Canonical: https://thonly.org/research/where-the-manual-adds-a-reading · Licence: CC0 1.0
This paper reports a machine-checked test of a classification of causal relations against the source text it summarizes. The classification is the Abhidhammattha-saṅgaha's grouping of the 24 conditions (paccaya) of the Paṭṭhāna, the seventh book of the Theravāda Abhidhamma, into six groups by the kind of thing that conditions and the kind of thing that is conditioned: mind, matter, or a concept (paññatti). The grouping was held out as an answer key and never used as an input. The Paṭṭhāna's 24 definitions, and then its detailed section for the wholesome triad, were read clause by clause into a table of 199 clauses, each checked on its line of the Chaṭṭha Saṅgāyana (CST) edition, and a classifier written in Lean 4 sorted the 24 conditions from that table alone. Three results follow. The definitions alone place 19 of the 24 as the manual does. With the detailed section, 22 placements, none involving a concept, follow under both readings of the Paṭṭhāna's phrase all dhammas, with no commentary: the detailed section, not the definitions, closes proximity, contiguity and the mental side of decisive support, and supplies result's matter. The remaining two placements, the object condition and decisive support, are the ones that involve a concept, and they turn on readings the sources leave open: whether designations count among dhammas, whether the person named among the decisive-support conditions is a designation or an aggregate-continuum, and whether the commentary's statements about concepts are read distributively or collectively. Checked each way, the count is 22, 23, 23, 24, 24 or 23. The paper also reports two earlier claims it withdrew on the same day, 23 from the definitions alone and 24 from the Piṭaka alone, and the instrument faults that were each caught once on a planted case. It is a study: it discloses no mechanism, asserts no priority, and every number in it is read from the generated artifact.
Keywords: Abhidhamma, Paṭṭhāna, conditional relations, twenty-four conditions, Abhidhammattha-saṅgaha, Theravāda Buddhism, classification, held-out answer key, formal verification, Lean 4, proof assistant, kernel evaluation, mutation testing, textual criticism, commentary and sub-commentary, designation (paññatti), Pāli canon, Chaṭṭha Saṅgāyana, pre-registration, study.
The paper uses Pāli vocabulary because it reads Pāli texts. An examiner or indexer searching in standard terms should find each concept under the term on the right.
| Term used here | Standard term |
|---|---|
| Paṭṭhāna | the seventh book of the Theravāda Abhidhamma Piṭaka; a treatise on 24 types of conditional relation (causal relation) |
| paccaya (condition) | a type of causal or conditional relation between two kinds of event |
| Abhidhammattha-saṅgaha (the Saṅgaha; "the manual") | a medieval systematic manual of Theravāda Buddhist psychology and metaphysics |
| nāma / mind | the mental side: consciousness and its mental factors (the Dhammasaṅgaṇī, §1316, also counts the unconditioned element under it) |
| rūpa / matter | the material side of the Abhidhamma's two-part analysis |
| paññatti / concept, designation | a name or a notion, as opposed to what it names; in the Abhidhamma neither mind nor matter |
| the six groups | a classification of the 24 relations by (kind of the conditioning term → kind of the conditioned term) |
| niddesa (the definitions) | the section of the Paṭṭhāna that defines each of the 24 conditions in one or a few sentences |
| pañhāvāra (the detailed section) | the section of the Paṭṭhāna that works each condition through a matrix of questions; here its first triad (wholesome, unwholesome, indeterminate) |
| dhamma | a phenomenon; the Paṭṭhāna's "all dhammas" is the phrase whose extent is in question |
| open side | a clause whose conditioning or conditioned term is a "whatever" phrase, is not named, or is the class term abyākata (indeterminate), which includes matter |
| held-out answer | a reference answer used only to score a result, never as an input to it |
| capture code | the script that turns a text into the data a check reads |
| the breaks | mutation testing: deliberate faults, each required to make the check fail |
| refuter | a reviewer with no authority to rescue a claim, instructed to find the case that kills it |
| layer | an editorial stratum of the text: Piṭaka · commentary · sub-commentary · manual |
This document and its contents are dedicated to the public domain under the Creative Commons CC0 1.0 Universal Public Domain Dedication, and are published so that they stand as prior art against any later attempt to enclose them. No patent has been or will be sought on any method, formalization, protocol, software or result described in this paper by HeartBank®, Factory 333™, THonly™, Silicon Wat℠, or any entity under their common control, in any jurisdiction, at any time. The authors and those entities commit not to assert any patent right against any party practicing any method disclosed here. The commitment is stated rather than implied, is permanent, and is not conditioned on reciprocity, attribution, or field of use. A publication grants nothing and frees nothing already enclosed. The generator, the Lean files and the scripts are published under CC0 at github.com/SiliconWat/formal-abhidhamma (files cited at commit 8663f69, directory matika/).
