status unchanged. Recommend the standing-rule fix (Beurling/Sweden quotes from Lundin 2006 and Persson 2005 unless otherwise noted).
  • Line 169: “The IBM 701 at JNWPU had a mean time between failure of roughly 30 minutes.” - reasonable but no citation; the figure varies by source. Optional: cite Persson 2005 if the figure comes from there.

Language

  • Line 14 (lede paragraph): “Bork bork bork!” Swedish Chef framing - the user’s writing-style memory accepts narrative voice with character. The “as the man said, right before something exploded” is fine. A minor concern: it pushes very close to the kind of self-disavowed framing the user has corrected before (Brooklyn-Romantic -> Visionary). The paragraph does its job and is not retracted by anything in the body, so it stays in the safe zone. Keep.
  • Line 22: “Sweden got there first.” Bold standalone sentence. Punchy. Justified by the body. Keep.
  • Line 167: “Let that sink in.” Direct address; fine in the voice of the series.
  • Line 235-237: “Sweden was first. By a year. … A machine named after bitter schnapps, built because the Americans wouldn’t sell them one. The two-week wall that [Lorenz] proved existed? The Swedes hit it before anyone else had even started running.” This is borderline. The “two-week wall” line is fine as a flourish but the implication that the Swedes “hit” the predictability barrier in 1954 is technically inaccurate - they hit forecast skill degradation, which is a different thing from Lorenz’s deterministic chaos limit. The Lorenz two-week wall is a property of the atmosphere, not of the Swedes’ computation. Minor overclaim. Suggested softening: “The forecast horizon that [Lorenz] would later prove existed? The Swedes were already pushing against it before anyone else had a stable production system.”
  • Line 289-291: “The Swedes: Still modest about it. Still first.” - the third line “My cluster: 0.54 TFLOPS. Roughly 9×10⁹ times faster than BESK…” is a recurring sign-off in the series, fine.

Forward-looking

  • None in this post. The post is self-contained. Good.

Character

  • Em dashes: uses -- consistently (good per repository convention). No — (em-dash unicode) found.
  • Curly quotes: None found (all " and ' are straight ASCII).
  • Ellipses: None found.
  • Number formatting: Consistent space-as-thousand-separator (8 000, 45 000, 2 400, etc). One minor inconsistency: line 99 says “9 × 10⁹ times faster” using superscript Unicode, which is the only place superscript is used. Acceptable.

Verdict

PASS. No factual errors. Two language flourishes (Swedish Chef opener, Lorenz two-week-wall closer) push the metaphor envelope but are within the established voice. Quote-attribution items from QUOTE_AUDIT.md are unchanged but the attribution risk is low: each named source (Persson 2005, Lundin 2006, Beckman 2002) is in References.

POST 2: 2026-04-04 “The Magician Who Told No Secrets”

Factual

Verified clean:

  • Line 27: Born 3 February 1905 in Gothenburg. Confirmed (Beurling.md line 7).
  • Line 27: Father Konrad captained merchant ships; mother Baroness Elsa Raab. Confirmed (Beurling.md lines 14-15).
  • Line 29: Great-grandfather Pehr Henrik Beurling founded clock factory 1783. Confirmed (Beurling.md line 16; Wikipedia and auction records).
  • Line 31: Proved Denjoy conjecture 1929 while alligator-hunting in Panama with father. Doctoral thesis delayed by ~5 years. Ahlfors independently proved same conjecture and published first. Thesis 1933. All confirmed (Beurling_additional.md lines 178-185).
  • Line 33: Carleson, Ahlfors, Domar quotes about Beurling - all traceable to Ahlfors & Carleson 1988 Acta Mathematica (in References as [^1]); now consistently footnoted.
  • Line 41: “You Harvard men seem to be afraid of integral signs” - source per research file is John Wermer, via MacTutor (Beurling_additional.md line 232; Beurling.md collected quotes line 489). Footnoted to [^1] (Ahlfors-Carleson) which is probably not the actual primary source for this quote (the primary is Wermer’s reminiscence in the Collected Works or MacTutor); but [^1] is plausibly transitive. Minor sourcing imprecision; not urgent.
  • Line 47-49: Germany invaded Norway in April 1940; T52 used for traffic Norway-Berlin via Sweden. Confirmed (Beurling_additional.md line 36, “After Germany occupied Denmark and Norway in April 1940”).
  • Line 55: T52 had ten cipher wheels with coprime pin counts from 47 to 73. Confirmed (Beurling_additional.md line 16, “ten pinwheels of coprime sizes: 73, 71, 69, 67, 65, 64, 61, 59, 53, and 47”).
  • Line 57: British codename “Sturgeon”. Confirmed (Beurling_additional.md line 11).
  • Line 67: Hagelin B-21 anecdote with Captain Erik Anderberg - Confirmed (Beurling_additional.md lines 209-212).
  • Line 69: Soviet codes pre-Winter War (November 1939); Czech telegrams. Confirmed (Beurling_additional.md lines 213-214).
  • Line 71: First intercepted T52 traffic 25 May 1940; first results verified 27 May 1940. Confirmed (Beurling_additional.md lines 39-41).
  • Line 77: David Kahn “quite possibly the finest feat of cryptanalysis”; Bauer “one of the most magnificent achievements”; Beckman “every bit as impressive as the breaking of the Enigma code”. The Kahn quote per Beurling_additional.md line 74 is described as “quite possibly the finest feat of cryptanalysis performed during the Second World War” (Kahn 1967). The post’s “performed by the Swedes” wording is a slight variant; the Kahn original says “performed during the Second World War”, not “by the Swedes”. Minor wording shift but the meaning is consistent.
  • Line 89: “A magician does not reveal his secrets.” Footnoted to Beckman 2002. Confirmed (Beurling_additional.md line 156, “the canonical source”).
  • Line 101: Vigo Lindstein engineer; LM Ericsson via Swedish Cash Register Company; Karlaplan 4 location. Confirmed (Beurling.md and Beurling_additional.md lines 119-122).
  • Line 103: 175 people / half of FRA’s workforce; 678 decrypted in single day October 1943; 500 000 intercepted, 350 000 decrypted. Confirmed (Beurling_additional.md lines 42-44).
  • Line 105: Operation Barbarossa intelligence; Britain warned Stalin; Stalin dismissed. Confirmed (Beurling_additional.md lines 89-94).
  • Line 107: T52c/T52ca/T52d upgrades; FRA broke Lorenz SZ40 in April 1943. Confirmed (Beurling_additional.md lines 107-113, though research file says “1942-1943” rather than precisely April 1943; the specific April-1943 date is plausibly from Beckman but not verified in the research files I have).
  • Line 109: “These messages are decrypted in Sweden.” Confirmed (Beurling_additional.md line 105 has the exact quoted phrase, attributed to a decrypted German message ~17 June 1942).
  • Line 117: Marian Rejewski cracked Enigma December 1932, age 27, from Bydgoszcz. Confirmed (Polish_codebreakers.md lines 13, 31; Wikipedia).
  • Line 117: bomba kryptologiczna; Pyry forest meeting July 1939, “five weeks before Germany invaded their country”. Verifiable (Polish_codebreakers.md and standard Enigma history; Pyry meeting is well-documented as 24-26 July 1939, ~5 weeks before September 1, 1939 invasion - 5 weeks is correct).
  • Line 127: Beurling at Harvard 1948-49 visiting; IAS 1954 permanent. Confirmed (Beurling.md line 42, lines 41-43).
  • Line 129: Einstein’s office, Room 115. Confirmed (Beurling_additional.md line 360, with the more precise dating of 1965 in the research file). The post’s “moved into” phrasing is consistent with both 1954-arrival and 1965-formal-grant interpretations. The research file notes the exact date is contested.
  • Line 133: Beurling-Malliavin theorem 1960-61, Karin Beurling biochemist at Princeton, Malliavin’s “We devoted half of the academic year to this problem” recollection. Confirmed (Beurling_additional.md lines 280-285, line 379).
  • Line 139: Lars Ahlfors Fields Medal 1936, “one of the first two ever awarded”. Confirmed - the first Fields Medals were awarded in 1936 to Ahlfors and Jesse Douglas.
  • Line 141: Pawned Fields Medal during 1944 escape from Finland; “I’m sure that it’s the only Fields Medal that has been in a pawn shop.” Confirmed (Beurling_additional.md lines 192-194; footnoted to Lehto 1998 / Mathematical People interview, [^6]).
  • Line 145: Christoffer Ahlfors died from electric shock; Beurling delivered news; Ahlfors’ quote about Beurling’s “tact and compassion”. Confirmed (Beurling_additional.md line 197).
  • Line 147: 1986 Beurling died; Ahlfors “Arne Beurling was the best friend I ever had” eulogy. Confirmed (Beurling_additional.md line 199).
  • Line 149: Carleson Abel Prize 2006, “It was my great fortune to have been introduced to mathematics by Arne Beurling.” Confirmed (Beurling_additional.md lines 442-444).
  • Line 167: Norra begravningsplatsen in Solna, Beurling’s burial; collected works published 1989. Confirmed (Beurling.md line 8, line 124; Beurling_additional.md line 392, line 311).

