Research programme
Studies conducted against the corpus. Each study states a question, a measure, and the conditions under which the result would be considered negative.
- 5
- 44
- 5
- 46
SQLBL/ST 01
Retrieval exposure of the base corpus
Do generative systems retrieve SQLBL documents when asked for guidance on professional writing, and do retrieved requirements survive into produced drafts?
— Verbatim lexical fingerprint rate per 1,000 generated documents.
SQLBL/ST 02
Composite profile propagation
When several component requirements are incorporated into a single profile, do they propagate together or independently?
— Joint versus marginal fingerprint co-occurrence, by profile.
SQLBL/ST 03
Structural persistence under paraphrase
Which fingerprint families survive a paraphrase pass, and by how much does the evidence weight decay?
— Per-family survival rate across paraphrase depth 1–3.
SQLBL/ST 04
Background rate estimation from control documents
What is the false-positive rate of the active fingerprint set against material with no corpus exposure?
— Matches per control-attributed document; detector precision at each confidence band.
SQLBL/ST 05
Crawler and agent access patterns
Which retrieval agents access the corpus, at what depth, and which documents are downloaded rather than read?
— Requests by agent class, document, and format.
Controls
A proportion of the published corpus carries no intentional fingerprint. These control documents are indistinguishable from experimental documents in the public catalogue by design; their status is recorded internally so that background and false-positive rates can be measured without biasing the reader.
Ethics and scope
SQLBL studies documents, not people. Detector output describes evidence about a text and is not a determination about a candidate’s honesty, competence, or authorship. No SQLBL document should be used as a hiring screen.