The Kitab's nameless witnesses

Status: Anticipated · untested

Status is derived only from the shepherd-authored triage/prediction data above -- community submissions and claims are a separate overlay and can never change it (see the participation panel below).

This is a proposed connection between two domains, generated by a language model. It is not an article and not evidence: it sits below the evidence/publication boundary. A quantitative prediction and a named kill-dataset are attached (when registered) so the claim stays falsifiable rather than merely evocative.

Claim (verbatim)

Sibawayhi's al-Kitab runs on two evidentiary regimes at once: a literary archive — roughly a thousand poetic shawahid, many with named poets — and live elicitation, the anonymous 'we heard some of the Arabs say'. If the book was built the way a fieldworker builds, the regimes should stratify by subject matter: syntax and poetic licence argue from the archive, where authority attaches to named poets; phonology and morphophonemics argue from the ear, where authority attaches to attested speech and a name adds nothing. Attribution density is then not scribal noise but a map of the work's evidentiary economy. Prediction: the proportion of anonymous testimonia (unattributed verses and 'some of the Arabs say' reports) will be at least twice as high in the phonological and morphophonemic chapters at the Kitab's end (imala, idgham, and their neighbors) as in the syntactic chapters of its opening third (primary clause: the twofold anonymity ratio; the verdict follows it). Exact computation: on the OpenITI text of al-Kitab, count attribution formulas — named citation verbs versus anonymous audition formulas — per chapter block, splitting the book at the transition into sarf and phonology. Kill: the shawahid indices of 'Abd al-Salam Harun's edition of al-Kitab and the OpenITI machine-readable text of the work.

Prediction clause (verbatim)

Prediction: the proportion of anonymous testimonia (unattributed verses and 'some of the Arabs say' reports) will be at least twice as high in the phonological and morphophonemic chapters at the Kitab's end (imala, idgham, and their neighbors) as in the syntactic chapters of its opening third (primary clause: the twofold anonymity ratio; the verdict follows it). Exact computation: on the OpenITI text of al-Kitab, count attribution formulas — named citation verbs versus anonymous audition formulas — per chapter block, splitting the book at the transition into sarf and phonology.

Kill-dataset (verbatim)

Kill: the shawahid indices of 'Abd al-Salam Harun's edition of al-Kitab and the OpenITI machine-readable text of the work.

Nobody has run this test. The kill-data is named above. If you can run it — or you know the paper that already settles it — claim the kill or submit the prior scholarship. Kills and prior scholarship are credited here, by name, as they come in.

Provenance

Run: Fresh agent generation · model: claude-fable-5

Fresh blind generation instance of claude-fable-5, 2026-07-16, wave M02 (the works) of the Minds & Works campaign, produced from model knowledge alone under the two-file blindness protocol.

Novelty / leakage triage

anticipated in the literature — this exact test has never been run

Sibawayhi's dual evidentiary regimes (named poetic shawahid vs anonymous 'we heard the Arabs say') are studied (Baalbaki), but the specific claim that anonymous testimonia are >=2x denser in the phonological/morphophonemic chapters than the opening syntactic third was not located as a measured test.

R. Baalbaki, The Legacy of the Kitab (Brill, 2008)

Predictions

Inconclusive registered 2026-07-16 calibration prediction (parent triage: leaked/adjacent)

Resolution: Inconclusive

Caveats: INCONCLUSIVE -- operationalization-unstable, DOWNGRADED from the compute's mechanical SUPPORTED by independent verification. The load-bearing statistic hangs on classifying every قال/أنشد/زعم citation into discourse functions -- named-poet vs anonymous-informant testimonia, but also teacher-authority citation, Qur'anic citation, hypothetical grammatical example ('[if] one says X'), and the dialectal-rule 'man qala X qala Y' device -- of which only two are 'testimonia' in the conjecture's sense. This was the single hardest judgement call in the build and the exclusion categories are NOT named in the registered criteria. Two careful independent efforts diverge sharply: the compute's elaborate exclusions give R = 2.12 (just over 2x), while an independent verifier's six from-scratch readings (two section-boundary methods x three classification strictnesses, including a deliberately generous stress test) all land R = 1.3-1.5, robustly BELOW 2x. The predicted DIRECTION is real and robust -- the phonology chapters ARE more anonymous than the syntax chapters (R > 1 in every reading) -- but the staked 2x MAGNITUDE is an artifact of one under-specified classification methodology, not a robust result. Honestly inconclusive. Independent re-derivation code in the build's independent_verification/ dir.

