{
 "schema": "s2s-bench-v1",
 "kind": "data-kit-index",
 "exhibit": "cove-voice-head-to-head",
 "exhibit_title": "The narrator’s chair, refused",
 "exhibit_url": "https://research.strata2signal.com/cove-voice-head-to-head/index.html",
 "published_utc": "2026-08-13",
 "status": "published 2026-08-13",
 "licence": "CC BY 4.0",
 "attribution": "strata→signal research, research.strata2signal.com",
 "contact": "hello@strata2signal.com",
 "outcome": "REJECTED — the candidate misses floor 1 (canon) and floor 4 (latency). A rejected candidate gets no threshold table: no number in this file is offered as an adoption target.",
 "contamination_caveat": "Published 2026-08-13. Models with a later training cutoff may have seen this set. We author fresh sets each cycle; this one is not a standard, it is our kit, yours to reuse.",
 "not_a_standard": "Our own trial, our own hardware, for our own seat. Not first independent numbers, and not a benchmark.",
 "own_scale": "These 0-10 scores live only inside this head-to-head. They share no ruler with the frozen voice trial published on the voice-trials exhibit: different prompts, different rubric, different posture, different judges. No delta between the two is computed anywhere — not in this file, not on the page.",
 "files": [
  {
   "file": "rows.json",
   "role": "one row per judged sample x arm — 13 prompts x 3 models = 39 rows, 117 scored cells: the reply as generated (the spoken line and the full envelope), per-sample timings with the contention flag and its re-run, all three blind judges' scores, canon verdicts and notes, the un-blinded canon findings and clause-fidelity flags, the 13 prompts with their canon anchors and provenance, the per-arm summary, the sensitivity cuts, and the counting rules",
   "media_type": "application/json",
   "bytes": 181616,
   "sha256": "1b818f416dc702e42ebfea2920142dc3590a0fdea7b438926b9e16407992a539"
  },
  {
   "file": "gate.json",
   "role": "the pre-registered gate: its five floors with their registered text verbatim, the verdict each returned and the evidence behind it, the interpretive call on the canon floor, the chair ruling on the register floor, what a pass would have earned, and what a re-bench would need. Carries the REJECTED verdict and the no-threshold-table rule",
   "media_type": "application/json",
   "bytes": 5837,
   "sha256": "735d0c64cad4fba11246798101f8fde4d706eb43a7d757c0168f028736ed2557"
  },
  {
   "file": "README.md",
   "role": "the kit's own README: licence, receipts, the contamination caveat, and how to reuse these rows",
   "media_type": "text/markdown",
   "bytes": 3964,
   "sha256": "552a22f1de797341c5f9f4e5d35e3a8a9a0c2d86795480c1dbdf000a4bfff696"
  }
 ],
 "self": {
  "file": "index.json",
  "note": "this listing; its own sha is not carried inside itself"
 }
}