Inferpedia - an encyclopedia of the missing

AI-generated conjecture · below the evidence/publication boundary

← All conjectures · Water, land & settlement

Zipf in the archive

Status: Prior

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)

Zipf in the archive. Mesopotamian city sizes inferred from tablet volumes obey Zipf's law, with the exponent flattening (primate-city dominance) under imperial integration. Falsify: archive-volume rank-size by period.

Kill-dataset (verbatim)

archive-volume rank-size by period.

Provenance

Run: Imported conversation (verbatim harvest) · model: claude-fable-5

Origin: operator conversation with Claude Fable 5 at max effort, conducted 2026-07-03, relayed verbatim by the operator into the shepherd session on 2026-07-04. No ModelRun exists for the original generation (it happened outside the pipeline); this transcript file is the canonical capture. Transcript path: docs/generated/conjecture_harvest_fablemax_20260703.md. Model (operator-attested, not pipeline-recorded): claude-fable-5. Novelty disclaimer (verbatim, load-bearing -- rule 4): "Same caveat as before, doubled: at 100 items across all of archaeology and history, some of these will have cousins in the literature I can't check. What I can guarantee is the format — each links two things not normally linked, and each names the dataset or measurement that would kill it."

Novelty / leakage triage

Adjacent (closely related prior work exists)

Rank-size analysis of Mesopotamian settlement systems is an active method (mounded-area based, with A-coefficients and reported deviations from Zipf), but the specific proxy — city size inferred from TABLET VOLUMES, with the exponent flattening under imperial integration — was not located. Adjacent: the statistical frame is established on a different size proxy. In-house adjacent-variant calibration test registered on papyri.info provenience counts (Egyptian documentary record), clearly labelled as a variant, not the Mesopotamian claim itself.

Predictions

Supported registered 2026-07-05 variant: normalized-places-v2 calibration prediction (parent triage: leaked/adjacent)

Resolution: Supported

Caveats: Calibration verdict on the adjacent-variant dataset; still says nothing about Mesopotamian city sizes. THE SENSITIVITY HEADLINE: the Zipf-like rank-size structure is robust to the instrument - slope moves only -1.008 -> -1.056 when the known false-splits are fixed, and the strict-vs-binding gap (-1.053 vs -1.056) shows the residual-tier semantics change is nearly irrelevant; across raw strings, v1, and v2 the conclusion is stable, so the pattern is a property of the surviving record, not of the mapping. Remaining honesty: R^2 dipped 0.973 -> 0.953 (still above the bar); nome-level keys retained by design (arsinoites, grown to 5,515 by comma-twin folds, is a district not a settlement - a settlement-only re-test would be a further semantics change, not done here); flagged-not-fixed v2 imperfections remain (Dios Polis and Ptolemais bare-key homonyms, Aphrodito/Aphroditopolis fold, hermoupolis spelling splits - none in the binding top 10); excavation/publication bias untouched by any normalization; population drifted 79,215 -> 77,606 across the three tests via sanctioned concurrent scriptome work, recorded per-run.

Under mapping v2 (comma-split + Arsinoe consolidation with homonym guards + oder/alternate fixes + region-residual tier, built blind to counts; docs/generated/conjecture_extracts/papyri_place_mapping_v2_20260704.json, commit cfa5a69), the papyri.info top-50 settlement-level rank-size relation remains approximately Zipfian - the mapping-sensitivity test the v1 resolution's caveats left open. Criteria fixed verbatim in docs/GOAL_CONJECTURES_MAPPING_V2_20260704.md (commit e9e0338), BEFORE the v2 mapping was built.

Resolution criteria: Exclude 'unknown', 'ambiguous:' and 'residual:' keys; OLS of log10(count) on log10(rank) over ranks 1-50 of canonical keys: SUPPORTED if slope in [-1.4, -0.8] AND R^2 >= 0.90; KILLED if slope outside [-2.0, -0.5] OR R^2 < 0.80; else INCONCLUSIVE. Narrative (non-binding, must be reported): slope shift vs v1 (-1.0077); the strict-comparison number computed WITHOUT the residual: exclusion; arsinoe's consolidated rank and count; how many top-50 entries changed identity v1->v2.