This document is a defensive publication establishing prior art as of 3 November 2026 for what it discloses. It asserts no priority for any textual fact, for the grouping itself, or for the idea of checking a classification by machine.
The quick census. A quick literature census of this study's claim was pre-registered before its first query (SiliconWat/formal-abhidhamma record, pushed 2026-10-13 18:01 PDT, commit 4956261) and run on 2026-10-13. Aperture, declared before searching: English web; Google Books and Scholar; archive.org full texts; named scholarship (U Nārada, Bhikkhu Bodhi, Nyanatiloka, Y. Karunadasa, Ledi Sayadaw in translation); 22 queries, logged. Not run: Burmese, Thai, Sinhala, Japanese and German scholarship; patents (out of scope for a study). Control: Bodhi's A Comprehensive Manual of Abhidhamma, ch. 8, which explains the grouping in English, was required to be found. It was found, but only by a direct full-text fetch at archive.org; the web search never surfaced its text. The census's negative rows are therefore weak, and a full literature census is owed before any first deposit.
| Conjunct | Result, in the aperture above, 2026-10-13 |
|---|---|
| A per-condition statement of what each of the 24 conditions relates | Known. Bodhi (ed.), CMA ch. 8 (1993); Nārada, Manual ch. VIII; Agganyani, Paṭṭhāna: Conditional Relations (abhidhamma.com, 2008). Cited, not claimed. |
| The six-way grouping | Known. It is the manual's own, §15–20 and §26, and is presented in the modern guides as the manual's summary. Cited, not claimed. |
| The grouping derived from the Paṭṭhāna's definitions | Narrowed. A derivation was not found in the aperture. Nārada's page states the grouping without citing a Paṭṭhāna passage (its four sentences were fetched again on 2026-11-03 and carry no citation). |
| Concepts as objects of consciousness, and the place of the concept in the group | Narrowed. Concepts as object are discussed in the guides; a link from the reading of all dhammas to the placement of the object was not found. |
| The definitions alone leave proximity, contiguity and decisive support open and the detailed section closes them | Narrowed lightly. Agganyani's Paccayaniddesa resolves the class term abyākata under proximity by a footnote naming mental states, not by the detailed section. The framing here was not found. |
| A machine-checked reproduction of an Abhidhamma classification against a held-out manual answer | Not found in the aperture, outside this series' own The Counts Check. Contrast, not collision: Barendregt's slides (2010), an E8 record on Zenodo (2026-06-16), and the Lean/Isabelle Buddhist-logic repository named in №4. |
Not found means not found in this aperture on this date. It does not mean new. The Burmese study tradition of the Paṭṭhāna is large and mostly untranslated, and the likeliest place for any of the "narrowed" rows to be killed is there.
The series The Abhidhamma Compiled treats the Abhidhamma as a specification a machine can check. Paper №4, The Counts Check, tested one such claim for chapter 2 of the Abhidhammattha-saṅgaha: that its statements of which mental factors combine with which types of consciousness, given twice, by factor and by type, are mutually consistent under a short set of general rules. That paper checked the manual against itself.
This paper checks the manual against the canon it summarizes. The manual's eighth chapter, on conditions, closes with a compact statement. The 24 conditions, it says, are of six kinds by what relates to what (abh07t.nrf.xml:2845, Saṅgaha ch. 8 §15; the CST reads Chadhā nāmaṃ tu nāmassa, pañcadhā nāmarūpinaṃ). Nārada's English reads: "In six ways mind is related to mind. In five ways mind is related to mind and matter. Again mind is related in one way to matter, and matter in one way to mind. In two ways concepts, mind and matter are related to mind. In nine ways are the two - mind and matter - related to mind and matter. Thus the relations are sixfold." The sentences are a summary; the manual is a medieval work, and the Paṭṭhāna, which defines the conditions, is far older. A summary can be tested against its source in a way the manual's chapter 2 cannot: the question is whether the grouping follows from what the Paṭṭhāna says each condition relates, and where, if anywhere, it needs something the Paṭṭhāna does not say.
Three reasons make the question worth a machine's time.