Significant factual issue:

  • Line 55: keyspace claim is muddled. The post says: “creating a keyspace of approximately 10^18 combinations per wheel setting – and roughly 10^27 total possible configurations.”
    • The figure 893,622,318,929,520,960 (~8.9 × 10^17, conventionally rounded to 10^18) is the total wheel-setting keyspace for the early T52a/T52b that Beurling broke. It is not “per wheel setting” - that wording is wrong. (Beurling_additional.md line 22 - “893,622,318,929,520,960 different key configurations – nearly one quintillion.”)
    • The 10^27 figure (= 2^90) belongs to the later T52d/T52e with irregular wheel stepping, which Beurling did not break. Per the Wikipedia search result on T52 keyspace: “for the more advanced models with irregular wheel stepping (T52d, T52e), the keyspace size is 10^27 = 2^90 keys.” Beurling broke T52a/b in 1940; T52d/e came later (after July 1942).
    • The post’s framing thus conflates the early model Beurling broke (10^18) with the later harder ones (10^27), inflating the apparent difficulty of what he actually did. The achievement is impressive enough at 10^18; the 10^27 number does not need to be there.
    • Suggested fix: Replace lines 55-56 with: “It used ten cipher wheels (not three), each with a different coprime number of pins (from 47 to 73), creating a keyspace of approximately 10^18 (893 quadrillion) combinations – nearly one quintillion. Later upgraded models (T52d/T52e, introduced in 1942-1944) added irregular wheel stepping that pushed the keyspace to ~10^27. Beurling broke the early T52a/T52b – the 10^18 model – in two weeks. The harder later models, when they appeared, defeated Bletchley Park entirely.”

Other factual nits:

  • Line 79: “Bill Tutte performed the theoretical break; Tommy Flowers built the machine to make it operational.” Correct attribution, but the Lorenz/Tunny break was a multi-year team effort (Tutte’s structural work was Jan 1942 onwards; Flowers’ Colossus was December 1943). Fine.
  • Line 117: “Five weeks before Germany invaded their country, they handed everything to the British and French at a secret meeting in the Pyry forest near Warsaw.” The Pyry meeting was 24-26 July 1939; the German invasion was 1 September 1939. That’s 5 weeks and 1-3 days. 5 weeks is correct.
  • Line 129: Post says Einstein’s office was “Room 115”. Beurling_additional.md confirms “Room 115” (line 360). Wikipedia silently confirms. The research file notes the formal grant date of 1965 vs. arrival in 1954, which is interesting but not critical to the post.