Registered against the OpenITI primary text of Sibawayhi's al-Kitab. Claim under test (primary clause): the proportion of ANONYMOUS testimonia (unattributed verses and 'some of the Arabs say' audition reports) is AT LEAST TWICE as high in the phonology/morphophonemics chapters at the book's end (imala, idgham and neighbours) as in the syntactic chapters of its opening third.

Resolution criteria: SHARED CONVENTIONS (identical across all rows of this packet). (a) TEXT SELECTION: use the PRIMARY OpenITI version of each named work (the '-ara1' primary per OpenITI convention, or the single most-complete version if no primary is flagged); if a work is split across several OpenITI version files, concatenate all parts of that same work's primary version in order. The build must FREEZE the exact OpenITI version URI(s) + a sha256 per work in a freeze_manifest before any metric is computed. (b) CLEANING: strip OpenITI mARkdown structural material (header block; lines/markers beginning with '#', '~~', page markers like 'PageVNN_NNN', 'ms', milestone markers 'Milestone###', editorial '@'-tags) to leave the running body text; strip Arabic tashkil (U+064B-U+0652, U+0670, U+0640 tatweel). (c) WORD = one whitespace-delimited token of the cleaned body text. (d) MARKER MATCHING is done on the tashkil-stripped text; alif/hamza variants are normalized for marker matching only (alif-forms أ إ آ ا -> ا) NOT for word counting. (e) All raw counts, rates, and the cleaned word totals per work must be reported alongside the verdict. (f) computed_at must postdate registered_at. || POPULATION: Sibawayhi's al-Kitab, single work, partitioned into section A = the SYNTACTIC opening third and section B = the PHONOLOGY/morphophonemics END (imala/idgham and neighbours). SECTION IDENTIFICATION (primary): by the work's own chapter (bab) structure -- A = chapters of the first third of the book by running word position; B = the chapters from the sarf/phonology transition to the end, located via chapter headings mentioning imala/idgham/idghаam and adjacent phonology topics. FALLBACK if headings are not machine-identifiable: A = first third by words, B = final third by words; the method actually used MUST be reported. MARKERS: ANONYMOUS-audition set (e.g. 'سمعنا', 'وسمعنا من', 'بعض العرب', 'بعض من نثق', 'حدثني من أثق به', plus shahid verses introduced without a named poet); NAMED-attribution set (a citation verb -- قال/أنشد/زعم -- bound to a proper name, or 'قال الشاعر' + name). ANONYMITY RATIO(section) = anonymous / (anonymous + named). PRIMARY = RATIO(B) vs RATIO(A). CLAUSE PRECEDENCE: (1) INCONCLUSIVE_BY_DESIGN if al-Kitab is absent, OR either section carries < 15 total testimonia, OR the two sections cannot be delimited by either method. (2) SUPPORTED if RATIO(B) >= 2 x RATIO(A). (3) KILLED if < 2x. Report RATIO(A), RATIO(B), the four raw counts, and the section-identification method used.

Known priors disclosure: Held: the conjecture's reasoning (syntax/poetic-licence argues from a named-poet archive; phonology argues from anonymous attested speech, so attribution density maps the work's two evidentiary regimes) and general knowledge al-Kitab is in OpenITI. NOT computed: the per-section attribution counts or anonymity ratios. The twofold threshold and the section topics are the conjecture's own.

Frozen OpenITI primary texts (sha256 in freeze_manifest.json; each work's identity independently re-confirmed from its own #META# header). Sonnet computed; a SEPARATE independent Sonnet verifier then re-derived every load-bearing number FROM SCRATCH with its own cleaning + tokenization (not permitted to read the compute's scripts/results), and stress-tested operationalization robustness. Verdict authored by Fable reconciling both. computed_at postdates registered_at (rule 6). R = anonymity_ratio(end phonology) / anonymity_ratio(opening syntax); R >= 2 for SUPPORTED.

Dataset: OpenITI text of Sibawayhi's al-Kitab, frozen. Anonymity ratio = anonymous / (anonymous + named) testimonia, computed for the opening-third (syntax) vs the end phonology/morphophonemics (imala, idgham, ...) sections.

computed 2026-07-16

Weigh in

No community feedback yet.

New here? Create an account first

Create an account or sign in and your feedback is tied to you — you can track it, get replies, and claim this conjecture so others know you’re working on it. Prefer not to? Just leave your take below as a guest — only the name you type is shown.

Add your take

Posted immediately (spam is removed). Community feedback is never an adjudicated verdict and never changes this conjecture's triage label or status above.