Dated manuscript production during Timur's campaign decades: a rise, not a trough

Provenance notice. This article exists because a conjecture — a falsifiable public prediction, staked in advance against a named dataset — was pre-registered (its kill/support clauses fixed before the decade-by-tradition computation ran) and killed, on the Inferpedia conjectures campaign's 1001-blind lane. The conjecture, fresh-fablemax-20260705-006, "Timur's fingerprint", remains killed and untouched at its own L1 lead status; nothing here reopens or revises it. The triage verdict on this conjecture was "adjacent" (the underlying Armenian production-decline-under-Timur observation is known historiography; the cross-tradition synchronized-trough statistic was not located), so the resolution is a calibration test of the instrument, not a novel-discovery claim, and this article makes none. This article was published in beta after a recorded Fable publication judgement (docs/generated/publication_judgement_615_20260716.json); every figure was recomputed from the committed decade-series artifact before publication.

Epistemic status

This article reports a measured pattern in surviving records: the numbers are directly attested in the cited datasets; what they mean is an open interpretation.

The pattern here is unusually simple to state and unusually easy to misread: a conjecture that predicted a production TROUGH during Timur's campaign years was pre-registered and killed because the counts in the cited catalogue roughly DOUBLE over the same years, in both traditions tested. That reversal is directly counted, not inferred. What remains genuinely open, and is the entire second half of this article, is what the doubling means: whether it reflects Timur-era manuscript production at all, or whether it is an artifact of how a modern catalogue of surviving, colophon-dated manuscripts is built.

Summary

In the vHMML (Hill Museum & Manuscript Library) catalogue of dated Eastern Christian manuscripts, Syriac and Armenian dated-manuscript counts were compared across two windows: a 1340s–1370s baseline (four decades) and a 1390s–1400s "trough" window (two decades), the window in which a conquest-driven production collapse was predicted. Syriac production rose from a baseline mean of 6.5 records/decade to a trough-window mean of 13.0 (+100%). Armenian production rose from 6.25 to 13.5 (+116%). Ethiopic production, included as an out-of-theater control expected to stay level, rose from 2.75 to 11.0 (+300%) — but its baseline sits below the 5-records/decade floor the pre-registration itself set as too thin to interpret. The conjecture predicted a synchronized DIP of at least 40% in both campaign-zone traditions; the record shows a synchronized RISE in both. The pre-registered kill clause — a rise, or a dip smaller than 15%, in either tradition — fires cleanly, and this is the first killed verdict in this lane's public record.

What is being inferred

Nothing about Timur's actual military or economic impact on manuscript production is asserted here, in either direction. The measured pattern is a catalogue statistic: dated-manuscript RECORDS in one library's metadata, not manuscripts PRODUCED. The gap between those two things is the article's real subject in its second half — vHMML dated counts measure catalogued survival within one institution's collecting and cataloguing scope, not production, and a genuine production collapse could in principle be masked or even inverted by survival and dating gradients (later manuscripts tend to survive better and be more often precisely, colophon-dated than earlier ones). This article does not conclude that Timur's campaigns had no effect on manuscript production; it concludes that this particular catalogue, read this particular way, does not show the predicted trough, and offers reasons the catalogue might not show a real trough even if one existed.

What is attested

A conjecture was pre-registered (fresh-fablemax-20260705-006, registered 2026-07-05T12:12:34Z): dated records were assigned to a decade by date midpoint; for each of three traditions (Syriac, Armenian, Ethiopic) the pre- registration fixed a baseline mean (1340s–1370s, four decades) and a trough-window mean (1390s–1400s, two decades), with SUPPORTED requiring both Syriac and Armenian to fall at least 40% below their own baselines while Ethiopic did not, KILLED requiring either Syriac or Armenian to show no dip of at least 15%, and INCONCLUSIVE otherwise — with an explicit additional floor, set at registration, that the test is INCONCLUSIVE BY DESIGN if any tradition's baseline mean falls below 5 records/decade (too thin a sample to interpret).

