When the Interval Is Smaller Than the Instrument: measurement dataset (campaign ledgers, per-cell counters, harness results, clock-discipline log)

The measurement dataset behind the accompanying software record: every aggregated result the paper's claims are recomputed from. Contents: the campaign ledger (one row per instrumented-benchmark cell: configuration, sign-separated discard counters, log size and SHA-256 of the source log); per-cell counter extracts; the Python-harness results (per-cell counts with sign-separated discards and per-run pacer-jitter percentiles, same-host and cross-host); the grid-membership inference output; the minute-by-minute clock-discipline log; the distributed-mode failure diagnostics; and the per-experiment CSVs for the inversion-mode campaigns (audit, priority, geometry, payload, trace). Headline totals: 223 instrumented OpenMessaging Benchmark runs; Kafka-driver corpus 10,913,263 discarded samples with 0 negatives; Redis-driver corpus 619,229 with 41,403 negatives, every one exactly -1000 microseconds. The schema is documented in the software record's supplement (S3). New in v2. Three artefact families were added for the IEEE TPDS revision, each generated by a committed script and pinned by tests. (1) traced_tail_slope.csv: survival, local slopes and windowed log-log indices recovered from the traced bpftrace run-queue histogram, giving a second and independent estimator of the stall tail (0.332 over 0.25-2 ms) against the 0.339 fitted from the payload sweep. (2) Audit-gated powered-transport artefacts (*_gated.csv in transport_rt/ and transport_rt2/) plus gate_sensitivity.csv: the earlier powered comparison had consumed audit-condemned runs, so the primary tables are now computed on audit-surviving runs only, with the sensitivity of the conclusion to that choice reported rather than asserted. (3) first_result_threshold_sweep.csv: a condition-level sweep showing that no cell of the withdrawn first result set becomes fully usable at any audit threshold up to 20%. A SHA256SUMS manifest lists every file in the archive so any single file can be verified independently of the zip. Data note. No StatsBomb event data is included. Aggregate workload statistics derived from it remain subject to StatsBomb's CC BY-NC 4.0 terms with attribution. Licence. This data compilation and its documentation are released under CC BY 4.0. The analysis code that produced and consumes it is MIT and lives in the linked software record; the manuscript files archived there are the author's copyright and are expressly not CC BY, pending journal publication.