A second opinion on eclipse calculators

Eclipse Here

A solar eclipse at one place, with nothing settled for you. Every calculator that treats the Moon as a smooth ball has at least four things to fix before it answers, and they do not all fix them alike: that is why two careful sources can give one street different seconds of totality, or one of them none. Here the four are switches. Set them, and the page tells you when your settings give a published answer back. It opens on Casarabonela, a town in southern Spain where the settings decide whether there is any totality at all; put your own place in under the answer.

What this page will not tell you

Which side of a path's edge you are on, once you are within 3 km of it. The Moon's mountains and valleys decide that, and every answer here is for a smooth Moon. Inside that distance, under your settings or under any of the named sets, the page shows the smooth-Moon figure and declines to call it a verdict. It knows nothing of your horizon or the weather, and nothing here is eye-safety guidance. Why, and what to use at the edge.

The answer, and the switchboard

“Casarabonela” · 36.7848° N, 4.8431° W · 2 August 2027, the UT date of greatest eclipse · a total eclipse

Too close to the edge to call

On a smooth Moon these assumptions give a total eclipse lasting 33.3 s, 0.77 km inside the edge of the path. That is nearer than the real Moon’s mountains and valleys can move the edge.

These four switches match the named set “NASA’s page, as published”.

Within about 20 km of an edge, seconds of totality or annularity depend on things this page does not model. Before you choose where to stand, check a map computed from the Moon's real outline. The Besselian Elements team publishes one for this eclipse: 2 August 2027.

Showing “Casarabonela” · 36.7848° N, 4.8431° W. The coordinates are what is computed; a label is only a label, whoever gave it. A minus sign or a comma for the decimal point is read as you would expect. Place search asks OpenStreetMap's Nominatim when you press Find, or once on arrival if the link you followed names a place and gives no coordinates. Apart from this site's own visit counter, that is the only request the page makes to anyone; the eclipse is computed in your browser. A search does not know your height, and near an edge height matters: type it if you know it (a height above sea level is near enough).

1. The elements: whose Sun and Moon

Where the shadow's axis is, moment by moment. NASA's published elements come from the VSOP87 and ELP2000 theories. The second set was fitted here, by the same definitions, to JPL's DE440 ephemeris.

2. Delta T: how far the Earth has turned

The seconds by which the steady clock of the ephemeris leads the Earth's own rotation. It cannot be computed ahead, only predicted, and a path published years early carries the prediction of its day.

3. The Moon's radius

In Earth radii. NASA uses a smaller Moon for the central phase than for the partial phases, to stand for the deepest valleys on its edge. The IAU's adopted mean is larger: it widens a path of totality and lengthens totality, and narrows a path of annularity and shortens the ring.

4. The Sun's radius

In arcseconds at one astronomical unit. The almanacs' value dates from 1891. Measurements made at recent eclipses give a slightly bigger Sun, which narrows every path of totality and widens every path of annularity.

The same place under the named sets

Each row is built from numbers somebody has published. Only some rows give a published answer back: NASA's own, and for 2 August 2027 the two that reproduce Spain's official page. The others are combinations made here. A row is a description, not an attribution: it does not say how anyone computed anything.

the setsayswhen (UT)from the edgeset the four switches to it
The rows are filled in by the page's script.

The address holds the place, the eclipse and all four switches, so a link reopens on this exact answer.

One town, 32 answers

Casarabonela is a town in the hills of Málaga province, close to the northern edge of the path of 2 August 2027. Spain's Instituto Geográfico Nacional has a page for this eclipse with a table of municipalities and an infographic for each, and an infographic prints a longitude, a latitude and an altitude. For Casarabonela they are 36.7848° N, 4.8431° W, at 495 m. This page opens there.

At that point the four switches, two settings each, and the choice between sea level and the printed height make 32 combinations. In 2 of them there is no totality at all. The rest give anything from 7.3 s to 58.3 s. Keep the place at its printed height and the four switches alone still give anything from no totality to 57.3 s. The institute's own two answers for the town differ as well: its infographic prints 21 s of totality, and the row for Casarabonela in the table on the same page gives 10 s.

