The Seven Hills of Sheffield

Correction — 12 September 2026

2026-09-12T09:00:00Z

The hill database was re-downloaded rather than re-read from the cache, and the download was v18.6. The sweep had “v18.5” typed into it beside a URL that serves whatever is current, and it writes that string into the file this paper prints as its source. So the paper was one re-run away from citing v18.5 while displaying v18.6, and nothing could have caught it: a warm cache never fetches the new file, and the suite proves the prose matches the data — which it did, on a wrong citation. The version is read out of the download now, and an unlabelled download stops the run rather than inheriting the last version that worked. What v18.6 actually changed: twelve of 361 authorities, none of them named anywhere in this paper, and not one published figure. The Geoff is still 40 m, Sheffield still seven of sixteen, Bradford seven, Barnsley seven of eighteen, and §2's margin figures are identical. One summit in the header moved — Moortop, outside the city, whose drop v18.6 revises from 58.0 to 56.8 m.

Revised — 12 September 2026

2026-09-12T09:00:00Z

§10's Rome now has the committed derivation it has been missing. The summit list was produced on 9 September by a sweep whose output was never checked in — for a paper that answers everything with “computed, not asserted”, the largest thing it could not show. Re-run from the cached contour sheets in the published window: 22 summits over 25 m inside the ISTAT boundary against the published 24, four over 40 against five, the same places to within metres, and drops running low by exactly the amount a 5 m contour surface quantised and then corrected upward to spot heights should run low. The run is recorded in romeReproduction.json with a test over it, so the next time the published file moves without a re-sweep, something says so. One curiosity worth keeping: the 100 m summit on the Baccano rim falls INSIDE a plain ray-cast of the boundary and outside ISTAT's own answer by 3.4 m, which is note 6's entire subject arriving unprompted in the verification.

Correction — 12 September 2026

2026-09-12T09:00:00Z

Every external reference on this paper was checked against a primary source, because nothing in the test suite can: the suite proves the prose matches the committed data, and says nothing about whether a quotation is real. Almost all of it held — both DOIs resolve to the papers named and to those authors, Chaucer's “Sheffeld thwitel” is Reeve's Tale line 3933, and the Rome sources are as quoted. One sentence did not. §11 had Ammerman's cores “found occupation surfaces between 3.2 and 6 m — Bronze Age ground”. Ammerman calls them dark anthropic soil horizons and says in his own supplementary material that whether they are in place “should be regarded as an open question”, with “a good chance that most of the seeds … have experienced some degree of displacement”. The elevations are his; the firmer nouns were this paper's, which is the Ecclesall Woods mistake in a citation. Two figures beside them — a deep sounding “already 7.6 m above the sea” and a water table “about 10” — are not in the open material at all: the only 7.6 there belongs to a different core. All three are gone. The depth is what the argument uses, and the depth is his.

Correction — 12 September 2026

2026-09-12T09:00:00Z

A full re-read against the committed data, and four of the paper's sentences do not survive it — all four in prose written during yesterday's corrections, and all four describing an instrument instead of reading it. §12 said “every drop above is measured down from a summit the survey puts lower”. The audit says the survey puts three of those seven HIGHER, including High Neb, which the very next sentence used as the open-ground control: the paragraph contradicted itself across two sentences. Three are read high, three are read low and Wincobank is read exactly. Note 7 said Frecheville and Waverley fail the floor “on either instrument”; on the survey those two stand 11.64 m apart and are not that city's seventh and eighth at all — Burbage Moor and Banner Cross are, by 0.63 m. The claim survives in a better form: the finer instrument did not break the 1967 tie, it moved it to two other hills. Note 7 also called a metre of floor “generous” when on open moor the two instruments part by 1.26 m of drop at High Neb, which is more than the floor — it is a minimum the data barely licenses. And §12 said two rises the 30 m grid could not resolve come in: one does, Banner Cross. The other is the Norton ridge with its summit moved 843 m along itself, and Burbage Moor was the radar's own first-one-out, promoted rather than discovered. All four are now computed in geoffInstruments.js from the two committed sweeps. The withdrawal of the 1936 unit is untouched by any of it.

