Reading a frozen snapshot of the extraction artefacts, taken 2026-08-15 07:05 UTC at commit e96504e288 files. Figures are recounted from that snapshot, not from a live pipeline.
The Work Record

Provenance

Every number on this site is recounted from the extraction artefacts when the page renders. Nothing is transcribed from a run report — those are written per run and a later witness overwrites an earlier one, so the copy on disk carries the last witness deposed under a name that reads like the project’s.

The witnesses, unmerged

Each catalogue is resolved against itself only. The same apparatus appears twice below — once as Fisher names it, once as Stevens does — and nothing on this site claims they are the same object, because no pass has yet established that they are. Cross-witness resolution is a single later pass over every deposed witness at once; running it pairwise would make the catalogue depend on the order the books were read in.

Handbook
1914 · primary
Fisher
1913 · secondary-interpreter
Stevens
1913 · secondary-interpreter
Advanced
1917 · primary
Elementary
1917 · primary
primaries
Montessori’s own four volumes
leaves with text1571482553684161376
material mentions355135140589311138
mentions per leaf2.2610.9120.5490.1582.2380.827
leaves naming nothing55.4%56.8%82.0%95.9%67.3%73.5%
materials after resolution673754 (4254 across 4 runs)13121246
materials per leaf0.4270.2500.2120.0350.2910.179
materials with a quote-verified age211009
age assertions2110014
materials with quote-verified control of error32402
plate sightings29296025
asserted orderings69200
pair readings1,5845401,632381,304
not asserted99.1%95.0%99.4%100.0%100.0%

The control-of-error cell for the primary corpus is empty on purpose. That corpus was extracted under a prompt that asked for control of error with no quote check, so its count is not comparable to a quote-verified one, and no substitute is offered in its place. The plate, ordering and not-asserted rows are blank for the same reason: those passes were run against the deposed witnesses individually, not against the pooled primary corpus.

The not-asserted rate is not comparable between witnesses as a bare percentage either. Its denominator is candidate pairs, which grows with catalogue size and with how often materials are named together — Stevens lists apparatus in bulk, so it reads far more pairs from fewer orderings.

The relevance gate

A verified quote is one that occurs on the leaf. That is necessary and it is not sufficient: one sentence about children’s vision at three was filed as the readiness age of four different materials, and the quote check passed it four times. Every assertion is therefore put to a second, mechanical question — does this sentence, read in its place on this leaf, make this claim about this thing? — and a no deletes it. What follows is what the gate removed, with the reason it logged.

witnessabstractioncontrol of errororder authorityorderingreadiness ageall
Handbook
1914
0/40%6/2921%2/367%15/3148%2/540%25/7235%
Fisher
1913
2/1712%2/2100%27/3675%1/119%32/6648%
Stevens
1913
4/1233%1/333%10/1759%3/933%18/4144%
Advanced
1917
0/10%3/3100%0/20%3/650%
Elementary
1917
4/2715%3/496%0/30%7/799%

Gate: google/gemini-3.7-flash, version handbook-relevance-v5+b277ef.s3. Calibrated 12/12 against hand-read controls before it was trusted; it was wrong twice before it was right, and both faults were caught by that suite rather than by reading the code.

Fisher’s readiness ages are 0 of 11. Every one was a general statement about children’s ages attached to whichever material the leaf happened to name — so Fisher now contributes no readiness age to this site at all, and its spine carries no age marks.

Reading arms

Each ordering question was put to two models independently over identical pages. Where they disagree the edge carries both answers and the disagreement is recorded, never resolved.

Handbook

google/gemini-2.5-flash-lite8/7921.0%
google/gemini-3.6-flash7/7920.9%

3 of 6 orderings carry both arms; 3 carry recorded dissent. On every witness the stronger model asserted no ordering the cheaper one did not.

Fisher

google/gemini-2.5-flash-lite14/2705.2%
google/gemini-3.6-flash13/2704.8%

7 of 9 orderings carry both arms; 2 carry recorded dissent. On every witness the stronger model asserted no ordering the cheaper one did not.

Stevens

google/gemini-2.5-flash-lite5/8160.6%
google/gemini-3.6-flash5/8160.6%

2 of 2 orderings carry both arms; 0 carry recorded dissent. On every witness the stronger model asserted no ordering the cheaper one did not.