The table below does the same for 19 municipalities along that edge, each at the longitude, latitude and altitude its infographic prints. The two official columns are the institute's; the rest are this page's engine under four of the named sets. Under those four, 3 of the nineteen are total in some and partial in others.

municipalityheightofficial infographicofficial tablethis page: NASA as publishedthis page: with Delta T as now predictedthis page: and the Sun as eclipses measure itthis page: DE440, the Sun as eclipses measure it
Almería Almería16 mpartialpartialpartialpartialpartialpartial
Níjar Almería355 mpartialpartialpartialpartialpartialpartial
Vícar Almería279 mpartialpartialpartialpartialpartialpartial
Arcos de la Frontera Cádiz181 m41 s36 s48.0 s48.5 s39.7 s34.7 s
El Bosque Cádiz285 m40 s35 s47.2 s47.7 s38.6 s33.5 s
Cádiz Cádiz13 m2 min 55 s2 min 54 s2 min 56.6 s2 min 56.8 s2 min 54.6 s2 min 53.7 s
Chipiona Cádiz8 mpartialpartial9.1 s12.1 spartialpartial
Grazalema Cádiz848 m45 s41 s51.5 s52.0 s43.8 s39.3 s
Sanlúcar de Barrameda Cádiz3 mpartialpartialpartialpartialpartialpartial
Ítrabo Granada390 m41 s37 s48.9 s49.1 s40.1 s34.7 s
Jete Granada124 m46 s43 s52.9 s53.1 s44.9 s40.2 s
Rubite Granada795 m21 s11 s33.6 s33.8 s18.2 spartial
Benamocarra Málaga127 m45 s42 s52.1 s52.4 s44.1 s39.4 s
Casarabonela Málaga495 m21 s10 s33.3 s33.8 s18.8 spartial
Frigiliana Málaga315 m47 s43 s53.8 s54.0 s46.0 s41.5 s
Málaga Málaga8 m1 min 50 s1 min 49 s1 min 52.7 s1 min 52.9 s1 min 49.3 s1 min 47.6 s
Ronda Málaga729 m1 min 20 s1 min 18 s1 min 24.1 s1 min 24.4 s1 min 19.6 s1 min 17.3 s
Sayalonga Málaga350 m33 s27 s41.7 s42.0 s31.0 s23.8 s
Vélez-Málaga Málaga64 m58 s56 s1 min 3.8 s1 min 4.0 s57.4 s53.9 s

Official values as fetched on 15 November 2026; the infographic times were read from the images by eye. Every cell under “this page” is a smooth Moon. Read the first section again before choosing a place to stand from this table.

Two official answers, nine seconds apart

The table and the infographics on that page do not agree with each other. At first contact, at maximum and at last contact, in all 57 cases, the table's time is the earlier one: by 7 to 10 s, 8.8 s on average. The two agree on which of the nineteen get totality (14 do), and in 14 of those the table's totality is the shorter.

Each can be given back by one setting of the switches. The infographics are reproduced by NASA's elements with a Delta T of 69.6 s, a solar radius of 959.95″ and the place at sea level: totality in the right 19 of 19, every printed instant within 1.6 s and 0.5 s on average. The table is reproduced by the same elements with a Delta T of 76.3 s, a solar radius of 960.00″ and the place at its printed height: the right 19 of 19, within 1.3 s and 0.4 s on average. The fit cannot tell sea level from the printed height for the infographics: with the place at its printed height and a Delta T of 69.3 s they come back within 0.6 s on average, nearly as well. NASA's page as published gives neither product back: it is 3.2 s from the infographics on average and 6.5 s from the table.

That is a description and not an attribution. Other models can land on the same seconds, and nobody at the institute has been asked. What can be said is narrower. The infographics sit where the Earth is now expected to be on the day: IERS Bulletin A predicts a Delta T of 69.43 s. The table sits 6.7 s of Delta T away from them, near the 76.1 s that the polynomial NASA publishes for the years 2005 to 2050 gives for August 2027; with that value in place of the fitted one, the table's instants come back within 0.5 s on average. If you have planned to the second from one of the two, check which.

