Rescue is a sampling method

Claim (verbatim)

In 2012-13, when armed groups occupied northern Mali, a reported three hundred thousand or more Timbuktu manuscripts were evacuated to Bamako in metal footlockers by Abdel Kader Haidara's network under SAVAMA-DCI, and when the occupiers withdrew they burned manuscripts at the state Ahmed Baba Institute (IHERI-AB) — a reported loss on the order of four thousand items, most of the institute's holdings having already been moved (the episode is documented in Charlie English's The Book Smugglers of Timbuktu and Joshua Hammer's The Bad-Ass Librarians of Timbuktu). The rescue was not a random sample: what could be evacuated was what was already inventoried, boxed, and known — the institutional and network-connected private libraries — while the deep tail of uncounted family libraries in Timbuktu and the surrounding desert could be neither burned on the record nor saved on the record, because no one held a list of it. The 2012 episode is thus a natural experiment in visibility: the rescued-plus-burned total is a lower bound on the known corpus, and the true corpus exceeds it by whatever the unlisted tail holds. Prediction: reconciling the SAVAMA-DCI evacuation inventories with the pre-2012 institutional catalogues, the evacuated-and-catalogued manuscripts will prove to be predominantly duplicate copies of already-known works, while post-crisis field cataloguing of previously unlisted family libraries keeps adding new distinct works at a rate implying the 2012 rescue captured well under half of Timbuktu's distinct textual population (primary clause: the rescued corpus is under half of the region's distinct works; the verdict follows it). Kill: the SAVAMA-DCI evacuation and digitization inventories (with the Hill Museum & Manuscript Library / T160K partnership) set against the pre-crisis IHERI-AB (Ahmed Baba Institute) catalogues; the reported burn and evacuation figures come from contemporaneous reporting and should be treated as estimates, not audited counts.

Prediction clause (verbatim)

Prediction: reconciling the SAVAMA-DCI evacuation inventories with the pre-2012 institutional catalogues, the evacuated-and-catalogued manuscripts will prove to be predominantly duplicate copies of already-known works, while post-crisis field cataloguing of previously unlisted family libraries keeps adding new distinct works at a rate implying the 2012 rescue captured well under half of Timbuktu's distinct textual population (primary clause: the rescued corpus is under half of the region's distinct works; the verdict follows it).

Kill-dataset (verbatim)

Kill: the SAVAMA-DCI evacuation and digitization inventories (with the Hill Museum & Manuscript Library / T160K partnership) set against the pre-crisis IHERI-AB (Ahmed Baba Institute) catalogues; the reported burn and evacuation figures come from contemporaneous reporting and should be treated as estimates, not audited counts.

Novelty / leakage triage

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

The 2012-13 SAVAMA-DCI evacuation and the IHERI-AB burning are documented (English; Hammer), and Hall & Stewart have shown the Timbuktu book market was dominated by a heavily duplicated core curriculum — the nearest prior art for the 'mostly duplicates' half. But the primary clause is a distinct-work saturation estimate: that the rescue captured well under half of the region's distinct textual population, with field cataloguing of unlisted family libraries still adding new works. No one has reconciled the evacuation inventories against the pre-crisis catalogues to a distinct-versus-duplicate count, and the reported totals are themselves journalistic estimates. The under-half claim is un-run.

Predictions

No prediction registered yet.