* feat(evals): add Hebrew fluency holdout cases * feat(evals): add Koine Greek fluency holdout cases * feat(packs): attach fluency lanes to he_core_cognition_v1 * feat(packs): attach fluency lanes to he_logos_micro_v1 * feat(packs): attach fluency lanes to grc_logos_cognition_v1 * feat(packs): ADR-0103 fluency lane attachment * test(packs): expect ADR-0103 fluency lanes on Hebrew Greek contracts * docs(evals): add Hebrew fluency holdout split note * docs(evals): add Koine Greek fluency holdout split note * docs(evals): note Hebrew holdout attachment * docs(evals): note Koine Greek holdout attachment * docs: add ADR 0103 placeholder * docs(adr): expand ADR-0103 fluency lane attachment * docs: index ADR-0103 and refresh frontier
2.1 KiB
ADR-0103 — Fluency Lane Attachment for ADR-0102
Status: Accepted Date: 2026-05-22 Author: CORE agents + reviewers Depends on: ADR-0091, ADR-0102
Context
ADR-0102 ratified hebrew_greek_textual_reasoning with only universal reasoning lanes attached:
inference_closurefabrication_control
At the time, the language-specific Hebrew and Koine Greek fluency lanes did not yet expose sealed holdout splits, so attaching them to ratified contracts would have violated ADR-0091 expectations around dev/public/holdout coverage.
Both lanes now ship plaintext holdout sets through:
holdouts/v1/cases.jsonl
and can therefore be attached to the ADR-0102 manifests without weakening the contract discipline.
Decision
Attach the following eval lanes to all four ADR-0102 manifests:
hebrew_fluencykoine_greek_fluency
Each lane declares:
{
"version": "v1",
"splits": ["dev", "public", "holdout"]
}
The following packs must remain uniform:
he_core_cognition_v1he_logos_micro_v1grc_logos_cognition_v1grc_logos_micro_v1
All four manifests advance provenance to:
adr-0103:reviewed:2026-05-22
Acceptance evidence
Accepted after the following landed together:
- Hebrew holdout cases
- Koine Greek holdout cases
- updated fluency contracts
- updated gap notes
- updated sibling-ratification invariants
- uniform lane attachment across all four manifests
Invariants
hebrew_greek_fluency_lanes_uniform
All four ADR-0102 packs must declare identical eval-lane coverage.
hebrew_greek_fluency_holdouts_present
Both:
hebrew_fluencykoine_greek_fluency
must ship non-empty holdouts/v1/cases.jsonl before appearing in ratified domain contracts.
Consequences
- ADR-0102 now carries language-specific fluency evidence in addition to universal reasoning evidence.
- Hebrew/Greek textual reasoning contracts now satisfy full
dev/public/holdoutdiscipline for attached fluency lanes. - Future language-track ADRs can extend morphology and construction coverage without changing Domain Pack Contract v1.