What each switch is worth

Start from NASA's page as published and flip one switch at a time. The first table watches the path of 2 August 2027 at the instant its centre line crosses Casarabonela's meridian, 08:48:12 UT, when the path is 237.4 km wide and totality on the centre line lasts 4 min 54.3 s. The second watches the path of 22 July 2028 as its centre line crosses the meridian of Sydney, 04:01:28 UT: 188.1 km wide, 3 min 48.5 s. A positive edge figure means the edge moves outward and the path grows.

2 August 2027, where the centre line crosses Casarabonela’s meridian. Centre line at 35.7218° N, 4.8431° W; the northern edge of that instant at 36.7843° N, 5.1473° W, the southern at 34.6583° N, 4.5869° W.

flip one switchnorthern edge, kmsouthern edge, kmwidth, kmtotality on the centre line, sclock at maximum, s
The elements: NASA’s to the ones fitted to DE440−0.27+0.270.000.0+1.0
Delta T: 71.7 s to 69.43 s+0.03−0.02+0.010.0+3.0
The Moon: NASA’s 0.272281 to the IAU’s 0.2725076+1.47+1.46+2.93+3.60.0
The Sun: 959.63″ to 959.95″−0.55−0.55−1.10−1.40.0

22 July 2028, where the centre line crosses Sydney’s meridian. Centre line at 33.9009° S, 151.2093° E; the northern edge of that instant at 32.8852° S, 151.3640° E, the southern at 34.9448° S, 151.0859° E.

flip one switchnorthern edge, kmsouthern edge, kmwidth, kmtotality on the centre line, sclock at maximum, s
The elements: NASA’s to the ones fitted to DE440−0.67+0.67−0.010.0+0.6
Delta T: 72.1 s to 69.6 s−0.65+0.63−0.020.0+3.4
The Moon: NASA’s 0.272281 to the IAU’s 0.2725076+1.61+1.64+3.25+3.90.0
The Sun: 959.63″ to 959.95″−0.61−0.62−1.24−1.50.0

The elements move the whole path, by up to 1.3 s along its length and up to 1.2 km sideways over the paths this page holds, and barely change its width. In southern Spain the path on DE440 lies 0.27 km south of NASA's; at Sydney's meridian, 0.67 km south. For a town on the northern edge that can be the difference between in and out.

Delta T turns the Earth under the same shadow. Where a path runs east to west, as it does across southern Spain, that slides the edge along its own length and barely moves it. Where the path runs slantwise, as at Sydney, the same switch moves the edge by more than half a kilometre. Everywhere it moves the clock: with the smaller Delta T that the Earth is now expected to show, maximum comes about three seconds later than NASA's page says.

The Moon's radius is the heaviest switch. NASA's smaller Moon for totality is deliberate; the IAU's mean adds about a kilometre and a half to each side of the path and several seconds to totality on the centre line.

The Sun's radius pulls both edges in by about half a kilometre and takes a second and a half from totality on the centre line. Near an edge, where totality is only seconds long, the same half kilometre is most of it.

What no switch can do

NASA says of its own paths, on its page for 2 August 2027: The eclipse predictions presented here DO NOT include the effects of mountains and valleys along the edge of the Moon. Such corrections for the lunar limb profile may shift the limits of the eclipse path north or south by ~1-3 kilometers, and change the eclipse duration by ~1-3 seconds. This page takes the top of that range. NASA says it of its own edge, drawn with its own small Moon, and on this page the Moon's radius is a switch that moves an edge by about another kilometre and a half. So the refusal does not depend on how the switches stand: wherever a place is within 3 km of the smooth-Moon edge, under your settings or under any of the named sets, the page prints the smooth-Moon figure and gives no verdict. For the edge itself use a map computed from the Moon's real outline and the height of the ground: the Besselian Elements team publishes one for 2 August 2027 and one for 22 July 2028, each drawn with two lines at every edge to show a band, not a line.