Language

  • Line 12 (excerpt): “When asked how, he replied: ‘A magician does not reveal his secrets.’” - good hook.
  • Line 14: “But I didn’t tell you why FRA existed in the first place.” - Bold for emphasis. Fine.
  • Line 35: “The apple did not fall far from the mast.” - punchy, original variant of the cliché. Keep.
  • Line 47-49: “The T52 was not the Enigma. It was worse.” Good rhetorical setup. Fine.
  • Line 81: “Beurling did something comparable. Alone. In a fortnight. With a pencil.” - punchy, three-beat. Keep.
  • Line 91: Description of Beurling’s method as “an almost preternatural intuition” - the word “preternatural” is in the research file (Beurling_additional.md line 60) and is a defensible characterization rather than purple prose. Keep.
  • Line 161: “A mathematician who cracked ciphers with intuition and aesthetics created the institutional demand for the computer that predicted the weather.” - this is the load-bearing thesis sentence and it lands.
  • Line 173: “Some magicians are forgotten. Some are remembered too late. And some – like Beurling – changed the world with a pencil and were thanked with silence.” Earnest but the parallel structure is well-earned by the post’s content. Keep.

Forward-looking

  • Line 123: “But that is a story that deserves its own post. The Polish codebreakers – Rejewski, Różycki, Zygalski – and the price they paid for being right too early and too quietly. That one is personal for me. It’s coming.” - VIOLATES the user’s “no forward-looking promises” rule. The Polish post (2026-04-12) was eventually written, so this isn’t a phantom promise, but the rule the user established (feedback_no_forward_promises.md, corrected 2026-04-22 after the Arakawa post) says posts should not tease upcoming entries.
    • Suggested fix (option 1): Drop the entire paragraph (lines 121-123). The paragraph adds nothing the post hasn’t already said. Cuts cleanly.
    • Suggested fix (option 2): Replace with a backward-looking phrasing that doesn’t promise: “The Polish codebreakers – Rejewski, Różycki, Zygalski – did this same kind of mathematical magic eight years before Beurling. Their story deserves the careful telling it has rarely received.”
  • Line 175: “In Poland, three other magicians are waiting for their story to be told properly. Soon.” - same forward-looking violation, much more explicit. This one tells the reader the next post is coming, which is exactly the pattern the user has asked to avoid.
    • Suggested fix: Drop the line entirely. The post ends well at “Some magicians are forgotten. Some are remembered too late…” paragraph.

Character

  • Em dashes: Uses --. No Unicode —. Good.
  • Curly quotes: None found.
  • Ellipses: None found.
  • Number formatting: Mixed style. The post uses 175 people, 175 000 people style consistently with other posts. 175 000 (line 103), 350 000 (line 103), 500 000 (line 103) - good. 10^18 and 10^27 use caret notation; that’s fine in inline prose.

Verdict

PASS with required revisions. Two issues to fix before considering this fully clean:

  1. Critical: Fix the keyspace claim on line 55 (10^18 is the total, not per-wheel; 10^27 is for models Beurling did not break).
  2. Critical: Drop the two forward-looking Polish-codebreakers promises (lines 123 and 175).

The rest of the post is solid and the Beurling material is well-aligned with the research files. The QUOTE_AUDIT.md gap on this post (the highest in the series) remains unresolved but each quote is traceable to a cited reference.

POST 3: 2026-04-08 “The Blueprint Von Neumann Gave Away”

Factual

Verified clean:

