Resolution: Killed
Caveats: PROPOSED BY THIS LANE, NOT ADJUDICATED -- the GM/Fable session must review before this verdict is treated as final (per the bridge goal doc's Pilot section, step 3). ACCESSION, NOT PRODUCTION (mandatory disclosure): FANKHA's counts measure what a 20th/21st-century Iranian national cataloguing project catalogued and how its cataloguers titled entries -- not medieval circulation or historical production directly; a 'killed' resting on this is a claim about the surviving-and-catalogued record's own title-dating pattern, not a secured claim about the true historical circulation curve. SINGLE-CATALOGUE REACH: this verdict rests entirely on one catalogue's (FANKHA's) own cataloguing conventions and holdings (57.47% of the in-house 'islamic'-tradition bucket, itself a coarser proxy for 'Persianate/Iran', not that region's full manuscript record) -- same A1/A2 discipline as the census pages requires saying so in the same breath as the verdict. UNIT MISMATCH WITH THE ORIGINAL VARIANT: this resolution's unit is catalogue records classified by TITLE alone (no author field exists in FANKHA, measured); it CANNOT reproduce the original fihrist-tei variant's per-manuscript, author-filtered COMPLETE/PARTIAL/MIDDLE classification, and instead tests a related but distinct operationalization (bound-title vs. split-title date clustering) of the same underlying conjecture -- a 'killed' here specifically kills THIS title-cataloguing-pattern test, not automatically the deeper codicological claim the original variant attempted and left inconclusive. HOMONYM CONTAMINATION (severe, only partially checkable): Population A (the 'Khamsa'/'Panj Ganj' bound-unit population, 335 records) is AT LEAST 21.5% (72/335) confirmed NOT Nizami's Khamsa -- FANKHA's own cataloguers cross-reference Amir Khusrau Dihlavi's khamsa to 'پنج گنج' via the title 'خمسه (خسرو دهلوي) = پنج گنج', directly contradicting this registration's own noetic prior that the bound-unit title defaults to Nizami; the KILLED direction and magnitude survive excluding these 72 rows (median gap -175 vs -183), but other rival khamsas (Khwaju Kirmani's, named in FANKHA but not exact-matching Population A's strict criteria) were not exhaustively ruled out. Population B (the 5 constituent-poem titles, 881 records) is a KNOWN, STRUCTURALLY UNCHECKABLE over-count relative to 'copies of Nizami's own text' -- FANKHA has no author field at all, so none of Layli u Majnun's, Khusraw u Shirin's, or the other titles' many non-Nizami imitators/predecessors (Hatifi, Jami, Hilali, Amir Khusrau, and, concretely found in this data, an apparently pre-Nizami 8th-century-CE Arabic Layli-Majnun tradition, min_year=740 CE) can be mechanically excluded; this was disclosed as a structural limitation at registration and confirmed as a real, non-hypothetical contaminant by this resolution, not merely a theoretical risk. DATA-QUALITY ARTIFACTS (disclosed, not corrected): Population B's reported max_year (5465) comes from one row with a visibly corrupted date_raw (a folio/page-count digit run swept into the Hijri parser); Population A contains at least 2 chronologically-impossible dated entries (755 CE and 1158 CE, both centuries before Nizami's own lifetime) whose cause (source cataloguing error vs. extraction error) was not determined; all are excluded in the disclosed robustness pass, which does not change the direction or order of magnitude of the finding. RECORDS-LANGUAGE LOCK: 'records' here are FANKHA catalogue rows, not physical manuscripts independently verified to exist outside this catalogue's own description of them (union-catalogue discipline: FANKHA describes copies its own contributing libraries reported, per docs/DECISION_FANKHA_MIRROR_INGEST_20260727.md). DATE-PRECISION LOCK: 33.5% (A) / 35.5% (B) of dated records carry only century-level precision, and the registered range-end convention (using date_end) systematically assigns such records their LATEST possible year -- applied symmetrically to both populations, so it should not by itself explain a directional gap, but it means the reported medians are not exact-year-precise for roughly a third of each population. WHAT WOULD CHANGE THIS VERDICT: an author-bearing FANKHA re-extraction (the volume-18 title;author-tail lead named in the works/authors probe report S2.2) that let Population B be author-filtered to Nizami specifically, and/or a hand-curated exclusion list for Population A's non-Nizami khamsas beyond the one confirmed here.
