---
title: "Models, tables, canons"
subtitle: "A record for the calculation tools of historical astronomy"
format:
  arxivai-typst:
    keep-typ: true
    book: false
author:
  - name: Gerd Graßhoff
    affiliations:
      - name: Philosophisches Institut
        department: ""
        city: Berlin
        country: Germany
      - name: Forschungsinstitut des Deutschen Museums
        city: München
        country: Germany
    orcid: 0000-0002-4450-357X
    email: gerd.grasshoff@hu-berlin.de
abstract: |
  A longitude in a historical astronomical text is the output of a
  calculation tool. The tool has a geometric model, the tables that
  tabulate its workings, and a canon, the instructions for using the
  tables. This paper describes a record in which each of those parts is
  stored with a provenance on every value, and an engine that returns a
  model reading and a table reading. The sky reading, from DE441, stays
  beside them. Five observations of Mars printed by Duke are compared
  with the Almagest model transcribed by van Gent. The model lies 2.6 to
  5.8 arcminutes from the longitude Ptolemy states. Thirty-five entries
  of Kepler's Martian ephemerides for 1609–1611, as transcribed by
  Carman, are compared with the area law at the eccentricity and daily
  motion Carman fitted. The equated anomaly differs by at most 8.3
  arcseconds, and the root-mean-square residual is 3.2 arcseconds.
  Several tools remain gaps, and the paper says which pages have not
  been read.
bibliography: calculation-tools-bibliography.bib
---

# A position is the output of a tool

Ptolemy's longitudes come from his model and from the tables that tabulate it. An Alfonsine almanac comes from Alfonsine tables used according to their canons. The solar longitudes in the Martian ephemerides for 1609–1611 come, on Carman's reading, from Tycho's solar tables reduced to a meridian 12½ minutes of time east of Uraniborg [@carman2026a]. To annotate such a sentence one needs the tool. The sky, computed from a modern ephemeris, is a third answer. The difference between the answers is evidence about the text.

The record described here keeps the three parts apart.

- The geometric model is the construction and its parameters: an eccentric, an equant and an epicycle, Tycho's arrangement, or an ellipse with the area law.
- The tables are the printed workings of that model, with the printed rounding, the step, and the misprints.
- The canon is the order of operations that turns a date into a position: the radix and the era, the meridian, which table is entered, and how the canon says to interpolate.

A table reading can differ from the model it tabulates, through the step, the rounding, the interpolation, and the misprints. That difference is part of the historical record.

# The record

Each tool is one folder. The YAML and CSV files are the source. A DuckDB file is rebuilt from them and is opened only for the length of a query. The schema is `calculation-tool.schema.json`. A command, `calc-tools validate`, rejects a magnitude that has no provenance.

Every magnitude is one of five kinds. `read` means the value was taken from a primary page, and the record names the collection, the document, and the page. `transcribed` means a named third party is the source. `reconstructed` means a modern scholar's reconstruction, cited. `derived` means the value was computed from others already in the record, and the formula is stored. `gap` means the page that would have to be read, and no number is supplied in its place.

A value also has a status, `active`, `revised`, or `revoked`. A revoked value is kept and is not used. A curator's correction replaces a machine reading.

The constructions are a closed list: `eccentric`, `equant`, `epicycle`, `mercury_crank`, `copernican_epicyclets`, `tychonic`, `ellipse_area_law`, `vicarious`. A new construction is a new version of the schema.

Relations between tools are stored only with a provenance. An edge whose page has not been pinned is stored as a gap, so the drawing of a descent does not outrun the citation.

# Three readings

The engine reads a record and returns one of three readings.

The model reading is the geometry in `model.yaml`. The table reading follows `canon.yaml` over the transcribed cells, with the interpolation and the rounding the canon states. The sky reading is the apparent ecliptic longitude of date from DE441, on UT1 [@park2021]. It is computed by the astronomical expert and is not folded into the tool.

The differences come apart. A stated longitude minus the table reading is a slip in the calculation, or a sign that another tool was used. The table reading minus the model reading is the effect of tabulation. The model reading minus the sky reading is the error of the model and of its parameters.

Where the cells of a table can be recomputed from the model, a cell that departs by more than the printed rounding is flagged. The facsimile decides whether the cell is a misprint or an error of the OCR, and that decision is a curation event. In this run the facsimile check is open for every extracted table. The flag is not yet a verdict.

# What is in the collection, and what is not

An audit on 17 October 2026 counted page folders, OCR pages, and `table` elements. It did not compare a cell with the facsimile.

| Document | Page folders | OCR pages | Table elements |
|---|---:|---:|---:|
| greek-astronomy/Ptolemy1984 | 672 | 4 | 1 |
| greek-astronomy/Ptolemaeus1898a | 589 | 5 | 0 |
| braheOO/braheOO-02 | 480 | 480 | 263 |
| kepler-prim/kgw-10 | 553 | 553 | 59 |
| copernicus-sec/Chabas2003 | 347 | 347 | 28 |
| copernicus-prim/copernicusGW-II | 476 | 476 | 56 |
| kepler-sec/Carman2026a | 15 | 15 | 2 |

Eight pages of KGW 10 and three pages of Chabás and Goldstein did not parse as XML. They were skipped.