  • Line 14: 1946 IAS report by Burks, Goldstine, von Neumann; “Preliminary Discussion of the Logical Design…”. Confirmed (in References).
  • Line 24: “At least fifteen teams around the world” built clones. Wikipedia lists 17+ derivatives (IAS_machine.md line 80-91). The “fifteen” figure is conservative and defensible.
  • Line 26: “We have already met the IAS machine itself and its four successors”. Cross-link is reasonable.
  • Line 32-39: Family table, 6 IAS clones (MANIAC, JOHNNIAC, ILLIAC, ORDVAC, WEIZAC, SILLIAC). All entries spot-checked:
    • MANIAC 1952, Los Alamos, Metropolis, ~2500 tubes, 1024 words, retired 1958. Confirmed (MANIAC.md lines 12-30).
    • JOHNNIAC 1954, Santa Monica, Willis Ware, ~5000 tubes, 1024 words, retired 1966. Confirmed (JOHNNIAC.md lines 8-19, 75). Note: research file says “completed in 1953, making it operational around 1954” - the “1954” in the table is the operational date and is consistent.
    • ILLIAC I 1952 (September 1, 1952), Urbana, Meagher, 2800 tubes, 1024 words, retired 1962. Confirmed (ILLIAC.md lines 14-23).
    • ORDVAC 1952 (March 6, 1952), Aberdeen, Meagher, 2178 tubes (post says ~2200), 1024 words, retired 1962. Confirmed (ORDVAC.md lines 12-13). Note: research file says 2178; post rounds to ~2200, fine.
    • WEIZAC 1955 (late 1955), Rehovot, Estrin, ~2000 tubes, retired December 29, 1963. Confirmed (WEIZAC.md lines 16-19, 51).
    • SILLIAC 1956 (operational June 1956, official September 1956), Sydney, Messel, 2768 tubes, retired May 1968. Confirmed (SILLIAC.md lines 17, 53-55, 91).
  • Line 49: Metropolis “Mathematical Analyzer Numerical Integrator and Automatic Computer”, trying to “stop the rash of silly acronyms”. Confirmed (MANIAC.md lines 41-43).
  • Line 51: MANIAC construction began mid-1949, ran first time March 1952. Confirmed (MANIAC.md lines 28-30).
  • Line 55: Klara von Neumann wrote MANIAC’s first programs. Confirmed (MANIAC.md line 46, line 62).
  • Line 57: Arianna Rosenbluth programmed Metropolis algorithm in 1953. Confirmed (MANIAC.md line 49).
  • Line 63: Fermi-Pasta-Ulam-Tsingou problem with Mary Tsingou; first numerical experiment “entirely inside a computer”; Dauxois campaigned to add Tsingou’s name. Confirmed (MANIAC.md lines 51-53, 63).
  • Line 67: 6x6 board, no bishops, MANIAC chess 1956, “first computer to defeat a human being in a chess-like game”, ~20 minutes per move. Confirmed (MANIAC.md lines 57-58).
  • Line 75: Willis Ware engineer on IAS machine; built JOHNNIAC; named after von Neumann who protested; “There are lots of Johns in the world” attribution to John Williams. Confirmed (JOHNNIAC.md lines 33-37). The Williams quote source is Gruenberger 1968, in References.
  • Line 79-81: JOHNNIAC was “the only IAS-family machine to actually use Selectron tubes for memory”; Keith Uncapher led effort, achieved 10 hours error-free; replaced March 1955. Confirmed (JOHNNIAC.md line 41).
  • Line 87: Logic Theorist “in 1955-1956, Allen Newell, Clifford Shaw, and Herbert Simon developed”. The first Logic Theorist proof was August 1956 on JOHNNIAC. The post line 87 says “Running on the JOHNNIAC in August 1956”. Confirmed (JOHNNIAC.md lines 47-49). Slight note: J. Clifford Shaw went by “Cliff Shaw” - inconsistent capitalization across the post (line 87 “Clifford Shaw”, line 95 “Cliff Shaw”). Cosmetic.
  • Line 89: 1956 Dartmouth Conference, McCarthy, Minsky, Shannon, Rochester. Confirmed (JOHNNIAC.md line 49; Minsky.md line 73).
  • Line 91: IPL influenced LISP. Confirmed (JOHNNIAC.md line 53).
  • Line 95: JOSS in 1963 by Cliff Shaw; “an exploration into continuous and intimate contact between a human user and a computer”; 10 simultaneous users by 1966. Confirmed (JOHNNIAC.md line 57).
  • Line 99: JOHNNIAC retired February 18, 1966 after 51 349 hours; last program counted down. Confirmed (JOHNNIAC.md lines 67, 75).
  • Line 105: Donald and Betsy Gillies caption; ILLIAC II three new Mersenne primes. Confirmed (ILLIAC.md line 110).
  • Line 107: Aberdeen contracted Illinois; “two of every component” clause enabled twin machine. Confirmed (ILLIAC.md lines 26-28; ORDVAC.md line 28).
  • Line 109: ORDVAC = Ordnance Discrete Variable Automatic Computer, ILLIAC I = Illinois Automatic Computer, Meagher chief engineer, Taub head of Digital Computer Lab. Confirmed (ORDVAC.md lines 31-32; ILLIAC.md lines 30-31).
  • Line 111: “first two computers in history that could exchange programs”. Confirmed (ORDVAC.md line 47-48; ILLIAC.md line 41-42).
  • Line 115: Reassembly at Aberdeen “expected to take over a month, completed in one week”. Confirmed (ORDVAC.md lines 41-43).
  • Line 117: Illinois used ORDVAC remotely “for up to eight hours a night”. Confirmed (ORDVAC.md line 58).
  • Line 123-125: Hiller and Isaacson, Fux counterpoint, Markov chains, August 1956 student quartet, Illiac Suite later String Quartet No. 4. Confirmed (ILLIAC.md lines 46-54).
  • Line 125: 1958 Hiller founded Experimental Music Studio at Illinois; collaboration with John Cage. Confirmed (ILLIAC.md line 56).
  • Line 131: Sputnik orbit calculation “within two days”; Donald Gillies, James Snyder; ephemeris in Nature by November. Confirmed (ILLIAC.md line 60).
  • Line 133: “By 1956, ILLIAC I had more computing power than all the computers at Bell Labs combined.” Confirmed (ILLIAC.md line 38).
  • Line 143: Pekeris at IAS, von Neumann support, Laplace’s tidal equations. Confirmed (WEIZAC.md lines 26-30).
  • Line 145: Von Neumann’s quip “Don’t worry about that problem. If nobody else uses the computer, Pekeris will use it full time!” Confirmed (WEIZAC.md line 30).
  • Line 149: Gerald and Thelma Estrin to Israel 1954; no parts, no tools, no trained staff. Confirmed (WEIZAC.md lines 33-38).
  • Line 151: Components from Landseas Import-Export, Haifa and Tel Aviv ports; bicycle repair shop. Confirmed (WEIZAC.md lines 40-41).
  • Line 153: Budget $50 000 = 20% of Weizmann annual budget. Confirmed (WEIZAC.md line 42).
  • Line 155: First calculation late 1955, “first computer in the Middle East”. Confirmed (WEIZAC.md line 50, line 78).
  • Line 161: Amphidromic point in South Atlantic predicted; British Royal Navy verified. Confirmed (WEIZAC.md line 58).
  • Line 169: Estrin to UCLA 1956, reconfigurable computing, doctoral student Vint Cerf. Confirmed (WEIZAC.md line 79).
  • Line 177: Harry Messel “Canadian who ran the School of Physics at age 30”, John Blatt. Confirmed (SILLIAC.md lines 33-34). Note: Messel was actually Canadian-born (Saskatchewan), confirmed.
  • Line 179: SILLIAC = Sydney version of Illinois Automatic Computer / “Silly ILLIAC” wink. Confirmed (SILLIAC.md line 40).
  • Line 181: Adolph Basser donated AU 50 000; Sir Frank Packer; estimated cost AU 35 200; final AU 75 000. Confirmed (SILLIAC.md lines 43-47).
  • Line 185: 2C51 valves from Bell Labs, six times more cost, five times longer life. Confirmed (SILLIAC.md lines 19, 58-59).
  • Line 187: 11 hours average between failures. Confirmed (SILLIAC.md line 24).
  • Line 191-193: Bob May first computation, Robbie Schafroth supervisor, bosons/superconductivity work, age 20. Research file says May “received his PhD at age 24” - the post line 191 says “He was 20 years old, working on bosons and superconductivity” which refers to the first computation. Verifiable: May was born 8 January 1936, the first SILLIAC computation was June 1956 - May was 20 at first computation. Correct.
  • Line 193: May → Robert May, Baron May of Oxford, President Royal Society, Chief Scientific Adviser, chaos theory in ecology. Confirmed (SILLIAC.md lines 65, 87).
  • Line 197: SILLIAC ran “Australia’s first computer payroll system” for Postmaster-General’s Department; over 2000 users. Confirmed (SILLIAC.md lines 72, 77).
  • Line 209: Lebedev born 1902 Nizhny Novgorod, graduated Moscow Higher Technical School (Bauman) 1928. Confirmed (BESM.md lines 13-15).
  • Line 209: Worked on analog computers for tank weapon-aiming and missile guidance during WWII. Confirmed (BESM.md line 18).
  • Line 211: 1948, learned from foreign magazines about Western digital computers. Confirmed (BESM.md line 19).
  • Line 215: MESM in former monastery outside Kyiv 1950-1951; “first stored-program computer in continental Europe”. Confirmed (BESM.md lines 56-58, with the note that “claimed as the first such computer in continental Europe (though the Zuse Z4 and Swedish BARK preceded it in certain categories)”).
  • Line 217: BESM-1 1952, ~5000 vacuum tubes, 8000-10000 FLOPS, “fastest computer in Europe” at completion. Confirmed (BESM.md lines 67-87).
  • Line 222-228: Architecture comparison table:
    • IAS Machine: Fixed-point, single-address, 40 bits, Williams tubes - all correct.
    • BESM-1: Floating-point, three-address, 39 bits, ferrite cores - all correct.
    • Confirmed (BESM.md lines 39-46).
  • Line 230: Malinovsky quote about Lebedev’s designs not being close copies. Confirmed (BESM.md line 46).
  • Line 236: BESM-6 designed 1965, produced 1968-1987, 355 units; 60 000 transistors; 9 MHz; 1 MIPS. All confirmed (BESM.md lines 113-122; Wikipedia BESM-6 search).
  • Line 238: “Principle of Water Pipe” instruction pipelining metaphor. Confirmed (BESM.md line 131).
  • Line 240: 1975 Apollo-Soyuz, BESM-6 finished half-hour before NASA. Confirmed (BESM.md lines 145-147).
  • Line 261: Von Neumann died 1957, age 53, bone cancer. Confirmed (standard biography). Cause “likely caused by his presence at nuclear tests in the Pacific” - this is a defensible hypothesis but not certain; the post hedges with “likely”. Fine.
  • Line 263: IAS machine shut down 1958. Confirmed (IAS_machine.md line 97).