Known priors disclosure: Seen at registration: the v1 result (slope -1.0077, R^2 0.9734) and v1's top-10 (oxyrhynchos 7,660; thebes 7,141; arsinoites 5,128; egypt 4,084; karanis 2,267; hermopolis 1,842; tebtynis 1,817; hermopolites 1,495; abu_mena 1,487; philadelphia 1,463); the v2 artifact's STRING-level stats only (978 keys, 324 changed strings, 48 strings folding to 'arsinoe', Peri Thebas split from 'thebes', 118 residual strings, 32 reviewed residual-token extensions). Direction genuinely unclear in aggregate: Arsinoe consolidation creates a large new key while the residual tier deletes v1's rank-4 'egypt' bucket and the Peri-Thebas split shrinks 'thebes'. NO v2 count, rank, or rank-size number has been seen. Population note: live papyri row counts have drifted twice today via concurrent sanctioned scriptome work; the computation will record the population it runs on.

OLS of log10(count) on log10(rank) over ranks 1-50, exactly as pre-registered in GOAL_CONJECTURES_MAPPING_V2_20260704.md (criteria committed e9e0338, before the v2 mapping was built; prediction registered before computation). Strict-comparison series (residual: keys included) computed as the mandated non-binding narrative. Script preserved at docs/generated/conjecture_extracts/v2_normalized_ranksize_20260705_script.py.

Dataset: papyri.info greco_roman_papyri records (77,606 with origin_place at computation time) joined to the committed blind-to-counts v2 mapping (papyri_place_mapping_v2_20260704.json, commit cfa5a69); 0 unmapped strings; binding series excludes unknown/ambiguous:/residual: (12,634 + 5,226 records); 978 canonical keys, top-50 used.

computed 2026-07-05

Supported registered 2026-07-04 variant: normalized-places-v1 calibration prediction (parent triage: leaked/adjacent)

Resolution: Supported

Caveats: Calibration verdict on the adjacent-variant dataset; still says nothing about Mesopotamian city sizes. The headline nuance is POSITIVE for the pattern: healing string fragmentation moved the slope from -0.898 to -1.008, essentially the canonical Zipf exponent, and consolidated Oxyrhynchos to rank 1 - the raw-string result was flattened by count-splitting, not inflated by it. Remaining distortions: the rank-4 'egypt' bucket (4,084 records) is a country-level residual, not a settlement (strings carrying no finer toponym), which the pre-registered criteria did not exclude; a handful of imperfect 'oder'-alternate folds exist in the key list (none in the top 10); the population changed 79,215 -> 77,606 mid-day via the sanctioned Wave-4a retag, so this measures normalization + population cleanup jointly; excavation/publication bias is untouched by normalization. Mapping-rule sensitivity untested (one mapping, one run - by design, to keep the pre-registration clean).

Under structure-based place normalization (the F1-A mapping, built blind to counts, docs/generated/conjecture_extracts/papyri_place_mapping_20260704.json), the papyri.info top-50 canonical-place rank-size relation remains approximately Zipfian - testing whether the raw-string result was an artifact of string fragmentation. Criteria fixed verbatim in docs/GOAL_CONJECTURES_FOLLOWONS_20260704.md (committed 0122a3d) BEFORE the mapping existed.

Resolution criteria: Exclude 'unknown' and 'ambiguous:' keys; OLS of log10(count) on log10(rank) over ranks 1-50 of canonical places: SUPPORTED if slope in [-1.4, -0.8] AND R^2 >= 0.90; KILLED if slope outside [-2.0, -0.5] OR R^2 < 0.80; else INCONCLUSIVE. Also report (narrative, non-binding): the slope shift vs the raw-string resolution (-0.8983), and Oxyrhynchos's consolidated rank.