From the pages that did parse, the extractor copied tables whose cells contain a digit: 27 tables from the first eighty pages of Dreyer's second volume, 29 tables from Chabás and Goldstein, and 59 tables from KGW 10. The Almagest mean-motion tables and the anomaly tables are not among them. Toomer's translation and Heiberg's *Syntaxis* are present as page images. The parameters used below are van Gent's transcription [@vangent]. He states that they correspond to Ptolemy's mean-motion tables. They were not re-read from the images.

The Alfonsine record points at Chabás and Goldstein, *The Alfonsine Tables of Toledo*, whose copyright page gives 2003 and the DOI 10.1007/978-94-017-0213-3 [@chabas2003]. A closed set of planetary parameters was not read out of that book. The Parisian Alfonsine tables are a different tool and are not in the collection. The Toledan Tables and the Handy Tables have no primary here. Their records are gaps.

*De revolutionibus* is its own tool, pointing at `copernicus-prim/copernicusGW-II`. Its parameters have not been read, so it computes no longitude.

The Prutenic Tables are not in the PDAI collection. A digitised copy, Internet Archive `ARes34403`, 14,810,897 bytes, was stored outside the synchronised tree on 17 October 2026. Its tables have not been transcribed. The dependence on *De revolutionibus* is stored as a gap until a page in the collection is cited for it.

# Ptolemy's five observations of Mars

Dennis Duke prints five observations of Mars and says that he converted Ptolemy's times to Julian day numbers relative to noon at Alexandria [@duke-outer]. The longitudes below are his column for Ptolemy. His own recomputed column is not used.

The model is the Almagest equant, with van Gent's parameters. The instant is Duke's Julian day. The same day number is recovered from the model's clock formula, so the clock is not a second unknown. The sky reading is DE441 at that instant [@park2021]. Mars is the barycentre, because part 1 of the kernel has no separate body for Mars. No latitude is stated. The anomaly tables are not transcribed, so there is no table reading on these plates.

| | Julian day | Stated | Model | Model − stated | DE441 − stated |
|---|---:|---:|---:|---:|---:|
| t1 | 1768888.54167 | 81;00 | 80.945° | −3.3′ | +33.4′ |
| t2 | 1770418.37500 | 148;50 | 148.736° | −5.8′ | +91.6′ |
| t3 | 1771974.41667 | 242;34 | 242.504° | −3.8′ | +29.8′ |
| t4 | 1771977.35903 | 241;36 | 241.556° | −2.6′ | +34.1′ |
| t5 | 1622092.75000 | 212;15 | 212.330° | +4.8′ | +21.6′ |

The first three are the acronychal oppositions used for the eccentricity and the apogee. The fourth, a few days after the third, is the one used for the radius of the epicycle. The fifth is the ancient observation used for the mean motion in anomaly. @fig-t1 is the first of them. The deferent, the equant, and the line of sight are drawn large enough to follow. The ecliptic window below the diagram carries the stated place, the model, and the sky.

![The first of Duke's five Mars observations. The construction is the Almagest equant at Julian day 1768888.54167. On the ecliptic, the stated longitude is 81;00, the model is 3.3 arcminutes west of it, and DE441 is 33.4 arcminutes east of it.](figures/mars-t1.pdf){#fig-t1 width=88%}

The model lies a few arcminutes from the longitude Ptolemy states. That is the residual of this implementation against the stated place. It is not a claim that the transcription has been checked against Toomer, and it is not Duke's recomputation.

# Tycho's solar theory, and his model of Saturn

Dreyer's second volume states a tropical year of 365 days, 5 hours, and 49 minutes, and a daily solar motion of 0;59,8,19,43, on the page whose OCR index is `page_024`. The record stores both as read from that page. Carman, discussing the *Calculi ad Corrigenda*, states that Tycho's solar eccentricity is 1/28 and that the model in that calculation is undivided [@carman2020]. The same paper's account of the 1609–1611 ephemerides says that their solar longitudes use the Tychonic model with the eccentricity bisected, for an instant 12 minutes 30 seconds of time earlier than noon at Uraniborg [@carman2026a]. Both statements are stored, with their pages. The epoch mean longitude was not found on the page that gives the daily motion, so a calendar date does not yet become a solar longitude. The comparison with the solar longitudes in Carman's table waits on that radix.

For Saturn, Carman's caption to his fourth figure says that the Prutenic elements are corrected by subtracting 22′ from the mean longitude, 3;30° from the apogee, and 1/36 from the eccentricity [@carman2020]. On the next page he assumes a Copernican eccentricity of 11390 parts where the radius is 100000, and he cites Swerdlow and Neugebauer 1984, p. 545, a book that is not in the collection. No opposition longitude was read off the paper. The plate for Saturn therefore states the corrections and draws no sky (@fig-saturn).