Registered as variant_key='fankha-v1' against the in-house iran-union-fankha catalogue (FANKHA, 323,111 ManuscriptRecord rows, tradition='islamic'), re-testing the SAME conjecture as the original variant_key='' prediction (registered 2026-07-16, resolved inconclusive-by-design: zero Nizami manuscripts securely dated before 1349 CE in the Fihrist-only in-house sample, a UK-collection artifact; that resolution's own caveats named exactly Iranian and Central Asian collections, e.g. the National Library of Iran, as the missing coverage). FANKHA is Iran's own national manuscript union catalogue and therefore the closest in-house instrument to that named gap, but it has one structural difference from the Fihrist test that this variant must work around rather than paper over: FANKHA carries NO per-work author field at all (measured, docs/generated/scriptome_works_authors_probe_20260729.md S2.0 -- every row's only author-shaped field is a constant citation-boilerplate string, byte-identical across all 323,111 rows). This variant therefore cannot replicate the original's per-manuscript, author-filtered COMPLETE/PARTIAL/MIDDLE classification; it instead tests the conjecture's own claimed date-stratification pattern ("single-poem circulation first, quintet codices arriving with the illustrated-atelier era") via a TITLE-LEVEL population comparison: catalogue records whose title is the bound quintet-unit ('Khamsa'/'Panj Ganj') versus catalogue records titled with one of the five constituent masnavis, and whether the bound-unit population's dated records cluster measurably LATER than the constituent population's dated records. Claim under test: the Khamsa is a bibliographic afterlife, not an authorial architecture -- FANKHA's own dated title-catalogue entries for the bound quintet-unit should be, in the aggregate, substantially later than its dated entries for the five poems catalogued separately, and the earliest dated attestation of any constituent poem should predate the earliest dated attestation of the bound unit.
Resolution criteria — the registered fine print
Resolution criteria: INSTRUMENT: source slug 'iran-union-fankha' (FANKHA, ManuscriptRecord rows, tradition='islamic', target_unit='catalogued_manuscript'). UNIT = catalogue records (one row = one title-heading-scoped copy entry; NOT a physical-manuscript row with unioned contained-work membership -- FANKHA's schema has no ContainedText layer and no cross-title shelfmark/repository join is attempted here, so this variant CANNOT reproduce the original fihrist-tei variant's per-manuscript COMPLETE/PARTIAL/MIDDLE classification; see known_priors_disclosure). NORMALIZATION: apply fold_arabic_script (NFKC + harakat-strip + tatweel-strip + alef-variant fold + ya/alef-maksura fold + teh-marbuta fold + kaf fold + hamza-carrier fold + bullet-strip + whitespace-collapse; docs/generated/scriptome_works_authors_probe_20260729.md S2.3, validated, zero false positives observed on 575 hand-checked clusters) to raw.work_title_raw (= ManuscriptRecord.title), THEN split on the cataloguer's own ' = ' alternate-title convention (S2.5: 13.0% of distinct FANKHA titles / 33.4% of all copies corpus-wide carry it) into segments; each segment tested independently. POPULATION A (BOUND UNIT, Grade A): a row where >=1 segment's folded form exactly equals 'خمسه' or 'پنج گنج' (both spellings of 'Khamsah'/'Panj Ganj' -- 'خمسة' folds to 'خمسه' via the teh-marbuta rule, so one target form covers both spellings). POPULATION B (CONSTITUENT, Grade A): a row where >=1 segment's folded form exactly equals one of: poem 1 'مخزن الاسرار' (Makhzan al-Asrar); poem 2 'خسرو و شیرین' or 'شیرین و خسرو' (Khusraw u Shirin, either citation order); poem 3 'لیلی و مجنون' or 'مجنون و لیلی' (Layli u Majnun, either citation order); poem 4 'هفت پیکر' or 'بهرام نامه' or 'هفت گنبد' (Haft Paykar / Bahram-nama / Haft Gunbad); poem 5 'اسکندرنامه' or 'اسکندر نامه' (Iskandar-nama, with/without internal space). A row matching both A and B (only possible across different '=' segments of one heading) counts as A only (a heading that explicitly asserts the bound-unit identity anywhere in its own alternate-title string is bound, not split) -- logged as an explicit override