Correction — 12 September 2026

2026-09-12T09:00:00Z

§11 said the Rome inside the Aurelian Walls has three Geoffs everywhere in the 1936 range, and named two of them. The third is Monte Testaccio, the heap of broken amphorae this paper's own ground rule says you do not count and which §11 excludes by name a dozen lines later. The one count where a third hill flattered the argument was the one count that did not say what the third hill was; it does now, and the count is also marked as the lower bound it is, because the Rome sweep emitted from 25 m and this bar sits under that. Note 6 quoted “the last hill to make Rome's count at all stands only 87 metres inside” — a figure computed at §4's scenario bar of 34 m, not at the 40 m the paper counts at, where the last hill in stands 1,086 m clear and the boundary decides nothing. Both bars are now named. Note 5 had Platner and Ashby writing “two years ahead of the diggers” about a levelling the same note dates to 1932. §13 said the traditional hills “clear every bar that city could have had” when two of the seven do and five clear none. §5 put Margery Hill four miles from Crookes; it is ten, and the distance is gone rather than corrected, because it was carrying nothing. §2's “earliest Sheffield anyone has found in English” is scoped back to English literature, which is the claim that was defended. §7 had a stranded “that” and pointed at the wrong table.

Correction — 12 September 2026

2026-09-12T09:00:00Z

Figure 2's alt text said the counting function “falls from sixteen to none as the required drop rises from zero to 108 metres”. At 108 m the count is ONE — High Neb, which is what puts the right-hand end of the axis where it is. A screen-reader user was given a chart ending in nothing where a sighted reader sees a final step of one. Both ends are computed from the curve now. In the same pass: one “bare-earth survey” appears on this page at three resolutions — 10 m in note 3, 5 in §12, 1 in §4 — with nothing connecting them, and §12 now says they are the same source read at different grids and why. And §9's Bristol counts at 21 and 22 m, and the twenty-two they give Sheffield, are the only figures on the page not read from a committed file: that sweep's output was never checked in, so nothing re-checks them. The note says so rather than leaving them to look like everything around them.

Revised — 12 September 2026

2026-09-12T09:00:00Z

The companion page at /seven/ gets the revision log it should have had on the day it shipped. It took three corrections that first evening — the headline ratio demoted, a five-year attestation list withdrawn, Blomefield redated — with hourly CI deploying each intermediate state and no record anywhere that a reader could check. This log records only what touches THIS paper, which is what the omission turned on. /seven/ was also missing from the sitemap, having been linked from here since it shipped and published from nothing.

Revised — 12 September 2026

2026-09-12T09:00:00Z

Housekeeping the log owed itself. Three of yesterday's four entries retract something and none of them carried the correction badge, which this file's own header promises. They carry it now. The 10 September entry closing “§7's tie is not contingent after all” has been walked back by note 4 since, and is marked superseded rather than left to be read as current.

Correction — 11 September 2026

2026-09-11T09:00:00Z

Correction to the entry below, which went live before this was caught. It said Beauchief left the seven “because its summit was the canopy of Ecclesall Woods”. The measurement was right and the explanation was invented: the point GLO-30 calls Beauchief's summit is on Dalewood Avenue, a residential street with twenty-nine buildings inside 150 m and one wood, and it is not in Ecclesall Woods at all. What is measurable is that its 161.6 m stands 7.0 m above the highest ground the bare-earth survey finds within 275 m, and 25 m above the ground in its own cell, which spans 128 to 146 m — a slope, not a summit. Manor Top is +5.2 m by the same test and Norton Lees +9.6. The claim that the surface model reads above the ground in built suburbs survives intact, with numbers instead of a wood chosen for sounding likely. Checked against a control the same way: at Wincobank the survey returns 161.00 m against the database's 161.0.

Correction — 11 September 2026

2026-09-11T09:00:00Z

§12 withdraws the 1936 unit, and this is a correction to the paper's central method rather than to a number. The 1936 sweep ran on a radar SURFACE model, which in a city reads roof and canopy. Swept again on the Environment Agency's bare-earth survey the table does not survive: Beauchief, fifth on the old list, leaves the seven because what the radar read as its summit stands 7 m above the highest real ground near it, on a built hillside; Manor Top goes the same way by 5.2 m, and two rises a 30 m grid could not resolve come in. The seventh and eighth drops of Orwell's city then sit 0.24 m apart, where the two instruments differ by more than that on open moor. So there is no 1936 unit to state, and the paper stops stating one: 25.47 m is withdrawn. The rule is now in the sweep rather than in the prose — a gap must clear a metre before the order between two drops counts as measured — and it retires a second claim with it. Note 7 used to offer the 1967 pair as the answer §9 cannot get for Bristol: given a finer instrument, the tie broke. Those two are 0.60 m apart on one instrument and 0.63 on the other. The tie never broke, and the note now says so. §11's Rome claim survives without the withdrawn figure, because it never needed a point: every bar either instrument allows the 1936 city runs 22.80 to 25.47 m, the Capitoline and the Palatine clear all of it, and the next canonical hill down clears none of it — which a guard now checks at both ends. One thing improved: the two instruments were said to disagree about the hill defining the modern unit, the satellite putting the Norton ridge at Norton Lees where the database puts it at Backmoor. On bare earth the summit is at Backmoor, at the height §4's own audit measures there. The database was right; the radar was reading the wood.

