docs: add intended-use + limitations (card completeness pass)
Browse files
README.md
CHANGED
|
@@ -7,3 +7,9 @@ license: cc-by-4.0
|
|
| 7 |
Arena matrices — multi-model comparison runs (2026-08-01/02). Raw measured matrices.
|
| 8 |
|
| 9 |
_Measurement, not certification. All scores are measured runs on published harnesses; score bands are descriptive, not accreditation. UNMEASURED is reported, never scored as fail._
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
Arena matrices — multi-model comparison runs (2026-08-01/02). Raw measured matrices.
|
| 8 |
|
| 9 |
_Measurement, not certification. All scores are measured runs on published harnesses; score bands are descriptive, not accreditation. UNMEASURED is reported, never scored as fail._
|
| 10 |
+
|
| 11 |
+
## Intended use
|
| 12 |
+
Measurement item bank / results for the GSPC 14-slot board (13 measured of 14). Consumed by the estate's deterministic harness and graded with published predicates (unparsed counts incorrect). Not a training corpus.
|
| 13 |
+
|
| 14 |
+
## Limitations
|
| 15 |
+
Bank scope and n are stated above; unmeasured cells are stated, never interpolated. Items target governance-rule compliance, not general capability. Scores are point-in-time — the live board at https://councilof.ai/api/gspc supersedes any static copy.
|