count, excluded from B. Each B row is tagged to every poem it matches; report both a per-poem breakdown and a pooled, row-deduplicated B total. GRADE B (narrative only, NEVER verdict-bearing): a segment that STARTS WITH one of the Population A or B target forms followed by a space and further text (e.g. a title explicitly qualified 'خمسه نظامی ...') -- reported separately since FANKHA's missing author field makes this weaker-population-but-stronger-per-row signal (an explicit textual Nizami-qualifier) worth surfacing even though it is not large or clean enough to anchor the verdict. HOMONYM DISCLOSURE (not resolved, cannot be mechanically resolved without an author field; applies to Population B only -- Population A's 'Khamsa' bound-title form is a weaker but real signal of Nizami-specific identity, being the eponymous eastern-Islamicate referent of the genre): the five constituent titles are historically NOT exclusive to Nizami. Khusraw u Shirin was also written by Hatifi (named, and confirmed as a real contaminant in the ORIGINAL Fihrist resolution's own independent-verification pass) and other imitators; Layli u Majnun is a pan-Islamicate legend retold by dozens of poets (Jami, Hilali, Amir Khusrau, Maktabi, Fuzuli among them), arguably the single most-imitated title in the set; Iskandar-nama collides with the separate, unrelated anonymous prose 'Iskandarnameh' romance tradition. Population B is therefore a KNOWN OVER-COUNT relative to 'copies of Nizami's own text', and this resolution does not attempt to net that out mechanically -- the verdict tests a bibliographic/cataloguing-pattern claim (does the catalogue's own bound-vs-split title practice show the predicted date stratification), not a strictly author-secure claim. This is a narrower, honest scope than the original fihrist-tei variant's, disclosed here and restated in the resolution's caveats. DATE: per row, dated_year = date_end if not null, else date_start if not null, else UNDATED (mirrors the original variant's 'ranges accepted, use the range end' convention for direct comparability; a date_start-based dated_year is also reported narratively as a bias check, since range-end dating skews the computed year later). WINDOWS (identical to the original variant, for direct comparability): PRE = dated_year < 1349 (CE; i.e. before 750 AH); NINTH_AH = 1398 <= dated_year <= 1495 (CE; i.e. 9th century AH). CLAUSE PRECEDENCE, evaluated strictly in order: (1) INCONCLUSIVE BY DESIGN if Population A has ZERO rows at all (dated or undated) -- FANKHA's cataloguers never use the bound-unit title form, blocking the comparison outright. (2) INCONCLUSIVE BY DESIGN if dated rows in Population A number fewer than 5, OR dated rows in the pooled Population B number fewer than 5 (thin-coverage guard; floor set at 5 rather than the original variant's 8 because this variant's unit is structurally different -- title-heading copy ROWS, not per-manuscript classifications over a fixed 163-row candidate pool -- so the original's floor does not transfer arithmetically; a fresh, disclosed choice, not an inherited one). (3) KILLED if, among all dated rows in both populations (any window), median(dated_year, Population A) <= median(dated_year, Population B) -- bound-unit copies are, on the whole, no later than split copies: the opposite of the predicted direction. (4) SUPPORTED if median(dated_year, A) minus median(dated_year, B) is >= 100 years AND min(dated_year, Population B) < min(dated_year, Population A) (the earliest attestation of any constituent poem predates the earliest bound-unit attestation). (5) otherwise INCONCLUSIVE (e.g. a positive but sub-100-year median gap, or the two sub-clauses of (4) disagree). NARRATIVE (non-binding, always reported): per-poem breakdown of Population B; PRE/NINTH_AH window counts for A and B reported side by side with the original resolution's own figures (population 163, dated 119, pre-1349 dated 0, ninth-AH tally COMPLETE 14 / PARTIAL 5 / MIDDLE 1 / ZERO 1); date_precision breakdown (exact/century/range/unknown) for A and B; the Grade-B qualified-prefix counts; the A/B override count; the composition clause (FANKHA's row count and its share of tradition='islamic' record count census-wide).
