Loading…
The rule the corpus is built under, stated so it can be checked against the corpus itself. Some of it is enforced by code rather than left to an author's care.
The policy holds all three and refuses to trade any of them away.
Lossless
Nothing that carries meaning is dropped. Tone marks and subdots are preserved, the translator is named, and the original stands beside the rendering rather than being replaced by it.
Most coverage
The corpus reaches for the widest documented record it can, including material a cautious project would leave out because sourcing it is work.
Least copyright exposure
Facts are taken freely, expression is not, and anything whose rights position is unclear is marked rather than quietly shipped.
Those three pull against each other. Coverage wants to include everything; copyright caution wants to include less; lossless fidelity wants to reproduce the original exactly, which is precisely what copyright restricts. The rest of this page is how that tension gets resolved in each case.
Taken freely: facts, ideas, dates, structures, findings. Copyright does not protect facts or ideas, only their expression. A date, a lineage, an archaeological finding, an etymology, a structural feature of an Odù, the argument a scholar makes: all of it can be restated. The corpus restates in its own prose and cites the source it learned it from.
Never taken: another author's expression, selection or arrangement. A compilation is protected as a compilation even when every item in it is public domain, so copying a collection's selection is an infringement even where copying any single entry would not be.
Never taken: a third-party translation, treated as free stock. A published translation is its translator's own expression and is fully protected independently of whatever it translates. Translations here are either our own, or public domain, or openly licensed, or quoted at a length fair dealing plainly supports, and are attributed in every case.
A proverb, an oríkì or an ẹsẹ Ifá cannot be paraphrased without destroying it, so primary texts are the one place the corpus reproduces rather than restates. For each one it uses the highest rung it can actually reach.
1. Public domain by age
Pre-1930 collections and vocabularies: Crowther (1843, 1852), Bowen (1858), Ellis (1894), and the early twentieth-century Yorùbá-authored collections. Copyright has expired and reproduction carries no exposure at all.
2. Openly licensed or open-access repository publication
A text a publisher or a university repository has itself put out openly. This is a licence, not an inference, so it is checked against the repository's own statement rather than assumed from the fact that a page loads.
3. Field-collected or contributed
With the contributor's consent recorded, and with attribution, region and dialect recorded alongside the text.
4. Quoted at fair-dealing length
A single proverb or a short verse quoted from an in-copyright work for comment and criticism, with full citation. Never a run of entries, and never the source's selection.
5. Not published
If none of the four above can be reached, the text does not go in and the gap is marked as a gap. This rung is the one the policy exists to make available: not shipping is always an option and is always cheaper than the alternative.
The ladder is what makes most coverage and least exposure compatible. A text can nearly always be reached at some rung; the work is finding which one, and the finding is done before publication rather than after a complaint.
Sourcing is also a question of what a source can establish, and three kinds of source establish three different kinds of thing.
What the tradition says of itself
Ifá verses, oríkì, origin accounts, what a lineage holds about its own founding. Real evidence of what the tradition holds. Not evidence of dates, migrations or events.
What the documentary and material record establishes
Archaeology, historical linguistics, dated documents, radiocarbon results, art-historical analysis. This can establish that a thing happened, within stated error bars.
What colonial, missionary and early ethnographic sources claimed
Often the only surviving written record for a period, and simultaneously produced by people with an explicit interest in what they found. Usable, never at face value, and always labelled as an artefact of its moment.
Blending them is how both the hostile account and the celebratory account go wrong: the hostile one takes the third register as fact, the celebratory one takes the first as history. Every claim in the corpus says which register it draws on.
Three things in this policy are not left to an author's care. They are checked by code.
0 across 694 documents
1,866,050
Every substantive claim carries an inline marker
A bracketed marker resolving to a named source in the file's own Sources block, which the corpus writing contract makes mandatory for every file.
A citation that does not exist gets rescued or the claim gets cut
A verification pass checks every citation against library catalogues and databases. Anything it cannot find goes to a grounded search that must locate a real published source supporting the same passage. The verdict is either a replacement source, or unsupported, in which case the passage is cut or softened. A source found in search this turn, or nothing.
A bare-domain URL is dropped
A citation URL is kept only when it has a path. A bare domain identifies nothing, and a citation that resolves to a homepage is not a citation.
The proverb corpus is the one place where the exposure concentrates, so its position is published rather than left implicit.
1,278 proverbs across 32 files come from a publicly hosted open etext collection, attributed, and are defensible as they stand.
410 proverbs across 26 files are cited by entry number to a book that is still in print. This set is being re-sourced before any commercial release, under the rules below.
Citing an entry number is not itself a problem, and quoting an individual proverb for comment is ordinary fair dealing. The exposure is at scale and in selection: several hundred entries taken in the order the book puts them in reproduces that book's arrangement, which is protected independently of the proverbs themselves.
Re-source each one to an independent attestation
Wherever an open attestation exists, so the proverb is carried by a source the corpus can point at rather than by the in-print book's arrangement. A grounded gather is finding those attestations, and its counts are checked file by file before any figure derived from them is published.
Prefer the open edition where the same publisher has made one
Where a publisher has itself put the material into an open repository, cite the open edition rather than the print one. That converts an exposure into a licence, and it is the outcome to prefer wherever it is available.
Keep or cut, where nothing independent is found
The entry is either kept as a single quoted proverb with full citation and comment, which fair dealing supports, or cut. It is not kept as part of an unbroken run of the source's own ordering.
Break the arrangement regardless
The corpus's own files are organised by subject, not by any source's entry sequence, so the ordering is the corpus's own selection rather than a reproduction of anyone's compilation.
No count of how many of those entries have been re-sourced is published until the work is complete and the results have been checked. Publishing a number that later moves would be worse than publishing none.
Attribution is not decoration on this corpus. The citations are the argument: the reason the project exists is that the widely circulated account of Yorùbá tradition was not sourced, and a reader deciding whether to trust this one is deciding on the apparatus. Reuse that strips the citations reproduces exactly the problem the project was built to fix.
Sources are named generously, linked wherever a link exists, and the corpus actively sends readers to the original authors and their books rather than positioning itself as a replacement for them.