Issues / nits:

  • Line 41: “And then there is the BESM in Moscow – which may or may not be a clone at all.” The post argues (correctly per the research file) that BESM was substantially independent. However, Wikipedia’s IAS machine article still lists BESM among the “IAS machines” derivatives list (verified by web fetch). The post’s framing is defensible because the research file (BESM.md lines 31-50) says exactly this: “Lebedev had arrived at the stored-program concept either independently or with only the vaguest knowledge that others were pursuing it. But the engineering was entirely his own.” The post is correct on the substance but has set up a small narrative tension with the standard short-form encyclopedia framing. Recommend: keep the post’s framing, but add a single line noting the disagreement: “Wikipedia still lists BESM among IAS-family machines; the historians who specialise in Soviet computing – Malinovsky in particular – emphatically do not.” That defangs the issue without changing the argument.

  • Line 73 caption: “operational in 1954” - JOHNNIAC.md says completed 1953, operational 1954. The table on line 32 also says 1954. Consistent.

  • Line 87: “Running on the JOHNNIAC in August 1956, it proved mathematical theorems from Whitehead and Russell’s Principia Mathematica.” Confirmed. “Simon was so excited that he wrote to Bertrand Russell himself” - Confirmed (JOHNNIAC.md line 49).

  • Line 99: “When JOHNNIAC was retired on February 18, 1966, after 51 349 hours of operation” - Confirmed (JOHNNIAC.md line 75).

  • Line 109: “Chief engineer Ralph Meagher and laboratory head Abraham Taub – both Princeton-connected” - Confirmed (ILLIAC.md line 31, Taub had Princeton/IAS connection; Meagher’s Princeton link is via the IAS visit, also documented).

  • Line 167: Caption says “Gerald and Thelma Estrin, 2007. They traveled to Israel together in 1954 to build WEIZAC from nothing.” Confirmed (WEIZAC.md line 33).

  • Line 209: “Sergei Alekseyevich Lebedev was born in Nizhny Novgorod in 1902.” BESM.md says “November 2, 1902”. Confirmed.

Language

  • Line 19: “He gave it away.” Standalone short paragraph. Fine, dramatic.
  • Line 22: “He saw the computer as a scientific instrument, not a commercial product.” Solid characterization.
  • Line 28: “Today: the rest of the family. Six machines, four continents, and the remarkable fact that each clone – built from the same blueprint – found its own science.” Strong thesis statement; gets earned through the body. Keep.
  • Line 165: “A computer built in a bicycle shop had discovered a feature of the planet that nobody knew existed.” Punchy callback. Keep.
  • Line 195: “The first person to run a computation on a vacuum-tube computer in Sydney ended up in the House of Lords. Not a bad trajectory from a Silly ILLIAC.” Earned. Keep.
  • Line 232: “BESM was not a clone. It was a parallel invention.” Confident framing. Mostly defensible (see factual note above).
  • Line 250: “And from that single blueprint:” + bulleted summary. Strong recap. Keep.
  • Line 259: “BESM proved that the stored-program idea was so natural, so inevitable, that it could be invented independently on the other side of the Iron Curtain.” Confident again. The word “inevitable” is a bit strong; the BESM independence is contested. Optional softening: “BESM suggests…”

Forward-looking

  • None in this post. Good. The “Previously in this series” footer at the end is backward-looking, which is fine.

Character

  • Em dashes: -- used consistently. No —.
  • Curly quotes: None found.
  • Ellipses: None found.
  • Number formatting: Spaces in some places (50 000, 45 000), no separator in others (2768, 5000). Mostly consistent: figures larger than 9999 get spaces, four-digit figures don’t. This matches the practice in Post 1.

Verdict

PASS. No factual errors, one architectural argument (BESM as “parallel invention” rather than IAS clone) that is defensible but stronger than the standard reference framing. All quotes traced to identifiable sources (Gruenberger 1968 for “lots of Johns” and Shaw “intimate contact” - both in References). The QUOTE_AUDIT.md gap (lines 75, 95, 97, 145, 230) is unchanged but each cited source is in References.