First, an answer that is a computation. The grouping is not a new doctrine. It is a classification of the 24 by two attributes (the kind of conditioning term, the kind of conditioned term), each of which the Paṭṭhāna's definitions state or leave open. That makes it a case where the answer exists in the text before any rule is applied to the text, and where the reproduction can be scored against a key that the reader of the rules never saw.
Second, a locating result is worth more than a confirming one. If a reading reproduces 24 of 24, nothing is learned about where it was lucky. If it reproduces 22 and shows exactly which two placements need what, a reader who disagrees with a reading knows which number to change.
Third, the method. The same instruments (a held-out key, capture code that may not read it, refuters, breaks) apply to every later classification the series checks. Section 3 reports them, with the faults each was seen to have, because an instrument that has only ever passed is an untested claim.
The Paṭṭhāna's list of conditions (abh03m7.mul.xml:37, repeated in the manual at abh07t.nrf.xml:2841) runs: root · object · predominance · proximity · contiguity · co-nascence · mutuality · support · decisive support · pre-nascence · post-nascence · repetition · kamma · result · nutriment · faculty · jhāna · path · association · dissociation · presence · absence · disappearance · non-disappearance. The manual's grouping, as the Lean file matika/Relata.lean encodes it from the manual's §16–20 and §26 (abh07t.nrf.xml:2861–2877, :2901) and as kala/ANSWERS.md records it, is:
group (conditioning → conditioned) n the conditions
--------------------------------------- -- -------------------------------------------------
mind → mind 6 proximity · contiguity · absence · disappearance
· repetition · association
mind → mind and matter 5 root · jhāna · path · kamma · result
mind → matter 1 post-nascence
matter → mind 1 pre-nascence
concepts, mind and matter → mind 2 object · decisive support
mind and matter → mind and matter 9 predominance · co-nascence · mutuality · support
· nutriment · faculty · dissociation · presence
· non-disappearance
---
24 (6 · 5 · 1 · 1 · 2 · 9; theorem sangaha_counts)The manual states the fifth group in one clause, §20: Ārammaṇavasena upanissayavaseneti ca duvidhā paññattināmarūpāni nāmasseva paccayā honti (abh07t.nrf.xml:2877): "in two ways, as object and as decisive support, concepts, mind and matter are conditions of mind alone." It is the group whose conditioning side names a concept.
The Paṭṭhāna defines each condition twice. The definitions (niddesa, abh03m7.mul.xml:37–265) give each condition one to several sentences saying what conditions what. The detailed section (pañhāvāra) then works each condition through a question matrix for each triad of the Dhammasaṅgaṇī's table of contents; for this paper it is read for the wholesome triad (wholesome, unwholesome, indeterminate), which matika/RELATA.md cites from line 5569 to line 6357.
The definitions are short and sometimes open. Three phrases matter below.
abh03m7.mul.xml:49): "all dhammas are conditions, as object, of the mind-consciousness element and what is associated with it.":53), and for proximity, Yesaṃ yesaṃ dhammānaṃ anantarā ye ye dhammā uppajjanti cittacetasikā dhammā… (:97). The conditioned side names mind; the conditioning side is "whatever".:165): "a person too is a condition, as decisive support." The clause names no conditioned term.Two sentences in the sources are the hinge of the rest.
The first is the Dhammasaṅgaṇī's. Its paññatti-couplet defines designation by listing its synonyms and closes ime dhammā paññatti (abh01m.mul.xml:8453), "these dhammas are designation". So the canon itself calls designations dhammas. Against that, the commentarial layer sets concepts apart from the mind-and-matter that the conditions otherwise concern (§5.3 below).
The second is the Dhammasaṅgaṇī sub-commentary's, on how a person can be a condition. In its own voice, at abh01t.tik.xml:2557, it says that a taking-up of the aggregate-continuum as a unity, as "a good friend, a bad friend, a person" (kalyāṇamitto pāpamitto puggalo), occurs together with the designation; that wholesome and unwholesome states arise in one who attends to that base with the perception person; and that for this reason puggalopi upanissayapaccayena paccayo is said. (The paraphrase is ours.) On this reading the person is the aggregates as grasped, which is mind and matter, not a bare concept.
