Record size corpusstable
Summary statistics — realistic corpus (N=90)
| Quantity | N | Min | Median | P95 | Max | Unit | §7 budget | Max/budget |
|---|---|---|---|---|---|---|---|---|
| (a) doc bytes, as-committed — informational | 90 | 2,144 | 2,950 | 6,497 | 20,211 | bytes | — | — |
| (a) doc bytes, compact JSON — MUST NOT exceed | 90 | 1,525 | 2,138 | 4,842 | 16,323 | bytes | 10,485,760 | 0.1557% |
| (b) raw Σ decoded — MUST NOT exceed | 90 | 0 | 0 | 602 | 985 | bytes | 6,291,456 | 0.0157% |
| (c) max single vector decoded — MUST NOT exceed (N=35 records carry >=1 inline vector) | 35 | 1 | 32 | 3,072 | 3,072 | bytes | 6,291,456 | 0.0488% |
| (d) spaces[] entries — SHOULD NOT exceed | 90 | 0 | 1 | 2 | 3 | entries | 64 | 4.6875% |
| (d) embeddings[] entries — SHOULD NOT exceed | 90 | 0 | 1 | 2 | 3 | entries | 64 | 4.6875% |
| (e) provenance.chain[] entries — SHOULD NOT exceed | 90 | 0 | 0 | 4 | 8 | entries | 1,024 | 0.7812% |
Corpus composition, N=90:
- carries ≥1 inline
metadata.raw[]segment: 29 - carries ≥1 by-reference
metadata.raw[]segment: 3 - carries ≥1 inline embedding (
vectororvector_b64): 35 - carries ≥1 by-reference embedding (
vector_ref): 56