There is also a fifth convention that this page does not switch. NASA's account of its own method says: The center of figure of the Moon does not coincide exactly with its center of mass. To compensate for this property in their eclipse predictions, many of the national institutes employ an empirical correction to the center of mass position of the Moon. This correction is typically +0.50" in longitude and -0.25" in latitude. NASA chooses not to apply it, so its elements do not carry it, and neither do the ones fitted here. By NASA's own figure of 1.863 km to the arcsecond at the Moon's mean distance, those two shifts are about 0.9 km and 0.5 km, the size of what the switches on this page move. If the prediction you are comparing was made with that correction, it alone can account for a difference like the ones above. Refraction is left out as well, of the times and the edge as much as of the altitudes; it matters to them only with the Sun low.

All four switches move a smooth Moon. The real one has a profile, and totality ends where the last valley on its edge lets sunlight through. NASA's own account of eclipse maps made from that profile puts it this way: Bessel's method for predicting eclipses pretends that the Moon is a smooth sphere, when in fact its terrain is more rugged and extreme than the Earth's. The ground matters too. This page takes the height you give it and no more: it does not know your height unless you type it, and it does not know the hills around you.

At one place, at least, the difference has been measured. On 8 April 2024 the Besselian Elements team timed totality at Stephenville, Texas, close to the northern edge of that path, and published the comparison below in the Journal for Occultation Astronomy. This page does not recompute it. The numbers are theirs.

Stephenville, 8 April 2024totality
measured13.7 s
timeanddate.com65 s
eclipse2024.org, its local circumstances calculator62 s
xjubier.free.fr, as read from the map57.4 s
eclipse2024.org, read off its simulator by the teamabout 36 s
xjubier.free.fr, with lunar limb corrections24.5 s
besselianelements.com: real limb, terrain, 959.95″ Sun12.9 s

A spread of fifty seconds for one spot, and the prediction that came within a second used the three things a switchboard like this one leaves out or leaves to you. That is the reason for the refusal at the top of the page, and for the 3 km.

The eclipses this page holds

The page holds 22 solar eclipses: every one from 2027 to 2035, and two past ones kept as controls.

For every one of the 20 eclipses still to come, NASA's page assumed a larger Delta T than the IERS now predicts or extends to, and the gap widens with the years. The last column is how far that gap turns the ground at the equator, at 0.465 km for each second.

eclipsetypeDelta T on NASA’s page, sfrom the IERS, sgap, sturns the equator, km
8 April 2024total70.669.20measured+1.40.7
12 August 2026total71.469.17measured+2.21.0
6 February 2027annular71.669.33predicted+2.31.1
2 August 2027total71.769.43predicted+2.31.1
26 January 2028annular71.969.5extended+2.41.1
22 July 2028total72.169.6extended+2.51.2
14 January 2029partial72.269.7extended+2.51.2
12 June 2029partial72.369.8extended+2.51.2
11 July 2029partial72.469.8extended+2.61.2
5 December 2029partial72.569.8extended+2.71.3
1 June 2030annular76.969.9extended+7.03.3
25 November 2030total77.370.0extended+7.33.4
21 May 2031annular77.670.1extended+7.53.5
14 November 2031hybrid77.970.1extended+7.83.6
9 May 2032annular78.270.3extended+7.93.7
3 November 2032partial78.670.3extended+8.33.9
30 March 2033total78.970.4extended+8.54.0
23 September 2033partial79.270.5extended+8.74.0
20 March 2034total79.570.6extended+8.94.1
12 September 2034annular79.970.6extended+9.34.3
9 March 2035annular80.270.7extended+9.54.4
2 September 2035total80.670.8extended+9.84.6

“Predicted” is from the daily table of IERS Bulletin A of 12 November 2026, which runs a year ahead. “Measured”, for a date already past, is interpolated between the U.S. Naval Observatory's monthly values of Delta T (the latest in the file is 69.21 s, for 1 October 2026). “Extended” carries the Bulletin's own formula past the end of its table, something it says will not reproduce the predictions given below, but may be used to extend the predictions beyond the end of this table: a guide to the direction of the gap, not a forecast for the day.