Revised — 11 September 2026

2026-09-11T09:00:00Z

Two additions from a corpus study run alongside the paper. The 1879 Pawson and Brailsford guide does not claim seven hills — every “seven” near “hill” in its 160,000 words is a distance or a surname — but it does record a civic boast, and it is a better one: that there was no street in the town from which the country could not be seen. The guide then tests it and concedes it is no longer literally true, which is more than the seven has ever had done to it. And §11's point that Rome counts spurs of one plateau as separate hills turns out to be a distinction Rome drew itself and then ignored: the Septimontium, the archaic festival whose name means seven mounts, counted two divisions of the Palatine and three of the Esquiline, and Festus's version of its list has EIGHT names — of which Platner and Ashby, the same source this paper uses for the Velia, write that one must obviously be omitted. An antiquarian striking a name to preserve a number.

Correction — 11 September 2026

2026-09-11T09:00:00Z

§2 said the SIZE of the gap between Sheffield's seventh and eighth hills was what earned it the convention — “the margin is the contribution, not the inequality”, and “it is the size of the gap rather than its bare existence that entitles”. Measured, that is false. Against the 133 UK authorities that can make exactly seven at some threshold, the median gap is 4 m and Sheffield's two beats 26% of them; under §4's own stricter reading of Backmoor it is 0.6 m and beats 2%. The margin does not distinguish Sheffield from anybody. What is actually scarce is the gap existing at all, which is Proposition 1's failure condition and which §9's Bristol fails. So the section keeps the fact and loses the wrong half of it, and both figures are now derived in geoffScenarios.js rather than asserted — asserting being the thing that went wrong.

Revised — 10 September 2026

2026-09-10T09:00:00Z

The conclusion has lost its ending. It used to move Sheffield’s boundary past High Neb and Back Tor, drop the unit to 34 m, and observe that at 34 m Rome has seven too — then admit in the next paragraph that no such city has ever existed. A coincidence staged in a city nobody has ever lived in is not a finding, and the paper spends thirteen sections refusing to do exactly that. It is cut. What replaces it was already sitting in the paragraph underneath: Rome’s count has only ever fallen as Sheffield grew — at least twenty-four hills when Orwell wrote, at least twenty-two after 1967, five since 1974 — and seven was skipped, the unit going from 26.07 to 40 metres in a single year. At every boundary either city has actually had, the sentence is true of exactly one of them. The search for an honest version was run first and came back empty: crossing every historical Rome against every historical Geoff, an exact seven is only reachable at bars of 31, 34 or 35 m, and Sheffield has never had any of them. The near miss is recorded here because it is close enough to be worth someone re-checking: inside the Grande Raccordo Anulare, at the Geoff of 1967–74, Rome holds at least seven natural summits — the Capitoline and the Palatine among them — and the eighth thing on the list is Monte Testaccio, which is a heap of broken amphorae and which this paper’s own ground rule says you do not count. It is a floor rather than a count, because the Rome sweep stopped at 25 m; Monte Mario, 25.8 m and inside the ring, is missing from the swept list and proves it. Settling it means re-sweeping Rome at a lower floor.

Revised — 10 September 2026

2026-09-10T09:00:00Z

A four-reader review of the whole paper. What it found, and what has changed. The Ammerman citation in §11 gave figures the cited article does not contain: it puts the Sant’Omobono cores at about 7.6 m above the sea, not 11.5, and records no gravel pavement at any elevation. The sentence has been rewritten from what the paper actually says. This is the second time a number has been attributed to that source and been wrong. §12 dated Dore’s arrival to 1933 twice while the register it cites, and §12 itself three paragraphs earlier, say 1 April 1934. Ecgbert’s “first overlord of all England” is now attributed to the stone on Dore green, which is whose phrase it is; the Chronicle calls him the eighth to hold that overlordship. §2 claimed to watch a tie break under a finer instrument twice; it happens once, and §12 says so. §5 said only High Neb kept Wincobank out, when losing any of the seven would let it in. The Trajan’s Column inscription is now given as usually read rather than as read. Platner and Ashby’s “about 28 metres” has its hedge back, and the Palatine bound with it: a grave floors the ground where it was dug, 181 m from the col, which is not a floor under the col, so the ceiling is stated as conditional. Pomezia’s 1938 farmland joins Fiumicino as a loss. The 1870 row leaves the historical table as the one unsourced area in it. Two things the paper knew and had not said: §12’s tie-breaking instrument is a radar SURFACE model, the class §10 declines to read Rome off, so its decimals are precision and not accuracy; and §4’s stricter reading of the reservoir costs not the count but the table, since at 38 m Bradford holds at least eight and §6’s first place is the database convention’s. The bare-earth audit behind §4 had checked Bradford’s sixth hill; its seventh, Steeton Moor, is the drop the tables turn on, and it now takes its targets from the data and comes out ground at +0.5 m. And §2 now answers the objection it used to assert its way past: every city could mint its own unit, Bradford’s would be 41 m, and what makes this one Sheffield’s is a tradition that wants testing, a failure condition it clears where Bristol cannot, and going first.