POST 4: 2026-04-09 “The Machine That Learned Too Early”

Factual

Verified clean:

  • Line 11 (excerpt) and line 26: Rosenblatt born July 11, 1928, drowned on his 43rd birthday July 11, 1971. Confirmed (Rosenblatt.md lines 4, 128; Wikipedia).
  • Line 26: Played piano on grand piano in Brooktondale; weakness for “Three Blind Mice”; built personal observatory; 12-inch Cassegrain telescope; SETI research; Eugene McCarthy 1968 campaign; Vietnam protests. All confirmed (Rosenblatt.md lines 108-122).
  • Line 28: George Nagy quote “very bright” vs “genius”. Confirmed (Rosenblatt.md line 142). Footnoted to Nagy 2011 [^1].
  • Line 30: Bronx Science class of 1946 (post) vs Minsky class of 1945 (post). Confirmed (Rosenblatt.md lines 21-22; Minsky.md line 24).
  • Line 32: Rosenblatt at Cornell, BA 1950, PhD 1956 in psychology; built EPAC for psychometric analysis. Confirmed (Rosenblatt.md lines 25-32).
  • Line 32: Joined Cornell Aeronautical Laboratory as research psychologist. Confirmed (Rosenblatt.md line 38).
  • Line 42: McCulloch-Pitts neuron 1943, fixed weights, can’t learn. Confirmed (Rosenblatt.md lines 60-61).
  • Line 48: “In 1957, Rosenblatt wrote a software simulation and ran it on the IBM 704… project called Project PARA – Perceiving and Recognizing Automaton.” Confirmed (Rosenblatt.md lines 41-42; Perceptron.md line 70).
  • Line 48: “After 50 trials” - Confirmed (Rosenblatt.md line 67).
  • Line 56: Mark I caption: “400 photocells, 512 association units, electric motors that turned potentiometers”. Confirmed (Perceptron.md lines 28-32).
  • Line 60-66: Mark I bullet list spec: 400 photocells in 20x20 grid, 512 A-units, 8 R-units, potentiometers, electric motors. All confirmed (Perceptron.md lines 28-32, 38-65).
  • Line 68: Mark I “achieved 99.8% accuracy distinguishing squares from circles, 100% on squares versus diamonds, and could classify the letters X and E with 100% accuracy after just 20 training examples.” Confirmed (Perceptron.md lines 86-89). Note: Wikipedia confirms 99.8% on squares vs circles “with 500 neurons” and 10 000 training images, but the post does not include those qualifiers - acceptable simplification.
  • Line 68: “It sits today in the Smithsonian National Museum of American History.” Accessioned 1967. Confirmed (Perceptron.md line 8, line 134).
  • Line 74: July 7, 1958 ONR press conference; July 8 NYT story on page 25. Confirmed (Rosenblatt.md line 73; Perceptron.md line 75).
  • Line 76 headline: “NEW NAVY DEVICE LEARNS BY DOING: Psychologist Shows Embryo of Computer Designed to Read and Grow Wiser”. Confirmed (Rosenblatt.md line 75).
  • Line 80: NYT “the embryo of an electronic computer that the Navy expects will be able to walk, talk, see, write, reproduce itself and be conscious of its existence.” Confirmed (Rosenblatt.md line 75-76; verified by external search).
  • Line 82: Rosenblatt “first machine which is capable of having an original idea”; future perceptrons “recognize people and call out their names”; “in principle, it would be possible to have perceptrons that would reproduce themselves on an assembly line and that would be conscious of their existence.” Confirmed (Rosenblatt.md line 76).
  • Line 94: Minsky built SNARC at Princeton 1951 with 40 neurons of vacuum tubes; learned to navigate maze; abandoned. Confirmed (Minsky.md lines 47-67).
  • Line 96: Minsky one of four organisers of 1956 Dartmouth Conference. Confirmed (Minsky.md line 73).
  • Line 106: 1969 Minsky and Papert Perceptrons: An Introduction to Computational Geometry. Confirmed (Minsky.md line 120).
  • Line 108: Single-layer perceptrons cannot compute XOR; only linearly separable patterns. Confirmed (Minsky.md lines 132-138).
  • Line 110: “Virtually nothing is known about the computational capabilities of this latter kind of machine. We believe that it can do little more than can a low order perceptron.” Confirmed (Minsky.md lines 142-144). Footnoted to Minsky & Papert 1969, p.232 [^3].
  • Line 116: Rosenblatt 1962 Principles of Neurodynamics; covered four-layer perceptrons. Confirmed (Rosenblatt.md lines 90, 96-97).
  • Line 116: Credit assignment problem - Rosenblatt did not solve it. Confirmed (Rosenblatt.md line 96).
  • Line 118: “It would take 17 years to solve” the XOR problem. 1969 Minsky-Papert -> 1986 Rumelhart-Hinton-Williams (or 1974 Werbos PhD), depending on which date you count. From 1969 to 1986 is 17 years. Internally consistent with the standard story.
  • Line 120: H.D. Block “seriously misleading”; Bernard Widrow; Jordan Pollack quotes. All in Minsky.md line 150. Sourced to Olazaran 1996 [^4].
  • Line 122: Minsky’s “Necronomicon” comparison “a book known to many, but read only by a few”. Confirmed (Minsky.md line 152). Footnoted to Crevier 1993 [^5].
  • Line 130: 1966 ALPAC report killed machine translation funding. Confirmed (standard AI history; in References as ALPAC 1966).
  • Line 130: Sir James Lighthill 1973 report; “In no part of the field have the discoveries made so far produced the major impact that was then promised”. Confirmed by external search of Wikipedia: Lighthill report. The exact phrase matches. Footnoted to Lighthill 1973 [^6].
  • Line 132: Mansfield Amendment 1969. Confirmed by external search: November 1969, Public Law 91-121, Section 203. The amendment is correctly placed.
  • Line 134: “the period from roughly 1974 to 1980 is called the first AI winter”. Confirmed (standard history).
  • Line 144: Eulogized in U.S. House of Representatives. Confirmed (Rosenblatt.md line 130).
  • Line 156: Hopfield 1982 recurrent network revival; Rumelhart, Hinton, Williams 1986 Nature paper “Learning representations by back-propagating errors”. Confirmed (in References).
  • Line 160: Backprop earlier - “Paul Werbos in 1974 and Seppo Linnainmaa in 1970”. Confirmed by standard AI history. Werbos PhD thesis 1974 (Harvard); Linnainmaa MSc thesis 1970 (Helsinki). Both correct.
  • Line 164: AlexNet 2012, Alex Krizhevsky, Ilya Sutskever, Geoffrey Hinton, eight layers. Confirmed (in References; external search confirmed).
  • Line 164: ImageNet results: “15.3% error versus the runner-up’s 26.2%”. Confirmed by standard reference.
  • Line 164: Yann LeCun “an unequivocal turning point in the history of computer vision”. Footnoted to LeCun 2019 NYT interview [^7].
  • Line 166: “AlexNet was trained on two Nvidia GPUs in Krizhevsky’s bedroom at his parents’ house.” Confirmed by external search (Wikipedia AlexNet, IEEE Spectrum). The two GPUs were Nvidia GTX 580.
  • Line 170: “He was just sixty years too early.” Cornell Chronicle 2019 quote. Confirmed (Rosenblatt.md line 140).