Advanced

google/gemini-2.5-flash-lite0/380.0%

0 of 0 orderings carry both arms; 0 carry recorded dissent. On every witness the stronger model asserted no ordering the cheaper one did not.

Elementary

google/gemini-2.5-flash-lite0/13040.0%

0 of 0 orderings carry both arms; 0 carry recorded dissent. On every witness the stronger model asserted no ordering the cheaper one did not.

How each catalogue was resolved

witnessblocked atembeddingadjudicated byforms rejectedprompt versions read
Handbook0.84google/gemini-embedding-001google/gemini-3.6-flash16handbook-edges-v1+ab394d.s2
handbook-authority-v2+b6d6cb.s2
Fisher0.84google/gemini-embedding-001google/gemini-3.6-flash3fisher-edges-v1
fisher-authority-v2 (superseding fisher-authority-v1)
Stevens0.84google/gemini-embedding-001google/gemini-3.6-flash21stevens-edges-v1+ab394d
stevens-authority-v2+b6d6cb
Advanced0.84google/gemini-embedding-001google/gemini-3.7-flash3advanced-edges-v1+ab394d.s3
advanced-authority-v2+b6d6cb.s3
Elementary0.84google/gemini-embedding-001google/gemini-3.7-flash55elementary-edges-v1+ab394d.s3

Where an artefact holds more than one prompt version, the site renders the latest and says which it superseded. Fisher’s order-authority file holds both a single-leaf pass and the three-leaf pass that replaced it; showing both would mix two different questions and double-count one declaration.

Check ten at random

Unfiltered, from every quote-bearing assertion on disk across every deposed witness — orderings, readiness ages, control of error, order authority, disputed edges included. Open the page and see whether the sentence says what we say it says.

41 quote-bearing assertions on disk · draw 0
readiness ageSolid Insetsage 2.5handbook
which can be done even by a child of two and a half.
no printed folio · leaf 150
control of errorSolid Insetshandbook
Here tlie child cannot help seeing his mistake in concrete form. He is perplexed, his little mind is faced with a problem which interests him intensely. Before, all the cylinders fitted, now there is one that will not fit. The little one stops, frowning, deep in thought. He begins to feel the little buttons and finds that some cylinders have too much room. He thinks that perhaps they are out of their right place and tries to place them correctly. He repeats the process again and again, and finally he succeeds.
p. 32 · leaf 60
control of errorPlane Geometrical Insetshandbook
Here also the control of the error is in the material, for the figure cannot be inserted perfectly except when it is put in its own place.
no printed folio · leaf 84
control of errorMusical Notation Materialhandbook
Then they verify their work by turning the discs over and reading the name.
no printed folio · leaf 148
control of errorMusical Notation Materialhandbook
If he is not sure he consults the numbered board (Fig. 33).
no printed folio · leaf 147
readiness ageMontessori Bellsage 5handbook
This exercise fascinates children from five years old upwards.
p. 63 · leaf 105
orderingBells → Musical Note Discssequence · weak · gemini-2.5-flash-litehandbook
Another exercise which the child can do at the same time is to place the disc bearing the name of the note on the rectangular base of the corresponding bell, whose sound he has already learned to recognize by ear in the sensorial exercise described above.
p. 99 · leaf 145
orderingPlane Geometrical Insets → Metal Insetssequence · strong · gemini-3.6-flashhandbook
The child has not actually performed a new exercise, because he had already performed all these movements when he touched the wooden plane insets.
p. 87 · leaf 129
orderingRough and Smooth Board → Touch Board with Graduated Stripssequence · strong · gemini-2.5-flash-litehandbook
After the board with the two contrasting sur- faces, the child is offered another board on which are gummed strips of paper which are rough or smooth in different degrees.
p. 38 · leaf 68
orderingSolid Insets → Broad Stairsequence · weak · gemini-2.5-flash-lite + gemini-3.6-flashhandbook
Let us look at the three sets of material which are presented after the exercises with the solid insets, i.e., the material for teaching size (the pink cubes), thickness (the brown prisms), and length (the green rods).
p. 103 · leaf 153