![The Saturn record. The corrections are those in Carman's caption. No stated longitude was read, so the plate draws no sky.](figures/tycho-saturn.pdf){#fig-saturn width=88%}

# Kepler's models of Mars

The models are not one model.

For 1604 and 1605, Carman reports that the heliocentric longitude was taken from the vicarious hypothesis and the distance from a model in which the distance varies as if the Sun moved uniformly on an eccentric [@carman2026a]. The paper that sets out those models is his account of the ephemerides of 1604 and 1605 [@carman2025a]. The numerical elements were not read out of it. The record names the two constructions and computes no longitude. His paper on the first use of Tycho's observations [@carman2025] and his paper on the early distance law [@carman2026] are in the collection on the same terms: present, and not yet reduced to parameters.

For 1609–1611 the OCR of his table gives 35 entries [@carman2026a]. The aphelion is 149° 10′ 22″ for 1609 and 1610, and 149° 12′ 38″ for 1611. Days are counted from 31 July 1609 and, for the later series, from 18 June 1611. He keeps the eccentricity at 9265 parts of 100000 and reports a least-squares daily motion of 0° 31′ 26″ 38‴ 27″″. Both of those are his fit. They are stored as reconstructions, not as values printed by Kepler. The manuscript folios were not re-read.

The area law is applied from aphelion, which is how the table counts the days. The zero of anomaly was not shifted to improve the fit. On the 35 equated anomalies the mean residual is −2.8″, the root-mean-square residual is 3.2″, and the largest absolute residual is 8.3″. Carman, allowing the instant of zero anomaly to move, reports a mean of 0.0″, a standard deviation of 3.2″, and a largest departure of 8.8″. The two comparisons are the same size. They are not the same reduction.

Recomputing the geocentric longitude from the printed solar longitude, the printed distances, the printed equated anomaly, and the aphelion closes to a few arcseconds on most rows. It does not close on 8 March 1610 (−26.5′), which is one of the two dates whose Martian distance Carman calls corrupt, nor on 7 April 1610 (+50.8′) or 27 May 1610 (−13.2′). Those cells are flagged. The facsimile has not been consulted, so the flag stays a flag.

@fig-kepler is the entry for 11 February 1609. The printed geocentric longitude is 35.735°. The area-law longitude, using the printed solar longitude and the printed distances, is 35.736°. The equated-anomaly residual is −6.2″. The closure of the printed parts is 1.1″.

![11 February 1609 in Carman's table. Earth, Sun, and Mars, and the geocentric line of sight. The stated longitude and the area-law longitude agree to a few arcseconds, so the two marks on the ecliptic coincide.](figures/kepler-1609-feb11.pdf){#fig-kepler width=88%}

The Rudolphine Tables are KGW 10. Fifty-nine digit tables were copied from its OCR. The volume has 553 pages and 59 `table` elements, and eight pages did not parse, so an extracted table is not yet treated as the printed Rudolphine table. The ellipse of 1609 is a separate tool.

# What a sentence can point to

A normalisation function, given a sentence and the page address the caller already has, returns the stated body and the stated number if the sentence contains them, a calendar reference, and the tool that belongs to the wording. It does not invent a longitude. If the sentence has no number, the pointer says that the stated position cannot be retrieved. The date stays unresolved until the calendar is named. An Egyptian year, month, and day, and a Julian or Gregorian day in the small set of date strings the calendar conversion accepts, are the two conversions this record can resolve. A sentence of Ptolemy points at the Almagest model. A sentence of Tycho about Saturn points at the Saturn record, and says that the stated opposition is still missing. A sentence that names 1609, 1610, or 1611 points at the ephemeris. Anything else is an admission, and a DE441 comparison would be labelled as a fallback.

# Coverage

| Tool | Status in this record |
|---|---|
| Almagest model | transcribed from van Gent; tables not read from the edition |
| Almagest tables | gap |
| Handy Tables | gap |
| Toledan Tables | gap |
| Alfonsine Tables of Toledo | edition present; parameters not closed; 29 digit tables copied |
| Parisian Alfonsine Tables | absent |
| De revolutionibus | edition present; parameters not read |
| Prutenic Tables | digitised copy stored; not transcribed |
| Tycho, solar | daily motion and year read; eccentricity read from Carman; radix missing |
| Tycho, Saturn | corrections read; no opposition longitude |
| Kepler, 1604 and 1605 | constructions named; elements not read |
| Kepler, distance law | paper present; parameters not read |
| Kepler, 1609–1611 | 35 entries transcribed from Carman; area law compared |
| Rudolphine Tables | OCR present; tables not yet trusted |
| Babylonian Systems A and B | gap |
| Islamic tables | gap |

The plates are the form a famulus annotation can open from a sentence: the report, the sky as the text states it, and the derivation of the tool that belongs to that text. Where the text does not state a place, the plate says so and draws none.