The definitions were read into a clause table by scripts/relata.py. Each row records the condition, the conditioning term, the conditioned term, the kind of each (mind N · matter R · concept P), the timing only where the clause's own words state it, the CST line, and a note marking every interpretive step. A --verify mode checks that each row's probe phrases occur on the cited line; at commit 8663f69 it reports 199 clauses, 222 probes, 0 missing. By source the 199 rows divide as follows (counted from matika/RELATA.md).
source rows what
-------------------------------------- ----- ----------------------------------------------
Paṭṭhāna definitions (niddesa) 49 rows 1–51 less the two person readings
the person of :165 as designation 1 a reading (the Puggalapaññatti's own title)
the person of :165 as aggregate-continuum 1 a reading (the sub-commentary, abh01t:2557)
detailed section, open-side rule 42 consulted for conditions whose definition is open
detailed section, full read 101 the other conditions, so none is read selectively
commentary, distributive 2 abh03a:7889 (object), abh03a:8057 (decisive support)
commentary, collective 3 abh03a:7817 (object, predominance, decisive support)
----
199The kinds follow a rule fixed before the second count. An open side is a whatever phrase, a relatum the clause does not name, or the class term abyākata (indeterminate), whose extension includes matter (sabbañca rūpaṃ, asaṅkhatā ca dhātu, abh01m.mul.xml:4669). An open side is read literally, as taking every kind it could name; and the detailed section is consulted for every condition that has one, where its enumeration replaces the open side. This is theorem consulted_where_open in Relata.lean: the conditions with detailed-section rows from the open-side rule are exactly the conditions whose definition has an open side. Nibbāna is counted under mind, by the Dhammasaṅgaṇī's own definition of nāma, which includes the unconditioned element (asaṅkhatā ca dhātu – idaṃ vuccati nāmaṃ, abh01m.mul.xml:8457). The classifier then assigns a condition to a group by the kinds that appear on each side; only "concepts, mind and matter → mind" admits a concept, and it requires one.
Result (vipāka) has no open side, so the open-side rule did not reach it. Its detailed-section clause (§428, abh03m7.mul.xml:6093) was consulted because the definitions-only count failed on it. That choice was selected by the outcome, and the table keeps result's clause in a source of its own so the counts show it. The full read that followed closed the gap: the detailed section was read for the other 19 conditions as well, so that no condition is read selectively. The Lean theorem full_read_moves_only_result states what it changed: of the placements the full read could move, only result's moves, under both readings of all dhammas.
The rule was "fixed before counting" only in the sense that it was fixed before the second count. The first count, which the rule replaced, had already been made, and was withdrawn (§6).
The manual's grouping (Saṅgaha ch. 8 §15–20 and §26) is the answer key, entered in kala/ANSWERS.md. Two things follow from holding it out.
First, a reading may not cite it. The repository's rubric item 7, extended on 2026-10-13 to capture code, forbids a script that builds a data table from reading a held-out answer into that table. relata.py carries a function answer_free() that matches any row whose text cites the manual's §15–26 or its lines abh07t.nrf.xml:2840–2901; it runs on every build and under --check in continuous integration. The case that justified it: the first version of the table had read decisive support's conditioned side as "mind, as the Saṅgaha (§20)", and the function flags that old row.
Second, a sorting error was found by holding out. Before this work the grouping itself sat among the repository's axioms, as a premise. It is a computation from the definitions, not a premise for reasoning about them, and it was moved to the held-out file; two axioms that had been cited to the manual were re-cited to the Paṭṭhāna (A82, A90).
Each version of the table was handed to a reviewer instructed to find the case that kills it. Two refuters shaped this paper, and the withdrawals they caused are reported in §6.
The paper's method contribution is a list of faults the instruments had, each of which was seen to fail on a case planted for it, and each of which was repaired. The list is a finding (F10, §7), not an endorsement. The refusal and detection paths are now planted once each in a self-test script (scripts/selftest.py) that continuous integration runs.
answer_free().2026-10-13-kv-8-2-antarabhava-control), an axiom row cited a line of the held-out answer's own passage (abh05t.nrf.xml:2409). The row was moved to the held-out file, and an overlap guard now enforces it; a planted citation fires the guard.kala/THEOREMS.md. ⚠️ This repair is to the proof-run harness. matika/run.sh applies each break to a copy of Relata.lean and requires only that the file as a whole then fail to compile; that each break fails "on its own theorem" is recorded in the commit messages, not enforced by the script (§9).Only item 1 governs the data of this paper directly; item 3 governs how its breaks are to be read. Items 2 and 4–7 were found in the runs that surround it and belong to the same instrument. The reader should weigh them as a record of how often an instrument that had only passed turned out to be wrong.
