061-host-scored-repeat-sets
Host-scored repeat sets for the selection gate (amends 029 and 041)
A proposed amendment of approved specs 029 and 041 (and of spec 006's command table through 029 3.13), for release 0.4.1. Today `rustev gate select` reads only repeat sets Rustev evaluated itself: every run is a spec 004 evaluation of a Rustev plan over a Rustev dataset manifest. A host whose baseline and candidate runs execute in its own runtime, and whose outputs are scored through `scoreViews` into `rustev.scores/1`, has no supported way in. This spec adds a host run record (`rustev.host-run/1`) carrying host dataset, flow, version and run identities with per-case outcomes, latency and cost; an import command, `rustev eval repeat import`, that verifies the host's views and scores and builds a host repeat set with spec 029's statistics; a host baseline; and selection over host sets. Host documents have their own schema names, so no existing document changes bytes or meaning and a host set can never be compared with a Rustev-evaluated one. Every figure is labeled host-asserted. Ratified (A-77) for release 0.4.1.
Read the full pinned specification ↗Relationships
depends on
depended on by
Claimed units
- extends
{ "nature": "amending", "spec": "004-evaluation-and-replay", "unit": { "kind": "directory", "path": "crates/rustev-eval/" } } - extends
{ "nature": "amending", "spec": "006-cli-surface", "unit": { "kind": "directory", "path": "crates/rustev-cli/" } } - extends
{ "nature": "additive", "spec": "031-public-api-and-release-family", "unit": { "kind": "directory", "path": "tools/rustev-release/" } } - extends
{ "nature": "additive", "spec": "001-boundaries-and-authority", "unit": { "kind": "file", "path": "Makefile" } } - extends
{ "nature": "additive", "spec": "001-boundaries-and-authority", "unit": { "kind": "directory", "path": "scripts/acceptance/" } }
Source headings
- 061: Host-scored repeat sets for the selection gate (amends 029 and 041)
- 1. Purpose
- 2. Territory
- 2.1 Binding principles
- 3. Behavior
- 3.1 The host run record
- 3.2 Import: `rustev eval repeat import`
- 3.3 Measures
- 3.4 Statistics, sources and bands
- 3.5 Derivation, lineage and labels
- 3.6 Provenance separation
- 3.7 Host baselines and selection
- 3.8 Held-out splits
- 3.9 Exit codes
- 3.10 Library and bindings
- 4. Out of scope
- 5. Observable negative cases
- Acceptance
- Verification
- Implementation record
- Decisions at approval
Frozen anchors
No frozen anchors declared.