cli: surface answer_mode=claim_lattice on ask/query (JSON variant)

Round-tripped end-to-end against Hermes-3-8B:
  query "what dinosaurs were in the first jurassic park film?"
  --answer-mode claim_lattice
  -> HYBRID 5/5 verified via claim_lattice in 4.9s

Hermes packed all five species into a single claim with five
evidence_ids, tripping TOO_MANY_EVIDENCE_IDS (max 2/claim). The
substrate's hard rule kicked in correctly — per-id checks all passed
but claim under-decomposition demoted STRICT to HYBRID. Reasoner-class
models (Qwen 3.6, Claude, GPT-4) should decompose into one claim per
species and clear STRICT.
This commit is contained in:
russell@unturf.com 2026-04-30 12:37:49 -04:00
parent 873cfa8b31
commit bc9438f900
No known key found for this signature in database

View file

@ -2852,10 +2852,12 @@ def build_parser() -> argparse.ArgumentParser:
)
ask_cmd.add_argument(
"--answer-mode", dest="answer_mode", default=None,
choices=["quote", "claim_lattice_pointer"],
choices=["quote", "claim_lattice_pointer", "claim_lattice"],
help=(
"answer schema. See `query --answer-mode` for full semantics. "
"Default 'quote'; 'claim_lattice_pointer' enables quote-by-pointer."
"Default 'quote'; 'claim_lattice_pointer' enables quote-by-pointer; "
"'claim_lattice' is the JSON variant (vLLM guided_json + lenient "
"pre-parser; pairs with grammar-constrained inference)."
),
)
ask_cmd.set_defaults(func=_cmd_ask)
@ -2949,7 +2951,7 @@ def build_parser() -> argparse.ArgumentParser:
)
query_cmd.add_argument(
"--answer-mode", dest="answer_mode", default=None,
choices=["quote", "claim_lattice_pointer"],
choices=["quote", "claim_lattice_pointer", "claim_lattice"],
help=(
"answer schema. 'quote' (default): model writes prose with "
"verbatim quote spans inline. 'claim_lattice_pointer' (G0 "
@ -2957,7 +2959,9 @@ def build_parser() -> argparse.ArgumentParser:
"map (E1, E2, …); model writes pointer-line prose ('Claim. "
"[E12]'); renderer interpolates literal spans. Synthetic-"
"elision-by-construction-impossible. No repair loop "
"(one-shot discipline)."
"(one-shot discipline). 'claim_lattice' is the JSON variant "
"(vLLM guided_json + lenient pre-parser; pairs with grammar-"
"constrained inference like Qwen 3.6 reasoner / Claude / GPT-4)."
),
)
query_cmd.set_defaults(func=_cmd_query)