Revised — 10 September 2026

2026-09-10T09:00:00Z

The lede's claim that 1936 is the earliest written reference was inherited from a page that honestly says “the earliest reference I have”, and every other source on the web traces back to it. So it was searched for: every digitised text on the Internet Archive catalogued to Sheffield, Hallamshire, Yorkshire or Derbyshire and published before Orwell wrote — 2,064 books, about a gigabyte of transcription. None attaches seven hills to Sheffield. Pawson and Brailsford's Illustrated Guide to Sheffield of 1879, which had every commercial reason to reach for Rome, calls the town “an amphitheatre of hills” instead. The date stands, and the lede now says “anyone has produced” rather than “known”, because books are not newspapers and OCR loses words. The same search turned up a West Riding town that did claim seven hills sixty years earlier, and it is not this one: William Smith's History and Antiquities of Morley (1876) refers to “the last of the ‘seven hills’” in quotation marks, as a saying already in circulation. The boast is portable. Also: §5's table of traditional districts against their nearest summits is now computed from the same coordinates the map is drawn from rather than typed into the page, which moved three of the seven distances by about ten metres.

Revised — 10 September 2026

2026-09-10T09:00:00Z

§11 said only pre-demolition survey mapping could say how high the ground stood over the two cols that were dug away. For the Palatine that mapping turns out to exist. Platner and Ashby's Topographical Dictionary of Ancient Rome was published in 1929, three years before the Velia was levelled, and records a grave found near the Arch of Titus at about 28 m above the sea. A grave is cut into the ground, so the ridge stood higher than that — which makes 28 a floor for the ancient col and 23.1 m a ceiling for the ancient drop, against 26.1 on today's ground. The Palatine was less prominent in antiquity, not more, by at least three metres. It changes no conclusion: 23 is no closer to 40 than 26 was. The Capitoline's col, under Trajan's cut, is still unmeasured and this paper is not going to lean on a contested inscription for it.

Revised — 10 September 2026

2026-09-10T09:00:00Z

The Chaucer claim, withdrawn earlier today as "the earliest literary mention of Sheffield anywhere", returns in a form that can be defended: the first mention in English literature. The two earlier ones are Domesday and a royal inventory of 1341, and both are Latin and both are administration — so the only thing this has to beat is English-language literature before about 1390, which is a small and well-catalogued corpus with no reason to name a minor manorial town. It would take an unnoticed Middle English poem to falsify it, and the log will say so if one turns up.

Revised — 10 September 2026

2026-09-10T09:00:00Z

§12 gains a note on how old the boundary is. Sheffield's line with Derbyshire once divided Mercia from Northumbria — the Limb Brook, the Sheaf, the Meers Brook — and both names describe it: Sheffield is the open ground at the Sheaf, from Old English sceāth, a boundary, and Dore is dor, the gate through it, where Ecgbert of Wessex took Northumbria's submission in 829 and became the first overlord of all England. Dore came to the city in the additions of 1933. A paper about boundaries deciding hills ought to mention that its own runs along the oldest one in England.

Correction — 10 September 2026

2026-09-10T09:00:00Z

The note on the name called Chaucer's line "the earliest literary mention of Sheffield anywhere". Neither source it was built from says that — the 1879 guide says only "the old poet Chaucer writes", and Macaulay says the whittles "had been mentioned by Geoffrey Chaucer". Sheffield is in Domesday in 1086 as Escafeld, three centuries earlier, and the same guide gives a "cultellum de Shefeld" in the Tower's Privy Wardrobe accounts in 1341, fifty years before the Reeve's Tale. The superlative was added by this paper and supported by nothing. It now says the town had been in the records since Domesday and its knives in a royal inventory since 1341, and that Chaucer's is the line everybody quotes — which is true, and a better fact.

Revised — 10 September 2026

2026-09-10T09:00:00Z

§2 now says where the name came from, which the paper had never explained: Geoff was a stand-in, of the kind that gets typed when a thing needs calling something and sticks for decades afterwards. Reading the 1879 Pawson & Brailsford guide to Sheffield turned up the better half of the note — the earliest literary mention of Sheffield anywhere is Geoffrey Chaucer's, a "Sheffeld thwitel" carried in a miller's stocking in the Reeve's Tale around 1390, the town already worth naming on a blade in Cambridgeshire. The unit was not named after him; it is now, in part.

