Bounding session delegation in remote assessment — measurements, code and provenance (CSI submission)

Companion deposit for a manuscript submitted to Computer Standards & Interfaces. Every quantity reported in the article traces to a file in results/, and every file traces to the command that produced it; that mapping is manifest.json, with a SHA-256 digest per entry. The study partitions the delegation of an examination session to a third party into attack modes and asks which a zero-knowledge authentication ledger can bound. Two of the four modes admit an upper bound on the transferred fraction, one through a detector complete for its mode and one by correcting the miss rate of an incomplete detector; two do not, for different reasons — one institutional, one because it lies outside the mechanism's claimed boundary. The main instrument is a parameter sweep over comparator separability and slot count: 163 cells, 62,200 executions. A single validation point on a real keystroke corpus anchors that surface. Quantities carry a SYNTH- or CORPUS- prefix and are never aggregated across the two. The deposit holds the twelve result files, the experiment, circuit, verifier and baseline code, the three figures with their script, the Groth16 verification key, and the claim register, which records the corrections applied during the work alongside the claims that survived. Two exclusions are deliberate and recorded in the manifest. The KeyRecs corpus is NOT redistributed here: only aggregate quantities derived from it are published, and the corpus is available under CC-BY-4.0 at DOI 10.5281/zenodo.7886742. The powers-of-tau and proving-key binaries are omitted as regenerable 15 MB artefacts; the setup script and the entropy it used are published instead, which is what allows a reader to confirm that the setup was a single contribution and not a ceremony. The verification key is included, without which the deposited proofs would not be checkable.