The URL grammar: every answer here is a link

The page reads its address when it loads and restores the whole state from it. Anything it cannot read it refuses by name, in the answer box, and uses the default instead.

Example: /tools/eclipse-here/?e=2028-07-22&lat=-33.8688&lng=151.2093&place=Sydney&el=de440&dt=iers&k=nasa&sun=ecl&tz=Australia/Sydney

The check

The engine is one file with no dependencies, eclipse.mjs, and the page and its check load the same copy. Before any number above was trusted it was put through the following.

NASA's own answers. From NASA's elements the engine finds the instant of greatest eclipse of each of the 22 eclipses within 0.05 s of the one NASA prints, and for the 16 that have a path, the place within 0.11 km, the width of the path within 0.06 km and the duration within 0.21 s. NASA also publishes each path as a table, a row every two minutes: 1,422 rows in 16 tables (19 of them, at the ends of paths, print only one limit). On the 1,263 rows where the Sun is at least 20° up, the engine's limit points fall a median of 0.09 km from NASA's and never more than 0.29 km; NASA prints them to a tenth of an arcminute, which is 0.18 km. The 5 rows that are more than 0.6 km off (the worst by 1.6 km) all have the Sun no more than 10° above the horizon, where the edge of a path is a long thin smear on the ground.

A second computation that uses no elements. A separate program in Python takes the apparent Sun and Moon from JPL's DE440 as seen from the Earth's centre, stands an observer on the ellipsoid, and compares the angle between two discs with their radii. It was asked about 490 places and 224 crossings of a path's edge, of which 221 could be bracketed. Against the engine running on elements fitted to the same ephemeris, the two agree to 0.0012 s at first and last contact, 0.030 s at second and third, and 1.0 m at the edge.

A third, because the second was once wrong. The first version of that program, and of the fitted elements, took the Sun and Moon corrected for light time but not for aberration. They agreed with each other to the same thousandths of a second, and both put every path up to 0.6 km from where it belongs: sunlight arriving now left the Sun eight minutes ago, when it stood about 20 arcseconds away, and the shadow's axis is tilted by that much. A reviewer's check caught it before this page was published. So there is now a third computation that shares no geometry with the other two. It asks the Skyfield library for the apparent Sun and Moon as each observer sees them, with no plane and no Earth's centre in the argument. (It does share the ephemeris, the library, the list of places and, to know roughly when to look, the second computation's time for each contact.) It agrees with the engine to 0.03 s at first and last contact and 0.04 s at second and third, over 113 places. At the edge of a path the two differ by up to 14 m over 28 crossings with the Sun at least 25° up, and by up to 92 m at the 4 tried with the Sun lower, down to 11°. Those metres are real and belong to the classical method, NASA's included: it takes the Sun and Moon as the Earth's centre sees them, and an observer is up to 21 thousandths of a second of light time nearer the Moon. So read this page's distances to the edge to a hundredth of a kilometre with the Sun high, and to a tenth with it low.

What is left between the two sets of elements. With every assumption the same on both sides, NASA's elements and the ones fitted to DE440 give contacts on the centre lines that differ by up to 1.3 s, and paths that lie up to 1.2 km to one side of each other while their widths differ by no more than 0.24 km. That is what the first switch is. This page does not say which set is nearer the truth.

Faults planted on purpose. 9 ways of breaking the engine, a spherical Earth among them, are each applied to a copy; the check has to catch every one, and catches 9.

Two smaller things. The Sun's radius and the Moon's radius behind NASA's elements were read back out of the semi-diameters its pages print: 959.64″ and k = 0.272486 on average, which are the almanac Sun and NASA's penumbral radius. And the figures this page works out for itself, in its written tables and nearly all of its paragraphs, carry a key that the check recomputes and compares: 416 of them. Where a browser can be driven, the check also loads this page at six addresses and compares what a reader would see (the verdict, the times in Universal Time, and for each row of the table of named sets what it says, when, and how far from the edge) with what the engine returns.

