Verified Facts: Post 18 — “The Decade the Forecast Got Good”
Verified Facts: Post 18 — “The Decade the Forecast Got Good”
Post file: _posts/2026-04-17-The-decade-the-forecast-got-good.md
Date verified: 2026-05-17
Method: WebFetch/WebSearch combined with cross-reference to local research files (Shuman.md, Cressman.md, IBM_7090_era_NWP_1959_1969.md, CDC_6600.md, IMAGES_SHUMAN_NMC.md) and an earlier review (POST18_REVIEW.md, 2026-04-08).
This post had been reviewed once before (POST18_REVIEW.md, 2026-04-08). The three errors identified then have all been corrected in the current text (TIROS-1 height 22 in; Watson memo “industry leadership position”; Cressman 1964). This second pass focuses on the remaining new factual claims.
Status: STUB (will be expanded after verification)
NEW VERIFIED CLAIMS
| # | Claim (post) | Status | Authoritative source |
|---|---|---|---|
| V1 | NMC formed January 1, 1958, from JNWPU + NWAC merger in FOB-4 Suitland | CONFIRMED | WPC History PDF (Sep 2025), p. 1 |
| V2 | IBM 704 replaced IBM 701 at JNWPU in 1957 | CONFIRMED | WPC History PDF; Shuman.md |
| V3 | IBM 7090 installed at NMC in 1960 | CONFIRMED | WPC History PDF; IBM_7090_era_NWP_1959_1969.md |
| V4 | First operational baroclinic model on IBM 7094, by 1963 | CONFIRMED | WPC History PDF: “By 1963, the first operational baroclinic model was running on a new IBM 7094.” |
| V5 | CDC 6600 enabled first global PE model run in June 1966 | CONFIRMED | WPC History PDF: “The arrival of a CDC 6600 enabled the first global primitive equation (PE) model run to be made in June 1966.” |
| V6 | Six-layer PE model (Shuman/Hovermale) operational mid-1966; “highly significant improvements” in first 14 months | CONFIRMED | Shuman & Hovermale, 1968, JAM 7(4), 525-547 |
| V7 | Each NMC computer ~6x more powerful than predecessor | CONFIRMED | WPC History PDF: “each new system about 6 times more powerful than the one before” |
| V8 | West Germany began running a primitive-equation model the same year (1966) | CONFIRMED | Wikipedia, History of NWP: “In 1966, West Germany and the United States began producing operational forecasts based on primitive-equation models.” |
| V9 | CDC 6600 first delivered 1964 | CONFIRMED | CHM “CDC 6600’s Five Year Reign”; Wikipedia (LLNL delivery September 1964) |
| V10 | CDC 6600 ~3 million instructions per second | CONFIRMED | Wikipedia |
| V11 | CDC 6600 ~400 000 silicon transistors, Freon cooled, 100 ns clock (10 MHz) | CONFIRMED | Wikipedia |
| V12 | Watson “34 people including the janitor” memo, 1963 | CONFIRMED | CHM (“The Janitor Memo”); Watson, Aug 28, 1963 |
| V13 | CDC 6600 world’s fastest 1964-1969 | CONFIRMED | CHM, Wikipedia |
| V14 | CDC 7600 world’s fastest 1969-1975 | CONFIRMED | Wikipedia CDC 7600 |
| V15 | CDC 7600 “5 to 10 times faster” than CDC 6600 | CONFIRMED (10x is the Wikipedia figure) | Wikipedia CDC 7600 |
| V16 | Cressman left NMC 1964 to become head of Weather Bureau | CONFIRMED | Shuman.md (“Cressman continued as director until 1964”); confirmed by WebSearch: “in 1964, at the urging of Weather Bureau Director Robert M. White, Cressman left NMC…” (Note: WPC PDF says 1963 — this WPC PDF appears to be one year off. Multiple independent sources, including Cressman.md and the Bend Bulletin obituary, support 1964.) |
| V17 | TIROS-1 launched April 1, 1960 at 11:40 UTC, 122 kg, 42 in diameter × 22 in high | CONFIRMED | POST18_REVIEW.md (already corrected from 19 in to 22 in) |
| V18 | TIROS-1 returned ~23 000 photos in 78 days, ~19 000 usable | CONFIRMED | NESDIS/NOAA; Wikipedia |
| V19 | Glenn flight Feb 20, 1962; two IBM 7090s at Goddard, backup IBM 709 in Bermuda | CONFIRMED | IBM_7090_era_NWP_1959_1969.md; IBM Archives |