Applying the standing test, remove the enforcer; does the guard break? — consulted_where_open and full_read_moves_only_result are properties of the table: change a row and the Lean check fails with no one watching. answer_free() is a rule enforced by the continuous-integration job, and it sees only the pattern it contains; a reading that paraphrased the manual's grouping without its section number would pass it. That limit is stated in §9.
Every row is labeled with its source, and so with its layer: Paṭṭhāna (Piṭaka) · Dhammasaṅgaṇī (Piṭaka) · Dhammasaṅgaṇī sub-commentary · Paṭṭhāna commentary (CST file abh03a.att.xml). A count that needs a commentarial row says so.
Reading only the definitions (the 49 rows of the first source in §3.1, with no person reading), with every open side read literally, the classifier places 19 of the 24 as the manual does. The five it does not place are named by theorem niddesa_alone_19:
object · proximity · contiguity · decisive support · resultFour of the five are conditions with an open side. Result is the fifth: the definitions give it as mind to mind (theorem niddesa_vipaka_is_mind_to_mind; the table's row for abh03m7.mul.xml:193 names only the mental aggregates), and the manual places it as mind to mind and matter.
placed of 24
definitions alone (niddesa) 19
+ detailed section, wholesome triad
(open-side rule + full read, §428 for result) 22 ← under BOTH readings of "all dhammas",
no reading of the person, no commentary
+ a reading that supplies a concept 22 … 24 ← six readings, §5.3Theorem the_22_without_concepts states that every condition other than the object and decisive support is placed as the manual places it, with the definitions and the detailed section as the only sources, under the exclusive reading of all dhammas (concepts not among them) and under the inclusive reading (concepts among them, abh01m.mul.xml:8453). The complementary theorem the_two_need_a_concept states that on the same sources the object and decisive support are the conditions not placed, and no others.
What the detailed section contributes is specific.
:85–93, :97: purimā purimā kusalā… akusalā… abyākatā dhammā, and yesaṃ yesaṃ dhammānaṃ anantarā). The detailed section's proximity clauses (abh03m7.mul.xml:5669–5725) name aggregates. For example the indeterminate-to-indeterminate clause reads purimā purimā vipākābyākatā kiriyābyākatā khandhā pacchimānaṃ pacchimānaṃ vipākābyākatānaṃ kiriyābyākatānaṃ khandhānaṃ anantarapaccayena paccayo (:5685). The table reads contiguity's parallel clauses the same way. Both are therefore mind to mind.:165 names none). The detailed section's decisive-support clauses (:5845–5981) name mental states on the conditioned side in every row of the table.:6093) states vipākābyākato eko khandho tiṇṇannaṃ khandhānaṃ cittasamuṭṭhānānañca rūpānaṃ vipākapaccayena paccayo, then the parallel at rebirth with kamma-born matter, and ends Khandhā vatthussa vipākapaccayena paccayo, "the aggregates are a result-condition of the base". It is the one row in the table that gives result matter (the other triads were not read, §9), and without it result is placed in the wrong group.The finding to hold: what the first version of the table had taken from the manual's own answer (that proximity, contiguity and decisive support's conditioned side are mental) is found in the Paṭṭhāna's detailed section. The manual's grouping did not have to add it.
Neither of the two remaining placements follows from the Paṭṭhāna's explicit enumerations alone, because the group they sit in is defined by the conditioning term including a concept, and the enumerations read here name none (a stem search for puggal- and paññatti in the detailed section's lines 5560–6360 finds no occurrence of either). Three readings supply one.