Issues / nits:

  • Line 60-66: The Mark I bullet list presents the figures (400 photocells, 512 association units, 8 response units) as if they were the canonical Mark I configuration. Wikipedia’s Mark I Perceptron article notes that “Different versions of the Mark I used different numbers of units in each of the layers”. The 400/512/8 configuration is the most commonly cited and is what the research file gives, so the post is fine, but a fastidious reviewer would note the variability. Optional clarification: “In its most-cited configuration, the Mark I had…” Not urgent.
  • Line 68: “could classify the letters X and E with 100% accuracy after just 20 training examples”. The Wikipedia article says “100% accuracy when trained with only 20 images” but adds the rotation-up-to-30-degrees qualifier. Perceptron.md line 89 confirms “20 examples (10 per class)” with “positional variation and rotation up to 30 degrees”. Post is consistent.
  • Line 116: “He had already addressed multilayer networks in his 1962 book Principles of Neurodynamics, which covered four-layer perceptrons with adjustable weights.” - Rosenblatt.md says “four-layer perceptrons with adjustable weights” is consistent with the 1962 book. Confirmed.
  • Line 144: “The exact circumstances are not well documented.” This is a careful hedge that matches Rosenblatt.md line 128.

Language

  • Line 11 (excerpt): “His high school classmate called it a dead end. He drowned on his 43rd birthday. Sixty years later, the world discovered he was right.” - strong four-beat opener. Keep.
  • Line 17: “What if the machine could teach itself?” - good cliffhanger.
  • Line 19/35: “And there, on an IBM 704, he built the first machine that could learn.” Standalone paragraph. Earned. Keep.
  • Line 50: “The perceptron wrote its own rules.” - punchy thesis. Keep.
  • Line 80: “the perceptron to be ‘the embryo of an electronic computer that the Navy expects will be able to walk, talk, see, write, reproduce itself and be conscious of its existence.’” - the post wisely lets the NYT line do its own damage.
  • Line 86: “The Cold War was on. Sputnik had launched nine months earlier.” - good context.
  • Line 100: “Not logic but learning. Not programming but experience. Not symbols but connections.” - parallel structure. Keep.
  • Line 112-114: “But the damage was already done. The book became a weapon.” - strong, defensible.
  • Line 134-136: “By 1974, funding for AI – especially neural networks – was nearly impossible to find. The period from roughly 1974 to 1980 is called the first AI winter. The perceptron was buried. And the man who invented it was already dead.” - chronological, blunt, earns its drama.
  • Line 142-148: Sailing/Chesapeake/birthday paragraph - factual, restrained. Keep.
  • Line 168: “Every time a neural network adjusts a weight during training, it is doing exactly what the Mark I Perceptron did with its electric motors in 1960.” - good closing thesis.
  • Line 170: “Frank Rosenblatt was right. He was just sixty years too early.” - earned.
  • Line 184: “The quiet psychologist with the grand piano and the telescope had been right all along.” - earned. Keep.
  • Line 186: “The machine that learned was the future. It just took the rest of the world sixty years to catch up.” - strong closer.

Forward-looking

  • None in this post. The “Previously in this series” footer is backward-looking and fine. Good.

Character

  • Em dashes: -- consistent. No —.
  • Curly quotes: None found.
  • Ellipses: None found.
  • Number formatting: Mostly consistent. Line 68 has “99.8%”, “100%”, “20 training examples” - all standard. Line 164 has “15.3%”, “26.2%” - all standard.

Verdict

PASS. No factual errors. The only minor nit is that the Mark I specifications in the bullet list could be slightly hedged (“most-cited configuration”) to match Wikipedia’s note about version variability, but the post’s numbers are the canonical ones. All quotes traced. The post lands its emotional beat without overclaiming. This is the cleanest of the four.

CROSS-CUTTING OBSERVATIONS

Forward-looking promise hygiene

Post 2 has two explicit forward-looking promises (lines 123 and 175). The user’s feedback_no_forward_promises.md rule (corrected 2026-04-22) says posts should not tease upcoming entries. These two lines were written before the rule was established (post date 2026-04-04). A choice for the user:

  • (a) leave them, since the rule was added after the post was published, or
  • (b) retroactively edit them out, since the user has done this for other posts.

If the policy is to keep the published archive consistent with current rules, the two lines should be cut.

Posts 1, 3, 4 are clean on forward-looking. Good.