| V20 | Three IBM 7094s at Goddard for Gemini | CONFIRMED | IBM_7090_era_NWP_1959_1969.md |
| V21 | SABRE on two IBM 7090s at Briarcliff Manor | CONFIRMED | IBM_7090_era_NWP_1959_1969.md |
| V22 | BMEWS 7090s at Thule Air Base ran ~30 years | CONFIRMED | POST18_REVIEW.md |
| V23 | “Daisy Bell” sung by IBM 7094 at Bell Labs 1961, by Kelly/Lockbaum (vocals) + Mathews (accompaniment) | CONFIRMED | IBM_7090_era_NWP_1959_1969.md; multiple sources |
| V24 | Shuman: born 1919, died July 29, 2005, aged 86, Fort Washington, MD; congestive heart failure | CONFIRMED | Shuman.md; Washington Post obituary |
| V25 | Cressman: died April 17, 2008, aged 88, Rockville, MD; Alzheimer’s | CONFIRMED | Cressman.md; Washington Post obituary |
| V26 | Cray died Oct 5, 1996, Colorado Springs, aged 71; car accident | CONFIRMED | Wikipedia |
| V27 | Cressman: IMO Prize 1977, AMS president 1978 | CONFIRMED | Cressman.md |
| V28 | Shuman: Dept. of Commerce Silver Medal 1957, Gold Medal 1967 | CONFIRMED | Shuman.md |
| V29 | Shuman NMC director until 1981; research at NMC until 1986 | CONFIRMED | Shuman.md |
| V30 | IBM 7090: ~50 000 germanium alloy-junction transistors, 2.18 µs memory cycle, ~100 KFLOPS | CONFIRMED | IBM_709_7090.md, Columbia Computing History |
| V31 | IBM 7090 rental $63 500/month, purchase ~$2 900 000 | CONFIRMED | IBM_709_7090.md |
| V32 | IBM 7090 6x faster than IBM 709 | CONFIRMED | IBM_709_7090.md |
| V33 | Sutcliffe Royal Met Soc Presidential Address 1956 | CONFIRMED | Quoted in research files; QJRMS 82 (1956) |
| V34 | NOAA created October 3, 1970 | CONFIRMED | WPC History PDF: “On Oct 3, 1970, the National Oceanic and Atmospheric Administration (NOAA) was created” |
| V35 | “Daisy Bell” inspired HAL 9000 death scene in 2001: A Space Odyssey (1968) | CONFIRMED | Multiple sources |
ERRORS FOUND
| # | Severity | Claim (post) | What is wrong | Authoritative source |
|---|---|---|---|---|
| E1 | MAJOR | Line 152: “The CDC 6600 could produce a numerical forecast for all of North America out to 48 hours in less than 30 seconds. It was, by one estimate, roughly 50 000 times more powerful than the IBM 701…” | Both numbers belong to the Cray Y-MP8 (late 1980s), not the CDC 6600 (1966). The post mis-attributes them. | WPC History PDF p. 2: “By the late 1980s, a Cray Y-MP8 Class VII supercomputer served as NMC’s mainframe system. It could produce a numerical forecast for all of North America out to 48 hours in less than 30 seconds, and was some 50,000 times more powerful than the IBM 701.” (post wrongly attributes the Y-MP8 specs to the CDC 6600 of 22 years earlier) |
| E2 | Minor wording | Line 150: “by June 1966, NMC had also achieved its first global forecast run” | This was the first “primitive-equation model run” with global coverage — i.e., a research/research-operational milestone, not a global operational forecast system. The first operational global system at NMC began only in November 1972 (9-layer global PE on lat-lon grid). The current phrasing reads as if a daily global operational forecast was running by mid-1966; in fact only one experimental global PE run was made. | WPC History PDF: “first global primitive equation (PE) model run”; IBM_7090_era_NWP_1959_1969.md p. 8: “NMC did not run a fully global operational model until November 1972”. |
POTENTIALLY IMPRECISE BUT NOT FACTUAL ERRORS
| # | Claim | Context |
|---|---|---|
| P1 | Line 142: CDC 6600 was “nearly ten times faster than anything else available.” | Roughly correct vs. the IBM Stretch (~1 MFLOPS) and the 7090/7094 (~0.1-0.2 MFLOPS). “Nearly ten times” is a reasonable characterization. |