| Reading | The choice | What it places |
|---|---|---|
All dhammas (abh03m7.mul.xml:49) | Inclusive: the Dhammasaṅgaṇī calls designations dhammas (abh01m.mul.xml:8453, ime dhammā paññatti). Exclusive: the commentaries set concepts apart from the mind-and-matter the conditions relate. | the object (inclusive) |
The person (:165) | A designation (the Puggalapaññatti's own title) or the aggregate-continuum under the perception person (abh01t.tik.xml:2557). The Paṭṭhāna commentary says of the pair, person and lodging, that they are stated as natural decisive support, and that their status as a condition is to be understood figuratively (Puggalopi senāsanampīti idaṃ dvayaṃ pakatūpanissayavasena vuttaṃ… Paccayabhāvo cassa panettha pariyāyavasena veditabbo, abh03a.att.xml:8053). | decisive support (designation) |
| The commentary | Distributive: the object is sevenfold, and designation is one of the seven (Sabbameva vā etaṃ kusalākusalavipākakiriyarūpanibbānapaññattibhedato sattavidhaṃ hoti, abh03a.att.xml:7889); decisive support is the four planes' dhammas saddhiṃ ekaccāya paññattiyā, "together with some designation" (:8057). Collective: object, predominance and decisive support are time-free because concepts are included with nibbāna (paññattiyā saddhiṃ nibbānassa saṅgahitattā kālavimuttāpī, :7817). | both (distributive); and it unseats predominance (collective) |
The commentary's two statements need separating. At :7889 the object's division into seven is the commentary's voice, and it is preceded by tattha ṭhapetvā paññattiṃ avasesaṃ, "leaving concepts aside, the rest…": it counts designation among the kinds of object and then divides the remainder by plane. At :7817 it speaks of three conditions together (ārammaṇapaccayo adhipatipaccayo upanissayapaccayoti ime tayo paccayā tekālikāpi honti), which it says are of all three times and also free of time, because concept is included with nibbāna. If the clause is read of all three conditions, predominance receives a concept too, and predominance then no longer sits in the manual's ninth group (mind and matter to mind and matter). If it is read as the commentary reads the object and decisive support at :7889 and :8057, it does not. Both readings are in the table as separate rows, because the commentary does not say which it means.
The counts, by reading (theorem counts_by_reading):
reading placed of 24
-------------------------------------------------------------- ------------
exclusive "all dhammas" · person as aggregate-continuum 22
exclusive · person as designation 23
inclusive "all dhammas" · person as aggregate-continuum 23
inclusive · person as designation 24
exclusive · continuum · commentary, distributive 24
exclusive · continuum · commentary, collective 23and the theorems that say why:
object_by_reading: among the sources in the table, the object is placed by the inclusive reading of all dhammas or by the commentary's concept at :7889, and by no other combination.decisive_support_by_reading: decisive support is placed if the person is a designation; read as the aggregate-continuum it is not placed until the commentary's "some designation" at :8057.collective_commentary_unseats_predominance: read collectively, :7817 gives predominance a concept and removes it from the ninth group.The layer matters. Rows 1–4 use the Piṭaka (the person as a designation is the Puggalapaññatti's title); the continuum reading of the person is the sub-commentary's; rows 5 and 6 need the commentary.
The detailed section has no slot for a designation: its indeterminate class is, by the Dhammasaṅgaṇī, result, functional states, all matter and the unconditioned element (abh01m.mul.xml:4669), and no designation appears in it. So its failure to give the object a concept is not evidence that the object has none. It is the silence of a text whose classes were not built to carry one. This paper does not treat the silence as a negative, and does not treat it as a positive either; it keeps the object's and decisive support's placements as readings.
For completeness: the manual's own list under natural decisive support (abh07t.nrf.xml:2897) repeats the Paṭṭhāna's items, Rāgādayo pana dhammā saddhādayo ca sukhaṃ dukkhaṃ puggalo bhojanaṃ utusenāsanañca, and so carries the person into the manual. That paragraph is outside the range held out as the answer key (§15–20 and §26), which is why it can be read here; no row uses it.
The table was corrected twice on 2026-10-13. Both claims are reported here because the instruments caught them, not the authors.
Withdrawal 1: "23 of 24 from the niddesa alone." The first version of the table (commit bf3cad6) reported that the definitions alone place 23 of the 24, result being the exception. Its refuter showed three defects:
:61 and as mind only at :97;The correction is the single rule of §3.1, applied to every open side. Under it the definitions alone place 19, not 23 (§4, niddesa_alone_19).
Withdrawal 2: "24 of 24 from the Piṭaka alone." The second version (commits 9d1dff0, 0dcae15) took the rule, added the detailed section, and reported 24 of 24 placed on the inclusive reading of all dhammas, "from the Piṭaka alone". The full-read refuter confirmed all 101 full-read rows and withdrew the count. It rested on two choices presented as one result: reading the person at :165 as a designation, and admitting the commentary at :7889 but not at :7817, where, read collectively, it gives predominance a concept and unseats it. The correction (commit 350dff0) states what is robust, the 22 of §5.2, and moves the object and decisive support to the three readings of §5.3, each encoded both ways. The figure 24 is not wrong arithmetic: it is the count on the inclusive reading with the person as a designation. What was withdrawn was the sentence saying it was what the Piṭaka says.