Correction — 10 September 2026

2026-09-10T09:00:00Z

§12 said Beauchief came to Sheffield with the abolition of Norton Rural District on 1 April 1934, and rested a good line on it: that when Orwell wrote, one of Sheffield's seven hills had been Sheffield's for twenty-three months. Chasing the unnamed "1918 map" turned up J. G. Harston's border series, which dates every extension from 1843 to 1994 — and lists Beauchief among the additions of 1900. Sheffield City Council's own account explains why both can be true: Beauchief was split, part going to the city in 1901 and the remainder in 1933, with the civil parish abolished the April after. Which transfer took the ground this summit stands on could not be established from anything reachable, so the claim is now stated as the open question it is: six of the seven had been the city's for a generation, and the seventh for either thirty-five years or twenty-three months. The same border series dates Burbage Moor to 1928 and names the 1918 map, both of which are now cited.

Revised — 10 September 2026

2026-09-10T09:00:00Z

The rest of the review panel's list, worked through. §2's Definition 2 now reads = 7 rather than ≥ 7, so a tie empties the set the way the prose says it does instead of leaving the Geoff to outlive its own failure condition; and §2 now claims the right thing for Sheffield — not that its seventh and eighth drops differ, which any two numbers do at enough decimal places, but that they differ by two clear metres on a surveyed list. §3 no longer says Sheffield will always have seven without the condition §2 attaches to it, and §13's opening sentence is scoped to the life of the phrase, with its caveat moved up beside it instead of sitting below the licensing paragraphs. §11 states Rome's shortfall as the convention it is, and distinguishes the two annexations: Sheffield's moves the unit for everyone, Rome's can only move its own. §9 acknowledges Edinburgh, which has eighteen and runs a race over its seven. Body text is now justified, and footnotes that follow a paragraph get a rule and some air instead of being glued to it.

Revised — 10 September 2026

2026-09-10T09:00:00Z

§7 now says the thing it had been leaving out. Barnsley holds eighteen summits to Sheffield's sixteen, and the tie at the top of the English table is broken on summits — so were Barnsley a city it would be first and Sheffield second. It is not one, so it does not appear, which means the hill §7 proposes to take is being taken from the only neighbour that would win the league it is barred from entering. Derived from the summit counts and guarded, so the admission cannot go stale while the numbers move.

Correction — 10 September 2026

2026-09-10T09:00:00Z

The ranking scare added earlier today is withdrawn in full, and the answer is the opposite of the one it feared. Asked whether Bradford's and Barnsley's hills had been checked as closely as Sheffield's, they had not — so all four hills at the deciding margin were pulled from the Environment Agency's bare-earth model at one metre to the pixel. Wincobank Hill, Bradford's Chellow Heights and Barnsley's Stainborough Castle each come out within 0.8 m of their database height and below it, which is whole-metre rounding; only Backmoor comes out above, by 2.8 m, and that excess is the reservoir. So the database is not reading the tank — the corroborating sweep is, and the sweep is not what this paper quotes. The unit stands, the neighbours' hills are ground, and §7's tie is not contingent after all. §4 reports the check rather than the worry, and §7's contingency paragraph is gone with its premise.

Superseded. The last clause has since been walked back. Note 4 now says what the stricter reading costs is the TABLE rather than the count — at 38 m Bradford holds at least eight, so §6's first place among English cities belongs to the database convention and not to the ground. The check reported here stands; the reassurance drawn from it went one clause too far.

Correction — 10 September 2026

2026-09-10T09:00:00Z

Within the hour, the note added above about the covered reservoir asserted that on the stricter reading Bradford and Barnsley overtake and Sheffield is second. Tom asked whether their hills had been given the same scrutiny. They had not — the claim rested on exactly the unexamined data the note was complaining about. Checking: the database records a summit feature for most hills at the deciding margin, and where it does, the surveyors are explicit that they stood on ground; Barnsley's Hunger Hill is logged as "ground at E side of covered reservoir", which is this distinction being drawn by someone standing on the hill. But four hills at that margin carry no feature note at all — Bradford's Chellow Heights and Barnsley's Stainborough Castle, on which their eighths turn, plus Wincobank Hill and Backmoor. Backmoor is the one that was swept, and it was on a tank. The ranking claim is withdrawn: the paper cannot say whether Sheffield is still first under its own stricter rule, and §4 now says so.