Known priors disclosure: The registrant has seen the raw-string result (slope -0.8983, R^2 0.9748) and the rank-2/rank-5 Oxyrhynchos split; consolidation plausibly steepens the top of the curve - direction anticipated, magnitude unknown. Also known at registration: the mapping's coverage stats only (2,864 strings -> 1,062 canonical keys, 11.28% ambiguous, 1.12% unknown - counts of STRINGS, not records); a handful of imperfect 'oder'-alternate folds spotted in the key list (e.g. 'oxyrhynchos_oder'); and that the underlying population changed 79,215 -> 77,610 records mid-day via the concurrent Wave-4a retag of mis-traditioned European codex records, so this test is normalization + population-cleanup vs the raw run, not normalization alone. NO canonical-place record count or rank has been seen.

OLS of log10(count) on log10(rank) over ranks 1-50 of canonical-place record counts, exactly as pre-registered in GOAL_CONJECTURES_FOLLOWONS_20260704.md (criteria committed 0122a3d, before the mapping existed). Script preserved at docs/generated/conjecture_extracts/f1_normalized_ranksize_20260704_script.py.

Dataset: papyri.info greco_roman_papyri records (77,606 with origin_place, after the concurrent Wave-4a mis-tradition retag) joined to the committed blind-to-counts place mapping (papyri_place_mapping_20260704.json); 0 unmapped strings; 12,608 records (16.2%) excluded as unknown/ambiguous per pre-registration; 1,062 canonical places kept, top-50 used.

computed 2026-07-04

Supported registered 2026-07-04 calibration prediction (parent triage: leaked/adjacent)

Resolution: Supported

Caveats: Calibration verdict on an adjacent-variant dataset — this says NOTHING about Mesopotamian city sizes; it shows the Zipf-like rank-size pattern appears in the surviving Egyptian documentary record's place-count distribution. The slope sits near the registered band's shallow edge (-0.8983 vs the -0.8 boundary). origin_place is unnormalized free text and demonstrably splits real places across strings (Oxyrhynchos appears at BOTH rank 2 and rank 5 under different formattings), which flattens or distorts the measured slope by an unquantified amount; a normalized gazetteer join would be the serious follow-up. Record counts proxy excavation intensity, preservation (Fayum/Oxyrhynchus bias), and publication practice at least as much as ancient documentary output. CDLI lane ABSTAINED by pre-registration (ID-ordered non-random 14.2% batch, unfit).

ADJACENT-VARIANT calibration test, clearly distinguished from the harvest claim (which concerns Mesopotamian tablet archives): in the in-house papyri.info dataset (79,215 documentary records from Egypt), record counts per mapped origin place follow an approximately Zipfian rank-size relation over the top ranks. This tests whether the rank-size PATTERN appears in a documentary-survival record at all; it says nothing about Mesopotamia or ancient city sizes.

Resolution criteria: Exclude unmapped/'unbekannt'/unknown buckets. OLS regression of log10(count) on log10(rank) over ranks 1-50. SUPPORTED if slope is in [-1.4, -0.8] AND R^2 >= 0.90. KILLED if slope falls outside [-2.0, -0.5] OR R^2 < 0.80. Otherwise INCONCLUSIVE. The CDLI lane is ABSTAINED by pre-registration: the in-house batch is ID-ordered and non-random (14.2% of 421,501), unfit for population rank-size inference per the Phase-A extractor caveat.

Known priors disclosure: The registrant has seen the Phase-A summary only (79,215 records, 99.99% carrying an origin_place string, largest bucket 'unbekannt' at ~13.4%) and has NOT seen the place-count distribution or any rank-size plot. Standing caveat acknowledged at registration: record counts per place proxy excavation intensity, preservation (Fayum bias), and publication practice at least as much as ancient documentary output.

OLS of log10(count) on log10(rank) over ranks 1-50 after exclusions, exactly as pre-registered.

Dataset: In-house papyri.info (Wave 3a) documentary-record counts per raw origin_place string, greco_roman_papyri tradition (79,215 records, 2,974 distinct place strings), full count list pulled read-only at computation time; 42 unknown-type buckets excluded ('unbekannt'/'unknown'/'sine loco'/'?'), top-50 ranks used, exactly as pre-registered. ADJACENT-VARIANT test (Egyptian documentary record), not the harvest's Mesopotamian tablet claim.

computed 2026-07-04

Weigh in

No community feedback yet.

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.