The history is in the artifact itself: matika/RELATA.md carries both withdrawals in a paragraph headed "History", and the commits above carry the refuters' findings in their messages. The refuters' own reports were not found as separate files in the repository, so this section rests on those messages and on that paragraph.
The findings are numbered. Each carries the break that would overturn it. Every count is read from matika/RELATA.md and matika/Relata.lean at commit 8663f69; the theorem names are those of the Lean file.
F1 · The definitions alone place 19 of 24. The five unplaced are the object, proximity, contiguity, decisive support and result (niddesa_alone_19). Break: any uniform rule for reading the open sides, fixed without reference to the key, that places more of the five from the definitions alone. The rule of §3.1 is one such rule; a different one would change this number, which is what the first refuter showed the first version lacked.
F2 · The detailed section, not the definitions, closes proximity, contiguity and the mental side of decisive support. The relevant clauses name aggregates or mental states on the conditioned side (abh03m7.mul.xml:5669–5725, :5845–5981). Break: a clause in those ranges naming matter or a designation on the conditioned side, or a reading of khandhā in them as including matter. The repository's break "a detailed-section proximity clause given matter" makes the check fail, so the check does depend on this.
F3 · In the table, result's matter comes from §428 and from no other row. The definitions give result mind only; §428 (:6093) gives it mind-born matter, kamma-born matter at rebirth and the base. Break: a result clause elsewhere in the Paṭṭhāna naming matter, or a reading of the definitions' result clause as including matter. The break "§428's result clause removed" fails the check.
F4 · Of the 19 conditions the open-side rule did not reach, the full read moves none but result. Under both readings of all dhammas (full_read_moves_only_result). Break: a change to any of the 101 full-read rows that moves another condition's placement. The result is conditional on the table; a different reading of a row can move it.
F5 · 22 placements follow from the Paṭṭhāna's definitions and the wholesome-triad detailed section under both readings of all dhammas, with no reading of the person and no commentary (the_22_without_concepts). Every one is a placement with no concept in it. Break: any reading of the table's rows that moves any of the 22, or an edition of the Paṭṭhāna (§9) that reads differently on the lines cited.
F6 · The two remaining placements, the object and decisive support, need a concept, and the enumerations read here name none. Without a reading that supplies one they fail (the_two_need_a_concept). Break: a Paṭṭhāna passage, not in the detailed section's wholesome triad, naming a designation as object or as decisive support; the other triads were not read (§9).
F7 · The object is placed by the inclusive reading of all dhammas or by the commentary's concept, and by no other combination of the sources in the table (object_by_reading). Decisive support is placed if the person is a designation; read as the aggregate-continuum it is not, until the commentary's "some designation" (decisive_support_by_reading). Break: a source for a concept in the table that this paper did not read; a reading of :165 that gives the conditioned side a concept.
F8 · The commentary's statement at abh03a.att.xml:7817, if read of all three conditions it names, removes predominance from the manual's ninth group (collective_commentary_unseats_predominance). Break: the commentary's own reading of its sentence. This paper does not decide it; a reader of the Burmese commentarial tradition could.
F9 · The count, by reading, is 22 · 23 · 23 · 24 · 24 · 23 (counts_by_reading, §5.3). Break: a seventh reading, or a row's re-reading; the table is ours (§9).
F10 · Instrument faults, each seen on a planted case (§3.4): capture code that read the key; an axiom cited to the key's passage; breaks that could not isolate a theorem; a leak check whose one-step threshold said "no" to a proven entailment; a verbatim check that was not anchored; and the limit of known-answer controls. Break: an instrument that passes the planted cases and still misses a fault of one of these kinds on the data of §5.
F11 · Two claims about the same table were withdrawn on one day, each by a refuter's finding (§6). Break: none; this is a record. It is listed because a study that reports only the survivor misreports its own reliability.
What this paper does not find: that the manual's grouping is the right grouping, that the manual derived it from the Paṭṭhāna (the manual's paragraphs read here, §15–26, cite no Paṭṭhāna passage), or that any of the three readings of §5.3 is the correct one.