Revised — 10 September 2026

2026-09-10T09:00:00Z

The paper has been holding two cities to different standards and saying so in two sections that never met. §10 filters Rome's spot heights to ground-coded points because unfiltered they put the Capitoline at 84.6 m, which is a roof; §4's own corroborating sweep reads Sheffield's seventh drop off the lid of a covered service reservoir. Rome is de-roofed and Sheffield is not, and it is Sheffield's number that decides both. §4 now states the price of the stricter reading rather than leaving it implied: the unit falls to 38 m, the seventh hill becomes Wincobank Hill — which is on the traditional list, so the tradition fares better under the stricter rule — Sheffield still has seven, and Bradford and Barnsley have at least eight apiece. Sheffield is second. §7's tie at the top of the English leaderboard is therefore contingent on the convention, and §7 now says so. The stated unit is unchanged, because the database figure is surveyed and citable and a unit derived by masking an earthwork would rest the whole leaderboard on this paper's own arithmetic.

Correction — 10 September 2026

2026-09-10T09:00:00Z

More from the same panel, all of it the paper being less careful than it tells its readers to be. §10 called Monte Mario the highest hill in the city while the paper's own data has a hill at La Storta, in the main body of the comune, 31 m higher and 11 m more prominent; it is now the highest ground overlooking the old city, which is what it is. §10's two distances to the summits the radar invented were phrased as radii containing nothing, when they are distances TO the nearest qualifying summit and rounded up — false by three metres, in the one sentence written to be careful about exactly that. §4 and §13 said High Neb sits four metres inside the boundary and Back Tor seventeen; three attempts to reproduce those gave three different pairs, so the metre precision is gone. §13's Roman counts for 1936 and 1967 are floors rather than totals, because the sweep stopped at 25 m and both bars sit just above it. §12 said the 1936 claim was unverifiable two paragraphs after computing the 1936 unit. And §12's instrument puts the Norton ridge 4 m higher than §4's database does, which is the hill that defines the modern unit — now said out loud rather than left for a reader to find.

Correction — 10 September 2026

2026-09-10T09:00:00Z

A review panel took §11's ancient argument apart, and it deserved it. The section had put figures on what Rome's hills would drop with the fill removed, citing Ammerman for a natural surface of 5.5 m in the Velabrum. That figure is a midpoint this paper invented: the cores give Roman gravel pavements at 5.96 and 6.87 m and a modern surface about 11.5, so roughly five metres of fill, not the fifteen claimed. Worse, the arithmetic was applied to the wrong kind of col. Only the Aventine's sits in a valley. The Capitoline's is at Piazza Venezia where Trajan's works cut the saddle to the Quirinal, and the Palatine's is the stump of the Velia, levelled in 1932 — both ridges taken away, so those two hills are more prominent now than in antiquity, the opposite of what the section said. The figures are withdrawn. What survives is structural and does not depend on the depth of anything: three of the seven have a col of their own and four share a saddle-ridge.

Correction — 10 September 2026

2026-09-10T09:00:00Z

Two citations and one instrument were wrong. The paper credited its relief to "the SRTM one-arcsecond model (public domain)"; every tile it actually reads comes from the Tilezen terrain build, which over Britain and Italy layers Environment Agency and EU-DEM data over SRTM and GMTED and requires both to be named. For a paper whose refrain is that the difference is the instrument, misnaming its own was the worst available error. The Luberti citation was wrong three ways — the title reads interdisciplinary not integrated, Pica was missing from the authors, and "reconstructed from 1,355 boreholes" is not their figure. ISTAT's CC BY 4.0 licence was unstated, the Environment Agency's prescribed attribution absent, and the Copernicus notice truncated and in the wrong form for derived use. All now stated.

Revised — 10 September 2026

2026-09-10T09:00:00Z

§11 now answers the ancient question, or as much of it as published data can. Rome's valleys hold up to fifteen metres of anthropogenic fill — several times the margin that decides any of this — and Ammerman's cores put the natural surface in the Velabrum at 5–6 m above the sea against a modern street above fifteen. Stand the three outer hills' cols back on that floor and the Capitoline, Palatine and Aventine each clear the bar. The other four do not move, because their cols are saddles on the plateau they share rather than valley floors, and no excavation separates a spur from its own ridge. So Rome's seven are three hills and four spurs, and were before the rubbish as well as after — which is why seven looks out of reach at any unit, not merely at today's. Stated as arithmetic rather than measurement: settling it needs the pre-urban surface reconstructed from 1,355 boreholes by Luberti and others, which is not published as a grid.

Revised — 10 September 2026

2026-09-10T09:00:00Z

The conclusion offered a world in which both cities have seven and did not say whether that world had ever existed. It has not. Sheffield only ever grew, so its unit only ever rose and Rome's count only ever fell — twenty-four hills at the 1936 bar, twenty-two after the additions of 1967, five since 1974. Seven was skipped in the step that mattered, which took the unit from 26.07 m to 40 in a single year. Going back further only lowers the bar, and a lower bar gives Rome more hills rather than fewer. To have seven, Sheffield would need the moors it gained in 1974 without the edge it has held since 1918, and no such city has existed for a day. Every unit Sheffield has actually had is now committed data, and a guard fails if any of them lands in the window that gives Rome seven.