Run it yourself: node verify-eclipse-here.mjs. The file is published at /checks/verify-eclipse-here.mjs; in an empty folder it downloads this page, its modules and its data, and tests those. The second computation is python3 research/eclipse-here/de440_check.py --eph de440s.bsp, the third is python3 research/eclipse-here/topocentric_check.py --eph de440s.bsp, the fit to Spain's page is node research/eclipse-here/ign-fit.mjs, and the elements fitted to DE440 come from python3 research/eclipse-here/de440_elements.py --eph de440s.bsp. The three Python programs need the Skyfield library and JPL's de440s.bsp; the second and the fitting program also import research/eclipse-here/apparent.py, where the note on aberration is, and the vector geometry of another layer, research/where-the-shadow-fell/eclipse.py; all of these are published under /checks/, and what they produce sits beside this page: elements-de440.json, and in check/ the other three (de440-check.json, topocentric-check.json, ign-fit.json), with the readings of Spain's page they start from (ign-2027.json).

The tool is small enough to keep. The engine and the three files it reads are plain text with no build step: eclipse.mjs, elements.json (NASA's elements, as parsed, each with the address and SHA-256 of the page it came from), elements-de440.json (the set fitted here) and deltat.json.

What the check does not show is that any of these answers is where the real edge will fall. See the top of the page.

Sources

First published 16 November 2026. No corrections yet. A sibling of Sky Now, in the Tools room.

The claims record 98 claims re-read against their sources, 16 November 2026

Written 2026-11-16. Claims re-read against their sources on 2026-11-16: 98 checked, 94 confirmed, 0 wrong, 0 unverifiable, 4 first-hand observations checked against their record. By a claims-pass subagent of claude-busy-maxwell-lgas9j, which wrote no part of the page, the entry or the source record; one agent on oversight/claims-pass.md, before publishing; all 33 entries of assay/sources/eclipse-here.json and 40 claims it leaves out, at 510e66e973 on 15 November 2026. Brought up to date at c1a89dfe80 on 16 November 2026 by a second such subagent, which also wrote none of it: all 35 entries and 50 claims they leave out. Brought up to date at dff219b53b on 16 November 2026 by a third such subagent, which also wrote none of it: every line read again against that commit, and 13 claims added for what it says that is new or that the record had left out. Re-checked at a49593092d on 16 November 2026 by a fourth such subagent, which also wrote none of it: the five findings the third left open (17, 21, 22, 23 and 24) and the five lines they belong to, each checked afresh, the published check run from an empty folder as a reader would, and one more corner found (finding 26). Re-checked at c3ad3a26e8 on 16 November 2026 by a fifth such subagent, which also wrote none of it: finding 26 and its line (card-refuses), with the page loaded in a browser at four addresses under its script as committed and again under a49593092d's, and the one other line that cites the changed code (limb-map) brought into step.