The computation (2026-07-05T12:13:18Z): Syriac baseline 6.5/decade, trough 13.0/decade (+100%); Armenian baseline 6.25/decade, trough 13.5/decade (+116%, with a noted internal spike — 24 records in the 1390s decade followed by only 3 in the 1400s decade); Ethiopic baseline 2.75/decade, trough 11.0/decade (+300%). The kill clause fires: neither Syriac nor Armenian shows any dip in 1390–1410 — both roughly double relative to their own baselines. The verdict is KILLED.

The resolution record names, honestly, a drafting defect in the pre-registration itself: the inconclusive-by-design floor ("any tradition's baseline mean below 5/ decade") also fires, via the thin Ethiopic control (2.75/decade baseline) — and clause priority between the kill clause and the inconclusive-by-design floor was not specified at registration. The recorded resolution reasons through this rather than silently picking a side: the kill clause concerns the two campaign-zone traditions themselves, both adequately sampled (6.5 and 6.25 records/decade baselines, well above the 5-record floor), and both move opposite to the prediction; the ambiguity belongs to the control arm, not to the tested traditions, so the kill is recorded as "the only defensible reading" rather than a clean, unambiguous clause match. This article carries that qualification forward rather than smoothing it away.

Why infer this entity

This article promotes a measured reversal, not a reconstructed entity: a conjecture staked a specific, falsifiable direction and clause, in advance, against a named catalogue, and the catalogue said no. It is included in the promotion ladder because the kill-loop's discipline (fix the clause, compute once, report the verdict) produced a result whose own acknowledged imperfection — the unresolved clause-priority defect — is itself a useful, documented example of how a pre-registration can be honestly wrong about which of its own guard-rails will bind, without that flaw rescuing the underlying claim.

Evidence ledger

Two primary records, both committed to this site's own repository, were read directly: the pre-registration record for fresh-fablemax-20260705-006 (docs/generated/conjecture_predictions_batch2_20260705.json, row 1) and its resolution/verdict record (docs/generated/conjecture_resolutions_batch2_20260705.json, row 1), which in turn point to the full decade-series computation artifact (docs/generated/conjecture_extracts/batch2_resolutions_20260705.json, key "timur_fingerprint"). No live model call read or summarized any external web page for this promotion.

Counterarguments

The central caveat is instrumental, not incidental: vHMML dated counts are catalogued SURVIVAL within one library's collecting and cataloguing scope, not production. Later manuscripts tend to survive in greater numbers and are more often colophon-dated (precisely datable by an internal scribal note) than earlier ones, a gradient that runs in the same direction as the measured rise and could produce or inflate it independent of any real production trend. The Armenian series shows an internal pattern that invites suspicion on its own terms — 24 records in the 1390s followed by only 3 in the 1400s — a shape more consistent with a single well-documented decade than with a smooth doubling, and worth a closer per-year look this article does not attempt. The Ethiopic control's own baseline (2.75/decade) is thin enough that its designed role — showing that traditions outside the campaign zone stay level — cannot be read with confidence either way; the drafting defect named above means the control's failure to clear the sample-size floor was not anticipated when the kill/support clauses were written. And catalogue coverage generally is not production: HMML's collecting scope, cataloguing priorities, and the underlying digitization program all intervene between an ancient manuscript being made and a dated record for it appearing in this dataset.

Confidence scores

• Direct attestation of the decade counts and the clause outcome: 85
• Existence warrant (this is a real, reportable pattern in the catalogue): 78
• Specificity: 74
• Reconstruction dependence (how much rests on interpretation beyond the counted rise): 35
• Counterevidence pressure: 45

What would change the score

A production-side dataset independent of a single library's cataloguing scope — ideally one that could distinguish manuscripts made in 1390–1410 from manuscripts made earlier but dated or catalogued later — would directly address the survival-versus- production caveat that is this article's main open question. A year-by-year (rather than decade-by-decade) reading of the Armenian series would clarify whether the 1390s spike is a broad trend or a single well-documented event. And a version of this test run with the clause-priority ambiguity resolved in advance — stating explicitly whether a thin control arm should block the whole verdict or only its own comparison — would remove the drafting defect this article otherwise has to carry forward honestly rather than paper over.