Correction — 10 September 2026

2026-09-10T09:00:00Z

Parkwood Springs was called made ground, twice, and it is not. The summit cell stands at 172.99 m of natural ground, a hundred metres outside the nearest historic landfill boundary, and the highest point on any tip there reaches 170.32 m — 2.7 m below it. It is a hill with rubbish on its flanks, not a hill made of rubbish, and its drop is measured to ground like the rest. This matters because §4 had used it to wave through the one Sheffield reading that really is taken off a structure: Backmoor's, which is the roof of a covered service reservoir. There was no precedent, the excuse is withdrawn, and the reservoir stands on its own. Monte Testaccio in §11 is unaffected — it is made ground all the way down.

Correction — 9 September 2026

2026-09-09T09:00:00Z

Rome was recomputed from Rome's own survey — Regione Lazio's 1:5,000 mapping, five-metre contours and ground spot heights, thirteen sheets swept at ten metres — with the boundary taken from ISTAT rather than OpenStreetMap. The count is still five, but only two of the five are the same hills: the radar model this paper used before put two summits where the survey finds no summit of even thirty metres' drop within four and eight kilometres, and a third falls short of the bar once measured on the ground. §11's ending went with them. It claimed seven Geoffs inside a circle of the comune's own area, and that was an artefact of the radar: on the survey the circle holds eight, and the radius band that holds exactly seven excludes Rome's actual area. The 'right size, wrong shape' conclusion is withdrawn. What replaces it is that Rome is two hills short of seven and both of them are next door, on the Baccano rim in Campagnano di Roma — the larger being the biggest drop anywhere near the city, whose summit falls three metres outside the line.

Revised — 9 September 2026

2026-09-09T09:00:00Z

§11 was reading a moving bar as a fixed one. A Geoff is Sheffield's seventh drop, and §12 sweeps a Sheffield whose seventh drop was 25.47 m — so "no historical Rome had any Geoffs" was only ever true at today's unit. At the 1936 bar the Rome inside the Aurelian Walls had three, two of them canonical: the Capitoline and the Palatine. The tradition was describing hills its own city's bar has since risen past, which is what happened in Sheffield too, and §11 now says so.

Correction — 9 September 2026

2026-09-09T09:00:00Z

Three sentences written while fixing the entries above were themselves wrong, and are recorded here because they were not the paper's first go at these. §10 called Monte Mario "the highest hill Romans can see from the middle of Rome" — a claim about sightlines made without computing any, and false: the Alban Hills stand six times higher and are visible on a clear day. It also said the comune's high ground was unnamed, when OpenStreetMap names Monte Santa Caterina on the same hill. And §11 called a hill outside the boundary a potential ninth when at this paper's own boundary it would be an eighth. The conclusion separately implied Sheffield had been robbed of two moors in the past; §12 is clear that the moors arrived in 1974 and that High Neb was the city's before then. §4's scenario is a future loss, not a past one, and the wording now says so.

Revised — 9 September 2026

2026-09-09T09:00:00Z

§11 gains Monte Testaccio, which the survey makes the most prominent summit inside the Aurelian Walls at 33.8 m — more drop than the Capitoline, and it is a tip: a mound of broken amphorae. None of the canonical seven reaches even 27 m. The conclusion changed with it. What the paper now says is that in §4's world, where a boundary review takes High Neb and Back Tor off Sheffield and the unit falls to 34 m, both cities have exactly seven — Sheffield by construction and Rome by accident, holding by about a metre at one end and two decimetres at the other. It is offered as a coincidence, not a measurement.

Revised — 9 September 2026

2026-09-09T09:00:00Z

The unit is corroborated. Sheffield's drops come from a curated hill database, and a curated list can leave things off, so the city was swept again from Environment Agency bare-earth LiDAR — sixty-four million cells at ten metres, no database involved. It returns the same eight summits in the same order and puts Wincobank eighth at 38.0 m, matching the database to the decimal. It makes the seventh drop 41.4 m rather than 40, but the ground it measures to is the top of a covered service reservoir fourteen metres from the summit: mask the earthwork out and Backmoor falls below Wincobank and the two change places, though what lies under the tank is what a bare-earth model cannot see, so 37.4 m is a bound rather than a measurement. The count survives either way. The stated unit is unchanged, because the database figure is surveyed and citable and the sweep is evidence about the list rather than a better measurement of it. §4 says so.

