Methodology
Availability grading — definition v2.0
Applies to every grade and KPI figure quoted by BookKorpus. Figures always cite the definition version they were computed under. Changes are versioned and logged in §8.
Each graded title is assigned one of four availability grades, and each lot carries two headline KPIs derived from them. “Digitised” can mean two very different things — a scan that anyone can read, or a scan locked away where only Google can use it — so we report both dimensions separately rather than collapsing them into one number.
The grades and KPIs describe the digitisation and accessibility status of a title at the check date. Nothing more — see §6.
Every graded title is checked against the German National Library (DNB), HathiTrust, Internet Archive and Open Library. For Grade A, antiquarian scarcity is additionally checked against trade availability (ZVAB).
| Grade | Meaning |
|---|---|
| A | No scan traceable anywhere in the checked sources, and scarce in the antiquarian trade. |
| B | No scan traceable anywhere in the checked sources. |
| C | A scan exists but is closed — HathiTrust “search-only” or Google snippet view. Only Google has access to the text. |
| D | An openly accessible full-text scan exists. |
| unknown | Match too fuzzy for a reliable grade (see §5). Remains in the denominator; never counts toward any KPI. |
KPI 1 — never digitised (A+B). The share of the graded sample for which no scan is traceable anywhere in the checked sources.
KPI 2 — not openly available (A+B+C). The share that is digitally inaccessible to every buyer except Google. For a training-data buyer, this is the number that counts: a Grade C title is, in practice, undigitised for everyone but Google — the physical volume is the only route to the text.
Both KPIs are computed over the full graded sample, including unknown-graded titles in the denominator. Fuzzy matches therefore lower our numbers; they never inflate them.
“Sample of 80 titles, graded 2026-08-24: 62% never digitised · 91% not openly available (A9 B41 C23 D5 U2). Full title-level grading sheet on request.”
Because all check sources are public catalogues, any row of a grading sheet can be re-checked by the buyer against the sources and date named on the sheet.
Check sources: DNB catalogue, HathiTrust, Internet Archive, Open Library; ZVAB for the scarcity component of Grade A. The source set and the date on which checks were run are stated on every grading sheet.
These databases change. A grade holds as of its check date; re-grading against a later date can be agreed as part of a transaction.
Lot KPIs are computed from a sample of titles identified from shelf and spine photographs. The sample size is stated with every figure. Figures are sampled, not guaranteed — we do not extrapolate a sample into a promise about every title on the pallet.
Bound periodical volumes, grey literature and pre-ISBN titles match fuzzily against catalogues. Where a match is not reliable, the title is graded conservatively as unknown — never as A.
The grades make no statement about copyright, usage or licence rights. They describe whether a scan of a title exists and who can access it — nothing about what any party may do with the text. We do not use, and buyers should not read into the grades, phrasings like “free to use” or “rights-cleared”.
We check against public catalogues. DNB, HathiTrust, Internet Archive and Open Library confirm nothing and endorse nothing; no grade is “verified by” any institution.
“If Google scanned it, it’s digitised, isn’t it?”
Yes — in a vault. HathiTrust “search-only” works are searchable but not readable, not exportable, and not available to third parties. For every AI lab except Google, the text is as inaccessible as if it had never been scanned. That is exactly why we report both numbers.
This definition is versioned; material changes increment the version and are logged here. Quoted figures always name the version they were computed under.
- v2.0 — 2026-08-24. Two-tier measurement: availability grades A–D and two lot KPIs (never digitised; not openly available), replacing the single duplication KPI.
- v1.0 — 2026-08-12. Initial publication: single duplication KPI against public digital corpora. Superseded.
Queries on the definition, or on applying it to a specific lot: [email protected] · +44 7398 705879. We reply within one business day.