abh07t.nrf.xml:2845–2901), states it. The modern guides reproduce it. Bodhi (ed.), A Comprehensive Manual of Abhidhamma (1993), ch. 8, explains it in English; Nārada's A Manual of Abhidhamma, ch. VIII (fetched 2026-11-03 at budsas.org/ebud/abhisgho/abhis08.htm), gives the four sentences quoted in §1. This paper reproduces it.--verify mode shows that a phrase is on the line, not that the row means what the Pāli means.05d5d3c. The line numbers are the working file's, and the Cambodian, Sinhala and Thai editions were not collated for these passages.matika/run.sh was run under Lean 4.34.1 at SiliconWat/formal-abhidhamma commit 8663f69 (Relata.lean last changed at 350dff0): the checks pass and every deliberate break fails as required. relata.py --check and --verify reported the generated files current and 199 clauses, 222 probes, 0 missing. A reader can repeat all of this.run.sh require the file to fail, not a named theorem. Each try applies one deliberate fault to a copy of Relata.lean and requires the compile to fail. That a given break fails "on its own theorem" is recorded in the commit messages and is not enforced by the script (§3.4, item 3).abh01m.mul.xml:8457), shared by the manual's grouping; a reading that placed it apart would change which rows carry which kind.answer_free() is a pattern match. It rejects a row that cites the key's section numbers or lines. A row that restated the key in other words would pass it. The refuters, not the function, are the defense against that, and each refuter is one reader.The commentary on the Paṭṭhāna already asks the object's question of itself: of the kinds of thing that can be an object, does a designation belong? It answers it in its own sentence, and the sub-commentary asks what the sentence about the person can mean. The commentary's marking of a reading it does not press, pariyāyavasena, "by way of a figure", is the move this paper's two-way encodings make mechanical: it differs only in saying exactly what each reading changes in a count.
The series' gate for a satellite paper is that the canonical material must yield a usable engineering artifact. The artifact here is the table of 199 rows with its checker, which is a second instance of the method of №4, applied to a different chapter of the manual and, this time, to the canon beneath it.
| Venue | Identifier |
|---|---|
| Primary canonical | <https://thonly.org/research/where-the-manual-adds-a-reading> |
| GitHub (paper) | <https://github.com/thonly/publications/blob/main/defensive-publications/where-the-manual-adds-a-reading.md> |
| GitHub (artifact, CC0) | <https://github.com/SiliconWat/formal-abhidhamma> (matika/RELATA.md, matika/Relata.lean, scripts/relata.py; files cited at commit 8663f69) |
| Software concept DOI (the formalization) | 10.5281/zenodo.23067241 (Formal Abhidhamma: the Theravāda Abhidhamma in Lean 4, v0.1.0) |
| Series, №4 | The Counts Check (the-counts-check), concept DOI 10.5281/zenodo.23020670 |
| Census pre-registration (public) | SiliconWat/formal-abhidhamma commit 4956261 |
| Independent timestamps | an OpenTimestamps proof anchored in Bitcoin, and RFC 3161 tokens from three timestamp authorities, one of them eIDAS-qualified; a timestamp proves that this exact text existed by its date and nothing about authorship, originality, or validity |
The author thanks the Vipassana Research Institute for the Chaṭṭha Saṅgāyana texts it gives away, and his father, with whom the transcription of the Cambodian Tipiṭaka proceeds page by page. The findings in §6 belong to the reviewers who made them.
github.com/VipassanaTech/tipitaka-xml (text commit 05d5d3c), romn/abh07t.nrf.xml:2841–2901 (read and checked with the corpus tools, 2026-11-03).romn/abh03m7.mul.xml: definitions :37–265; detailed section, wholesome triad, :5569–6357; §428 at :6093 (checked 2026-11-03).romn/abh01m.mul.xml: §583 at :4669; §1315 at :8453; §1316 at :8457 (checked).romn/abh01t.tik.xml:2557; the paragraph read is the sub-commentator's own voice on the person (checked).romn/abh03a.att.xml:7817, :7889, :8053, :8057 (checked).budsas.org/ebud/abhisgho/abhis08.htm (fetched 2026-11-03: the four sentences of the grouping, quoted in §1).abhidhamma.com. (Pages 1–4 of the Paccayaniddesa were read in the 2026-10-13 census; not re-fetched for this draft.)thonly.org/research/the-counts-check. Concept DOI 10.5281/zenodo.23020670.thonly.org/research/patthana-typed-causation-vocabulary. (The 24 conditions as a typed-causation vocabulary; this paper's table is independent of its paraphrases.)thonly.org/research/abhidhamma-executable-process-specification.SiliconWat/formal-abhidhamma: matika/RELATA.md, matika/Relata.lean, matika/run.sh, scripts/relata.py, scripts/selftest.py, runs/PROVE-RUBRIC.md; commits bf3cad6, 9d1dff0, 0dcae15, 350dff0, census 4956261; run 2026-10-13-kv-8-2-antarabhava-control.