Correction — 9 September 2026

2026-09-09T09:00:00Z

Cities level on Geoffs were ordered Sheffield-first, on the stated grounds that Sheffield defines the unit — which is true, and is also the paper awarding itself the argument in the one table where it is tied. Ties now go to the city holding more summits, then by name, which puts Sheffield above Bradford on sixteen summits to ten and has nothing to do with whose unit it is. The rule is stated under the table. Sheffield is 1st of 38 English cities either way; the change moves Wakefield above Sunderland, and Doncaster up among the cities with one.

Revised — 9 September 2026

2026-09-09T09:00:00Z

Named J. G. Harston, whose survey of the seven hills the paper had been citing twice and agreeing with twice — including §5's point that two methods with nothing in common put the same places on the same ridge — while linking to it as an anonymous hyperlink. Credited in the sources as well as in the text.

Correction — 8 September 2026

2026-09-08T09:00:00Z

Eleven corrections from a fresh review. The traverse in §4 was described as forty kilometres and is thirty-four: the distance was summed between snapped grid cells, which measures a staircase rather than the straight line drawn, and inflated it by fifteen per cent — an instrument artefact of exactly the kind §4 warns about. Losing High Neb and Back Tor promotes Leeds to five sevenths, not three; that unit and its consequences are now computed rather than typed. Monte Mario is Rome’s highest hill, not its most famous — the Capitoline and the Palatine are, and both are on the list. The 1967 additions transferred land out of Chesterfield Rural District, which was not abolished until 1974. Waverley is no longer claimed for the 1918 boundary, which the working notes decline to establish. §5 named five districts and gave three drops; only three were swept. Figure 5’s alt text miscounted three ways, and Definition 1 measured the drop inside the boundary when only membership is bounded.

Revised — 8 September 2026

2026-09-08T09:00:00Z

Four diagrams. §4 draws every summit in the city as a bar from its col to its top, so the height and the drop can be read off the same picture, and a walk over all seven from west to east with the true crossings between them marked. §1 draws what a drop is, from the ground: the highest route out of Parkwood Springs, its col, and the 45 m between. §2 draws the counting function itself, including the value it never takes — two hills tie at 31 m, so it falls from twelve straight to ten.

Revised — 8 September 2026

2026-09-08T09:00:00Z

§5 is now “The traditional hills” and a third shorter. What the unit does when a hill is taken away moved to §7, where the same mechanism is already explained in the other direction.

Revised — 8 September 2026

2026-09-08T09:00:00Z

Revisions moved to their own page and grouped by day, with a feed of them.

Revised — 8 September 2026

2026-09-08T09:00:00Z

The conclusion now says what the paper established rather than only landing the joke: that seven is fixed by definition and 40 m is measured, and that the traditional list stopped being right in 1974 rather than through carelessness.

Revised — 8 September 2026

2026-09-08T09:00:00Z

Added §12, Orwell's Sheffield: the unit and the seven hills as the city stood in 1936, with the boundary reconstructed from the county borough's own register of changes. Added a map of it.

Revised — 8 September 2026

2026-09-08T09:00:00Z

§5 now sweeps the traditional names rather than reporting the database's silence about them: they are summits of 4–14 m. Added the ridge that Harston's survey and the sweep independently agree is one hill, and what the unit would be without High Neb.

Correction — 8 September 2026

2026-09-08T09:00:00Z

Corrected after review: High Neb is four metres inside the boundary, not thirty (the thirty was to the nearest vertex, not to the line); Upperthorpe, Netherthorpe and Crookesmoor are across the Don from Parkwood Springs, not on it; Rome's boundary has not always moved outwards (Fiumicino left in 1992); five authorities that are not cities, not six; the Bristol impossibility is at this instrument's resolution; and the database does not list every summit in Britain, as §9 says itself.

Revised — 8 September 2026

2026-09-08T09:00:00Z

The title is set as a masthead, the opening takes a drop cap, and the header names its hills, summits and places when you point at them.

Revised — 7 September 2026

2026-09-07T09:00:00Z

Added the map of Rome, the three-dimensional header, and a colour for each kind of statement. Renamed the counting function to H.

Correction — 7 September 2026

2026-09-07T09:00:00Z

Eight factual corrections, including the Hewitt definition, the Leeds count and §5's contradiction of its own table.

Correction — 6 September 2026

2026-09-06T09:00:00Z

Bristol's count was reported as a fact about Bristol when it was a fact about the hill database's 30 m floor. Proposition 1 was handing every region its own Geoff.

Revised — 6 September 2026

2026-09-06T09:00:00Z

Stated the seven as an axiom, cited the sources, and admitted the method's weakest point in the back matter.

Published — 4 September 2026

2026-09-04T09:00:00Z

First published: the Geoff defined, Sheffield's seven counted from the hill database and the ONS boundary, and the traditional seven checked by location rather than by name.