* 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
1.6 KiB
Koine Greek fluency eval lane (Phase 5.3)
What it measures (v1, honest scope)
Whether the deterministic articulation layer
(generate/articulation.py) produces a Greek-script surface in
subject-object-predicate word order when given a (subject,
predicate, object) triple drawn from the
grc_logos_cognition_v1 / grc_logos_micro_v1 seed packs.
This is the C01 (simple declarative) gate only.
Why v1 is C01-only
Identical reasoning to the Phase 5.2 Hebrew lane: the realizer's
tense/aspect/quantifier/negation logic in generate/templates.py
and generate/morphology.py is English-only. Greek requires its
own morphology + rhetorical templates before C02–C13 can be
measured honestly. See gaps.md for the v2 unblock path.
Inputs
public/v1/cases.jsonl- 3 cases — one per triple from the seed packs, all C01.
dev/cases.jsonl- 1 case — smoke check for runner integration.
holdouts/v1/cases.jsonl- 5 sealed plaintext Koine Greek realization cases added for ADR-0103.
These cases are not referenced by the development split and complete
the
dev/public/holdoutrequirement for attachingkoine_greek_fluencyto ADR-0102 ratified contracts.
Scoring rubric
A case passes if the realized surface:
- Contains Greek script (U+0370..U+03FF or U+1F00..U+1FFF for polytonic)
- Is in
accept_surfacesOR satisfies allconstraints(must_contain,max_words)
Provenance
Same realize() entry point as
tests/test_articulation.py::test_realize_greek_surface_uses_greek_script_and_compact,
generalized into a lane with multiple cases and the standard scoring
shape.