Why Seven Hills

Correction — 12 September 2026

2026-09-12T09:00:00Z

Every quotation, date and attribution on this page was checked against a primary source rather than against the sweep that found it. The dates hold: Hutchinson's Durham volume three is 1794 (the Archive dates the scan to the set's 1785); Walford's Scientific Tourist is 1818; Barrett's Bristol 1789; Smith's Morley 1876, and its “upwards of 12,000 souls” is its own 1876 population. Orwell's line is verbatim from the diary of 3 March 1936. Platner and Ashby say what they are quoted as saying, and the count of seventeen distinct names across the four lists, with only the Palatine on all four, comes out exactly right when the names are counted by hand — Lydus gives TWO lists, which is where the seventeenth comes from. Two things needed fixing. The Nicomachus quotation is real but comes at third hand from W. Wynn Westcott, a Victorian occultist, who has it as Minerva rather than Athena; the page now quotes the scholarly text and names Westcott for the handsomer wording. And the Fornham blockquote was dated 1739 two lines above a sentence saying the carriers run from 1781 — both true, and the page never said why: the wording is Blomefield's, and the corpus's earliest carrier is Mostyn John Armstrong copying him forty years later. It says so now.

Revised — 12 September 2026

2026-09-12T09:00:00Z

The claim that Sheffield's own 1879 guidebook never reaches for Rome is the one NEGATIVE result on this page, and a negative cannot appear in either sweep's output, because both record hits. So it was asserted in prose, in a paper whose whole method is not doing that. Counted: in 160,432 words the phrase appears not once, and the three mentions of Rome are the Papacy, a marble statue and a masonic order. The counts are committed and a test fails if they stop being true.

Correction — 12 September 2026

2026-09-12T09:00:00Z

§2 said “Hills lead, and the runner-up gives the game away” and named oaks as the runner-up. The table directly above it is sorted by ratio, as the prose says, and puts windmills second on THREE occurrences; oaks is third. On the raw count oaks leads hills outright, 440 to 291. Neither ordering is stable either — the Archive caps each query at 1,600 and ranks by relevance, so the subset changes between runs and so can the order, which is exactly how a typed name goes stale. The row is now read from the data, and windmills is named as the thin row it is rather than passed over in silence: a ratio on three occurrences is arithmetic, not evidence.

Correction — 12 September 2026

2026-09-12T09:00:00Z

The composition of the 200 prose hits summed to 199. Rome 72, abroad 34, barrows 29 with 6 more as a bare place-name, civic claims 20, unmatched 38 — and one hit classified as scripture, computed by the sweep, counted in the total, and never rendered by the page. It is Laodicea, “built upon six or seven hills”, out of a scripture gazetteer. The sentence is now a loop over the categories, so it cannot fail to add up again.

Correction — 12 September 2026

2026-09-12T09:00:00Z

§5 said the count “cannot answer” whether the phrase was ever attached to Sheffield, “because it records only numbers and nouns, never the sentences they sat in”. That stopped being true when the sweep began keeping concordances, and §2 two screens earlier says so. It can answer, and the answer is stronger than the hedge: Sheffield is named in none of the 291 “seven hills” the sweep captured, and the towns that do make the claim are Bristol, Durham, Edinburgh, Norwich and Bath. The method section is scoped too — concordances are kept for that one phrase and no other row, which is what the attestation table cannot be built from.

Revised — 12 September 2026

2026-09-12T09:00:00Z

§6's conclusion read “the largest prime you can hold in your hand”, which contradicts the mechanism the same section runs on: one hand holds five, and “five hills” is the claim it says can be settled in a second. The argument needs ten fingers and its own closing line says ten fingers. And the paragraph about the unplaceable hits went on to produce Lisbon, Prague, Constantinople and Providence — all four of which the sentence above had just counted as placed, abroad. It is about the 34 abroad, and now says so.

Revised — 12 September 2026

2026-09-12T09:00:00Z

This log starts here, four days late. The entries below were reconstructed from the commits after a re-read pointed out that the page had been revising in public with no record of it.

Correction — 11 September 2026

2026-09-11T09:00:00Z

§4 said the Fornham wording “survives verbatim through 1805, 1831, 1843, 1855 and 1873”. The corpus supports none of that list as stated: its carriers are 1781, 1805, 1832, 1843 and one undated 1873 text, so two of the five years have no book behind them, two carriers were missing, and the prose is demonstrably not verbatim between any two — the Flemings are put to the sword in one and routed in another. What survives is the name arriving with its own correction attached, which is the point the years were standing in for, and that is now computed from the concordances rather than typed. The sweep gained --dates to make it computable.

Correction — 11 September 2026

2026-09-11T09:00:00Z

The page led on a departure ratio of 105.3x. Its numerator is solid and its denominator is a two-step extrapolation from four fitted points, which leaves two degrees of freedom; applied consistently the same statistic calls ten 2,189x and puts windmills, on three occurrences, in second place. It is no longer the headline. The page leads on the raw contrast and on the per-TEXT counts, which one talkative book cannot inflate — seven appears in 150 books against 25 at six and 23 at eight — and the interval on the prediction is published beside it.

Correction — 11 September 2026

2026-09-11T09:00:00Z

Blomefield is 1739, not 1781. Durham's quote is in the volume the Archive dates to the set's 1785 but whose preface puts after January 1790. And Barrett's 1789 Bristol makes Bristol the earliest English claim rather than Durham, which moves Sheffield from 151 years behind to 147. Barrett was missed because the sweep could not read “feven hills”: eighteenth-century OCR renders the long s as an f far more often than as the character itself. Line-break hyphens were being missed too.

Correction — 11 September 2026

2026-09-11T09:00:00Z

The method section said the page “reads nothing else” than sevens.json while printing an attestation table the sweep cannot produce — it records numbers and nouns, never sentences. It says so now, and says the rows were found by hand.

Published — 11 September 2026

2026-09-11T09:00:00Z

Published.