Record size corpusstable
Designed size/entry-count outliers (excluded from the realistic corpus above)
| File | Kind | Doc bytes (committed) | Doc bytes (compact) | Raw Σ decoded (b) | Raw segs | Max vector decoded (c) | Carriage | Spaces | Embeddings | Chain | Target code | Why designed (conformance/manifest.json) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| conformance/invalid/ver103-record-exceeds-hard-cap.json | invalid fixture | 4,127 | 3,172 | 448 | 3 | — | — | 1 | 1 | 0 | VER103 | A conformant L1 record run against a deliberately small hard cap. The cap is a deployment policy (default 64 MiB), so the fixture exercises it with --max-bytes 1024 rather than shipping a 64 MiB file. |
| conformance/invalid/ver104-record-size-policy.json | invalid fixture | 2,099,609 | 2,098,937 | 0 | 0 | 8,388,616 | bare vector[] (proxy) | 1 | 1 | 0 | VER104 | An inline vector of 1,048,577 components: 8,388,616 bytes at the documented 8-bytes-per-component proxy, past the 6 MiB single-vector advisory. VER 1.0 sets no normative size limits, so this is a warning; VER 1.1 drafts real limits. |
| conformance/invalid/ver105-record-size-exceeded.json | invalid fixture | 1,574,913 | 1,574,345 | 0 | 0 | 6,291,464 | bare vector[] (proxy) | 1 | 1 | 0 | VER105 | An inline vector of 786,433 components: 6,291,464 bytes at the documented 8-bytes-per-component proxy, one component past the 6 MiB single-vector limit of draft section 7 quantity (c), which is a MUST NOT. |
| conformance/invalid/ver106-size-should-cap-exceeded.json | invalid fixture | 22,768 | 15,181 | 0 | 0 | 32 | bare vector[] (proxy) | 1 | 65 | 0 | VER106 | 65 embeddings, one past the 64-entry cap of draft section 7 quantity (d). The quantity measured is entries, not bytes: a record this wide is a bundle of records wearing one record's identity. |