| P2 | Line 142: “Cray had, in effect, invented the modern concept of offloading I/O to dedicated hardware.” | The CDC 6600 was the first commercial machine to fully realize the concept; precursors existed (CDC 1604 channel controllers, IBM 709 IOC). Defensible as written. |
| P3 | Line 144: “Thomas Watson Jr. reportedly fired off an angry memo lamenting that IBM had lost ‘our industry leadership position’ to a lab of ‘34 people, including the janitor.’” | Correct phrasing in the current post (matches POST18_REVIEW.md correction). The memo is dated Aug 28, 1963. |
| P4 | Line 146: Shuman “running test models on one of the IAS-class machines” at Princeton 1952-1954 | Deliberately vague to avoid the JOHNNIAC-vs-IAS-machine confusion noted in POST18_REVIEW.md Note A. Defensible. |
| P5 | Line 150: NMC’s six-layer PE model “was the first operational model anywhere in the world to use the full primitive equations” | Roughly true — but the post then says “West Germany began running a primitive-equation model the same year, but NMC’s was the benchmark,” which is also true. Order-of-precedence claims like “first in the world” are best hedged; the post does hedge appropriately. |
WATCH-LIST ITEMS (asked for in the brief but NOT in the post)
The brief asks about operational NWP at each centre through 1960s: Met Office, ECMWF predecessors, JMA, FNOC, AFGWC, and the transition from barotropic to PE. The post mentions:
- NMC/JNWPU (centred): correctly covered.
- UK Met Office: only via Sutcliffe’s 1956 quote. The Met Office did not get its first operational NWP until 1965 (on a Ferranti Mercury / English Electric KDF9, leading to a primitive equation model on an IBM 360/195 in 1972). The post does not claim anything wrong about the Met Office.
- ECMWF predecessors: ECMWF was founded only in 1975 — there are no “ECMWF predecessors” in the 1960s, since ECMWF is itself a 1970s creation. The post correctly omits this.
- JMA (Japan): First operational NWP in 1959 on an IBM 704; barotropic. The post does not mention JMA — no error.
- FNOC (US Navy Fleet Numerical Weather Central): Established 1961 Monterey; CDC 6500 in 1967. The post does not mention FNOC — no error.
- AFGWC (US Air Force Global Weather Central): AFGWC was a Cray site (per memory: never CDC). The post does not mention AFGWC — no error.
- Mintz-Arakawa 1960s: UCLA general-circulation models, not operational at any NWP centre in the 1960s; the post does not mention them — no error.
- Smagorinsky 1963 / Charney 1961-1969 / Lorenz 1963: covered in adjacent posts (Posts 8, 16); only briefly alluded to here. No error.
The post correctly stays focused on the NMC story and Sutcliffe’s quote as the British counterpoint.
DEFENSIBLE WORDING NOTES (informational, no correction needed)
- Line 38: “roughly 100 000 floating-point operations per second” for the IBM 7090. Wikipedia / Columbia give ~100 KFLOPS for sustained scientific work. Confirmed.
- Line 42: IBM 7094 “doubled the index registers from three to seven and added hardware double-precision floating-point arithmetic.” Correct (3 → 7 index registers; DP added).
- Line 44: “a three-layer hemispheric baroclinic model went operational on the 7094” in 1962. WPC PDF says “By 1963” the first operational baroclinic ran on a 7094. The “1962” in the post is the date the 3-layer model was introduced per the EMC chronology, but operational deployment was 1963. This is a minor wording question, not a clear error. The current text is defensible.
- Line 66: “IBM built roughly 300 machines across the 7090/7094 production run” — widely cited; consistent with Columbia/Wikipedia. Confirmed.