Known-priors disclosure — what the registrant already knew
Known priors disclosure: Seed: the conjecture's own triage (verdict: adjacent, shepherd tier, search_date 2026-07-16) records that Nizami's five poems' separate composition across three decades for different patrons, and the Khamsa as a later assembled package, are documented in F. de Blois, Persian Literature: A Bio-bibliographical Survey, vol. V (2004) -- the SAME seed as the original variant_key='' prediction; this fankha-v1 variant tests the identical underlying thesis via a new instrument, not a new claim. Resolving dataset exposure: (1) The ORIGINAL variant_key='' prediction and its resolution have been read in full (docs/generated/conjecture_prediction_khamsa_20260716.json, docs/generated/conjecture_resolution_khamsa_20260716.json): inconclusive-by-design, zero Fihrist-held Nizami manuscripts securely dated before 1349 CE among 163 in-house Fihrist rows (119 dated), the whole-population unwindowed tally PARTIAL 91 / COMPLETE 57 / MIDDLE 5 / ZERO 10, and the resolution's own caveats explicitly named Iranian, Turkish and Central Asian collections (Topkapi Sarayi, National Library of Iran, the Suleymaniye, IOM St Petersburg) as the missing coverage -- FANKHA (Iran's national union catalogue) is a direct, deliberate answer to that named gap, not an arbitrary new instrument choice. (2) docs/generated/scriptome_works_authors_probe_20260729.md has been read in full, sections 1 and 2 especially: FANKHA has 323,111 rows (confirmed stable this session, re-queried below), a validated fold_arabic_script normalizer (S2.3) and the cataloguer's own ' = ' alternate-title convention (S2.5) are both reused verbatim from that report's already-validated machinery, not re-derived here; the probe's own top-30-by-copy-count works table (S2.1) and every other title/genre example it quotes were re-checked by this registrant and contain NO Nizami/Khamsa/Layli-Majnun/Haft-Paykar/Iskandar-nama/Makhzan-al-Asrar title anywhere -- confirmed Khamsa-free by this registrant's own re-reading, not merely trusted from the design doc's own claim. (3) Immediately before this registration, three GENERAL (non-Nizami, non-Khamsa, no title-content filtering at all) schema-level queries were run against FANKHA and the wider scriptome DB, per the ordering rule: (a) reconfirmed iran-union-fankha's total row count -- 323,111, unchanged from the probe report; (b) the source x tradition breakdown for the whole scriptome DB (used for the composition clause below); (c) FANKHA's own date_precision distribution corpus-wide -- exact 145,358 / unknown 105,055 / century 72,527 / range 171 -- and date_start/date_end non-null count 218,056/323,111 (67.5%). ZERO queries filtering on title text, work_title_raw content, or any Nizami/Khamsa/poem-name string have been run against FANKHA before this registration. Composition clause (from query (b) above): tradition='islamic' totals 562,235 ManuscriptRecord rows across 7 sources (iran-union-fankha 323,111; hmml-vhmml 138,291; namami-kritisampada 80,957; fihrist-tei 15,187; bnf-manuscripts 3,227; papyri-info 1,450; fragmentarium 12) -- FANKHA alone holds 57.5% of the tradition='islamic' bucket. 'islamic' is the schema's actual (coarser 'Islamic world') tradition tag, the closest in-house proxy for 'Persianate/Iran', not an Iran-specific or Persian-language-specific tag -- it also includes Christian-Arabic HMML material and Indo-Persian NAMAMI rows, disclosed here so the 57.5% figure is not misread as 'FANKHA is 57.5% of the Iranian/Persian corpus' more narrowly than the tag actually supports. Noetic prior honestly held (NOT verified against FANKHA before registration): the registrant's own background knowledge of the specific canonical Persian/Arabic title forms used in resolution_criteria (the five masnavi titles, their known word-order variants, and the specific named poets -- Hatifi, Jami, Hilali, Amir Khusrau -- whose own same-titled works create the disclosed homonym risk) is standard literary-historical knowledge, not derived from this session's FANKHA queries; it was used only to author the matching spec, never to check FANKHA's actual title inventory before this row was created.