Quote attribution status

QUOTE_AUDIT.md (2026-04-19) flagged dense quote-without-citation issues in:

  • Post 1 (Sweden): 7 flagged items, mostly traceable to Persson 2005 / Lundin 2006 (both in References).
  • Post 2 (Beurling): 16 flagged items - the largest cluster in the series. All traceable to Ahlfors-Carleson 1988 / Beckman 2002 (both in References) plus the Malliavin tribute (not in References).
  • Post 3 (Blueprint): 5 flagged items, traceable to Gruenberger 1968 / Dyson 2012 / Malinovsky 2010 (all in References, though “Ware 1959 memo” should be added).
  • Post 4 (Perceptron): 5 flagged items, traceable to Cornell eCommons memorial / Olazaran 1996 / Crevier 1993 / specific LeCun NYT interview (most in References).

None of these constitute a legal risk. The user’s feedback_writing_style.md preference for full-sentence attributions (“as X said in Y (YYYY)”) is unevenly applied across the batch. Post 4 is best (every quote is footnoted to a numbered source); posts 1-3 are looser but still defensible.

The user’s writing style memory says recurring characters should be cross-linked. Check:

  • Rossby (post 1) - cross-links to Bergen School and Charney’s ENIAC forecast and Phillips’ climate. Good.
  • Beurling (posts 1 and 2) - post 1 mentions Beurling without a self-link; post 2 cross-links back to post 1. Asymmetric but acceptable since Beurling is introduced as a side-character in post 1 and as the protagonist in post 2.
  • Phillips (post 1, 3, 4) - cross-linked from posts 1 and 4 to “first climate model” (post 30). Good.
  • Charney (posts 1, 4) - cross-linked from post 1 to “tamed the equations” (post 29). Good.
  • Lorenz (post 1) - cross-linked to “butterfly” (post 31). Good.
  • Richardson (post 1) - cross-linked to “pile of hay” / “pencil” (post 24). Good.
  • Klara von Neumann (post 3) - mentioned without cross-link to post 18 (the Women post). The Women post (2026-04-18) is after post 3 in publication date, so no link possible at the time of writing post 3, and adding one retroactively would be a reasonable polish. Optional.
  • Mary Tsingou (post 3) - mentioned without cross-link. Same chronological constraint.
  • Norman Phillips (post 1) - cross-linked to “first climate model”. Good.
  • Bert Bolin (post 1) - introduced as a key character. The IPCC and Crutzen Nobel connections in post 1 are striking but Bolin only re-appears centrally in post 17 (“decade got good”) and post 13 (“forecast that reached the Nobel”). Post 1 doesn’t link forward to those - and shouldn’t, per the no-forward-promises rule.

Number formatting

Per the user’s memory (feedback_formatting.md): no comma thousands separator, use scientific notation. The four posts comply: spaces or no separator. No commas in numbers found.

Em dashes / curly quotes / ellipses

All four posts use only -- for em-dashes (good per repository convention) and straight ASCII quotes. No ellipses found. No Unicode em-dash, en-dash, or curly quote contamination. Good.

  1. POST 2, line 55-56 (CRITICAL): Fix the keyspace claim. The 10^18 figure is the total wheel-setting keyspace for the early T52 Beurling broke; it is not “per wheel setting”. The 10^27 figure belongs to the later T52d/T52e models, which Beurling did not break.
    • Suggested replacement: “It used ten cipher wheels (not three), each with a different coprime number of pins (from 47 to 73), creating a total keyspace of approximately 893 quadrillion – close to 10^18 – combinations. The encryption operated in two stages: first an XOR substitution, then a conditional transposition of bit pairs controlled by a plugboard. This two-stage process meant that even knowing the plaintext didn’t directly reveal the key. Later T52d/T52e variants (introduced from 1942) added irregular wheel stepping that pushed the keyspace to ~10^27 – and Beurling never tackled those; Bletchley Park failed to break them either.”
  2. POST 2, lines 121-123 and line 175 (HIGH): Drop the forward-looking promises about the Polish codebreakers. The user’s feedback_no_forward_promises.md rule says don’t tease upcoming posts.
    • Cut lines 121-123 entirely, or rewrite as backward-looking.
    • Cut line 175 entirely. The post ends well at the “Some magicians are forgotten…” paragraph.
  3. POST 1, line 24 (LOW): Soften “a barotropic model developed at Stockholm University” to acknowledge the American origin (Phillips ported the ENIAC code to BESK).
    • Suggested: “…running a barotropic model – adapted from the ENIAC forecast code – on a computer called BESK…”
  4. POST 1, line 237 (LOW): Soften the Lorenz “two-week wall” closing flourish - the Swedes did not “hit” the deterministic chaos limit in 1954; that’s a property of the atmosphere, not their computation.
    • Suggested: “The forecast horizon that Lorenz would later prove existed? The Swedes were already running into it before anyone else had a stable production system.”
  5. POST 3, line 232-259 (LOW): Acknowledge the BESM-as-clone debate one more time. The current “BESM was not a clone. It was a parallel invention.” is more confident than the historical record.
    • Suggested addition (one sentence): “Wikipedia still lists BESM among IAS-family machines; historians of Soviet computing emphatically do not.”

  6. POST 4, line 60-66 (OPTIONAL, COSMETIC): Add “in its most-cited configuration” or similar before the Mark I bullet list, to acknowledge that Wikipedia notes versions varied. Not urgent.

  7. All posts (STANDING HYGIENE): Apply the QUOTE_AUDIT.md fixes incrementally - especially the Beurling post’s quote density. A single sentence at the head of each References section (“Quotes from X are drawn from Y unless otherwise noted”) would resolve most of the concern.

VERDICT (BATCH)

Post Verdict
1 (Sweden) PASS. Very clean. Two minor language softenings recommended.
2 (Beurling) PASS WITH FIXES. Keyspace error needs correcting. Two forward-looking promises should be cut.
3 (Blueprint) PASS. Solid. One optional clarification on BESM.
4 (Perceptron) PASS. The cleanest of the four.

Overall: a strong batch. Post 2 has the most issues but they are all mechanically fixable. The factual base across all four posts is well-aligned with the research files - this is what good fact-checking habit looks like.

Reviewer recommends: apply the priority-1 and priority-2 fixes to Post 2; the rest are polish.