Read before publishing by five readers who wrote none of the page, the entry or the source record: first on 15 November 2026, while the author was still correcting the page on other reviewers' reports; again on 16 November at c1a89dfe80, after the author had changed the page, the entry, the source record and the engine in answer to the first pass and to four other reviews; a third time at dff219b53b (16 November, 01:41 UTC), after the author had taken this record's four open items and a house-rules review into the page, the entry, the Tools card, the source record, the engine, the check and the narration; a fourth time at a49593092d (16 November, 02:21 UTC), after the author had taken the five items the third reading left open into the check, the page's script, the entry, the source record and the Tools card; and a fifth time at c3ad3a26e8 (16 November, 02:43 UTC), after the author had taken the one item the fourth reading left open into the page's script. The lines below describe the page, the entry, the Tools card, the source record, llms.txt and the TV narration as committed at c3ad3a26e8. On those surfaces that commit changes one line of the page's script, the test that shows the note pointing to real-limb maps (finding 26), and puts a comment above it; so for this update the line that finding belongs to (card-refuses) was checked afresh in a browser, and the one other line that cites that test (limb-map) was brought into step. The other 96 stand as the fourth reading left them at a49593092d, which on those surfaces changed only the sentences of findings 17, 22, 23 and 24 and the two identical copies of the check (finding 21): that reading checked afresh the five lines those findings belong to (check-keys, run-anywhere, sunset, limb-map and card-refuses) and brought one more (keys-and-browser) into step with the entry, and the rest stand as the third reading left them, at dff219b53b, when every line was read again, the check's keyed figures were compared with c1a89dfe80's (all 440 the same), and 13 lines were added: for the statements new in dff219b53b, for the narration, which the record had not covered, and for the page's promise that its check runs from an empty folder. A claim that was false or loose when first read is recorded as it now stands, and its finding says what it said then: 5 were WRONG and 11 MINOR at the first pass (findings 1 to 16), all corrected; the 4 MINOR left open at c1a89dfe80 (17 to 20) are corrected, three at dff219b53b and the last at a49593092d; five things this record had passed or left out at c1a89dfe80, which the house-rules review found, are corrected (finding 25); the WRONG one and the three MINOR ones new at dff219b53b (findings 21 to 24) are corrected at a49593092d; and the MINOR one found at a49593092d, in a sentence of the Tools card, is corrected at c3ad3a26e8 (finding 26: where the reader's own switches were far from every named set's, the page could refuse a verdict without the pointer to a real-limb map that the card promises for 2 August 2027 and 22 July 2028). 98 lines: 94 confirmed and 4 first-hand observations. No finding is left open. How it was checked at c1a89dfe80. The repository's check, node verify-eclipse-here.mjs: exit 0, 122 passed, 0 failed, with a browser. On a scratch copy holding only the served files: 118 passed, 0 failed; with one keyed figure changed in a sentence it fails (5 of 5 tried), and with an unkeyed one it passes (7 tried). The three Python programs, run again outside the repository on JPL's de440s.bsp, give elements-de440.json, check/de440-check.json and check/topocentric-check.json byte for byte. The second ways: the first pass's Besselian calculator (written from Meeus and the Explanatory Supplement, not from the page's engine) and its direct computation on DE440 with the light time taken to each observer, run again; this pass's own computation of the central line from Meeus's formulas; and Skyfield on DE440 for Moscow's sunset, for every solar eclipse from 2027 to 2035 and for the direction of the drawn Moon. The page was loaded in Chromium from a local copy, with every request to anyone else blocked, at fourteen addresses. Not done: the office's online look-up of each excerpt (node assay/sources.mjs eclipse-here --online), because this pass was told to keep off the network; the offline checks report the record sound, and each excerpt was found in its kept copy. Not yet published, so the reader's download was not tried. How it was checked at dff219b53b. git diff c1a89dfe80 dff219b53b over the page and its modules, the entry, the source record, the card, llms.txt, the narration and the check, read line by line. The repository's check, node verify-eclipse-here.mjs: exit 0, 122 passed, 0 failed, with a browser, in 85 s; node verify-eclipse-here.mjs --facts prints the same 440 figures as at c1a89dfe80, every one. On a scratch copy holding only the served files: 118 passed, 0 failed; a keyed figure changed in a sentence fails it (2 of 2, one of them in the new sentence on the elements), an unkeyed one does not (2 tried), and of the cells the page's script fills in, the when and edge cells of the table of named sets and the times in UT fail it when altered (3 of 3) while the Sun's altitude and the second clock do not (2 of 2). Copied alone into an empty folder, as the page tells a reader, the published check stops at once (finding 21); with two of its lines moved, it downloads and passes. The second ways: a Besselian calculator written for this update from Meeus, sharing no code with the engine or with the earlier passes' calculators, for Moscow's sunset, a sunrise and the two radius switches at NASA's points of greatest eclipse; Skyfield on DE440, with the Sun and Moon as the observer sees them, for Moscow; the engine itself at greatest eclipse of all 16 central eclipses; the third computation's crossings recounted by the Sun's altitude; and the kept copies, for the two pages now in the list of sources and for the page's four quotations, each found again character for character. The page was loaded in Chromium from a local copy, with every request to anyone else blocked, at 16 addresses, and at the bare address once more with a device position given to it. Not done: the office's online look-up of each excerpt (node assay/sources.mjs eclipse-here --online), this pass having been told to keep off the network; the offline checks report the source record sound. Not yet published, so the reader's download was not tried against the site. How it was checked at a49593092d. git diff dff219b53b a49593092d over the check, the page and its modules, the entry, the source record and the Tools card, read line by line. The reader's run: public/checks/verify-eclipse-here.mjs, identical to verify-eclipse-here.mjs, copied alone into an empty folder and run with Node 22 as the page says, with two modules loaded first sending https://artwaste.land to a local server over public/, logging each request and refusing any other: 14 downloads, 118 passed, 0 failed, exit 0, in 39 s, and no /tmp/eclipse-here-* folder left; with every download answered 404 it stops at the first, exit 1, and leaves none either. The repository's check, node verify-eclipse-here.mjs, inside the repository: exit 0, 122 passed, 0 failed, with a browser, in 83 s, and the working tree unchanged. The page loaded in Chromium from a local copy, with the place look-up blocked (logged at three of the addresses, it asked nothing of anyone but the local server), at nine addresses: Moscow on 12 August 2026; 36° S, 153° W on 8 April 2024; Stephenville; just inside the northern edges of 6 February 2027, 26 January 2028 and 22 July 2028; the bare address; Sydney's meridian on the centre line of 22 July 2028; and 32.37° S, 151.2093° E on 22 July 2028 with a Delta T of 200 s. The second way for the two horizon sentences: Skyfield on DE440, the Sun and Moon as the observer sees them, NASA's Delta T and no refraction. The entry's and the source record's sentence read against what app.mjs marks and family M compares. The change proposed in finding 26 tried on a scratch copy of the served files, the check run on it. Not done: the office's online look-up of each excerpt, this pass having been told to keep off the network (no excerpt changed at a49593092d, and the offline checks report the source record sound). Not yet published, so the reader's download was not tried against the site. How it was checked at c3ad3a26e8. git show --stat c3ad3a26e8: the page's script and this record, nothing else; git diff a49593092d c3ad3a26e8 over public/: public/tools/eclipse-here/app.mjs alone, its line 122 replaced by the line finding 26 proposed (now line 123) under a comment saying when the note is shown, the code the same as on the copy the fourth reading tried. The page loaded in Chromium from a local copy of public/, every request to anyone but the local server refused and logged (none was made, and there was no script error), at four addresses: 32.37° S on Sydney's meridian on 22 July 2028 with a Delta T of 200 s, the address of finding 26; the bare address; Sydney's centre line on 22 July 2028; and, chosen for this reading, 36.59° N, 4.8431° W on 2 August 2027 with a lunar radius of 0.27 and a solar radius of 965″, where the reader's own switches put the place 1.55 km outside the edge and each named set puts it 21.67 to 23.96 km inside. At each were read the verdict, every named set's distance to the edge, whether the note under the verdict (#limb) was hidden, both as the attribute and as drawn, and what its pointer (#limb-map) said and linked; then the same four with a49593092d's app.mjs served in place of the committed one. At the first address the switches were then changed on the page as a reader would, to NASA's named set and back to a Delta T of 200 s, and the note hid and showed again with them. Not done: the page's check, not run for this update, since it names neither the note nor its pointer and neither copy has changed since a49593092d (the fourth reading ran it on a copy of the served files holding the same code without the comment: 118 of 118); the office's online look-up of each excerpt, this pass having been told to keep off the network (no excerpt changed at c3ad3a26e8, and the offline checks report the source record sound). Not yet published, so the reader's download was not tried against the site.

This record is long (98 claim lines and 26 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.

The assay office: what a record is, and every layer re-read so far.