Method and dataset — how it was measured
Exactly the registered fankha-v1 criteria (packet prompt_hash 9baf27304185bf300c3d395163dda2cd29b520bc28187923e4a03592937326c3). fold_arabic_script (docs/generated/scriptome_works_authors_probe_20260729.md S2.3, transcribed verbatim; 10/10 fresh sanity checks against explicit \u-escaped codepoints passed before the real run, since presentation-form/bare-letter Arabic glyphs are not reliably eyeball-distinguishable) applied to each ' = '-delimited segment of every FANKHA row's title. Population A = any row with a segment folding exactly to 'خمسه' or 'پنج گنج' (335 rows). Population B = any row with a segment folding exactly to one of the 5 constituent-poem target forms -- both citation orders for poems 2/3, three name variants for poem 4, two spacings for poem 5 -- tagged per poem, deduplicated by row (881 rows). A takes precedence on the 7-row A/B overlap. dated_year = date_end if present else date_start (registered range-end convention, mirroring the original variant; the date_start-only variant also computed as a bias check). Windows PRE (<1349 CE) and NINTH_AH (1398-1495 CE) identical to the original fihrist-tei variant for direct comparability. Clause precedence evaluated exactly as registered; clause 3 (KILLED) fired: median(Population A)=1592 <= median(Population B)=1775. Compute: docs/generated/instrument_builds/khamsa/khamsa_fankha_compute.py (Sonnet). A DISCLOSED, NOT-pre-registered robustness/diligence pass then followed (mirrors the original variant's own independent-verification step): inspected the outlier dates and sample titles the primary compute surfaced, and re-ran the identical classification (a) excluding 1 row whose date_raw is visibly corrupted (a folio/page-count digit run swept into the Hijri month-day-year parser, yielding a nonsensical CE year of 5465 on an Iskandar-nama row) and excluding all dated_year<1160 CE rows on both populations (Nizami's earliest attributed work, Makhzan al-Asrar, is conventionally dated ~1166 CE per de Blois vol. V -- the conjecture's own seed source -- so nothing genuinely his, nor a bound assembly of his work, can predate ~1160 CE on any reading); and (b) additionally excluding 72 Population-A rows that FANKHA's own cataloguers explicitly cross-reference to Amir Khusrau Dihlavi's (not Nizami's) khamsa via the title 'خمسه (خسرو دهلوي) = پنج گنج' (found by inspecting Population A's sample titles post-hoc, not anticipated at registration -- a direct, concrete instance of the homonym risk the registration disclosed in the abstract). The KILLED direction and approximate magnitude (median gap -183 as-registered; -183 after date-cleaning; -175 after also removing the Amir-Khusrau-qualified rows; -116 using the date_start bias-check variant) survive every cut. Compute: docs/generated/instrument_builds/khamsa/khamsa_fankha_robustness.py (Sonnet). computed_at postdates registered_at (rule 6).
Dataset: In-house iran-union-fankha catalogue (FANKHA, apps.scriptome): 323,111 ManuscriptRecord rows (CatalogueSource 'iran-union-fankha', tradition='islamic', target_unit='catalogued_manuscript'), Iran's own national manuscript union catalogue (Dirayati/NLAI, vols. 1-34 ingested 2026-07-27 via the Ghaemiyeh/archive.org mirrors, docs/DECISION_FANKHA_MIRROR_INGEST_20260727.md). FANKHA carries NO per-work author field (measured, docs/generated/scriptome_works_authors_probe_20260729.md S2.0 -- its only author-shaped field, raw.attribution, is a constant citation-boilerplate string identical across all 323,111 rows); this resolution is therefore a TITLE-LEVEL population comparison (bound quintet-unit title vs. the five constituent masnavi titles) over ManuscriptRecord.title (= raw.work_title_raw), not an author-filtered per-manuscript classification like the original fihrist-tei variant. FANKHA holds 323,111/562,235 (57.47%) of the whole census's tradition='islamic' bucket (the schema's coarser 'Islamic world' tag -- the closest in-house proxy for 'Persianate/Iran', also covering Christian-Arabic HMML and Indo-Persian NAMAMI material) -- row counts reconfirmed stable both at registration and at compute time.
computed 2026-07-29