Config reference: synthetic

• Enterprise tuning surface

  ---

  Defaults + constraints are rendered directly from Pydantic.

• Env keys when available

  ---

  Many fields have an env-style alias (from TriBridConfig.to_flat_dict()).

• Tooltip-level guidance

  ---

  If a matching glossary entry exists, you’ll see deeper tuning notes.

Config reference Config API & workflow Glossary

Total parameters: 11

Group index

• generator
• judge
• quality_gate

generator

JSON key	Env key(s)	Type	Default	Constraints	Summary
synthetic.generator.concurrency	—	int	4	≥ 1, ≤ 16	Concurrent generator/judge requests sent to the LiteLLM gateway per synthetic run. Forced to 1 when the selected alias is the single-stream local vLLM serving row.
synthetic.generator.evidence_quote_max_chars	—	int	200	≥ 50, ≤ 1000	Max characters for evidence quote field
synthetic.generator.expected_answer_max_chars	—	int	400	≥ 50, ≤ 2000	Max characters for expected answer field
synthetic.generator.max_tokens	—	int	1200	≥ 100, ≤ 16000	Max tokens for generator LLM response
synthetic.generator.question_max_chars	—	int	180	≥ 50, ≤ 500	Max characters for generated question text
synthetic.generator.source_excerpt_max_lines	—	int	80	≥ 10, ≤ 500	Max lines of source chunk content sent as context to generator/judge
synthetic.generator.temperature	—	float	0.0	≥ 0.0, ≤ 2.0	Temperature for synthetic generator LLM calls

judge

JSON key	Env key(s)	Type	Default	Constraints	Summary
synthetic.judge.max_tokens	—	int	400	≥ 100, ≤ 4000	Max tokens for judge LLM response
synthetic.judge.temperature	—	float	0.0	≥ 0.0, ≤ 2.0	Temperature for synthetic judge LLM calls

quality_gate

JSON key	Env key(s)	Type	Default	Constraints	Summary
synthetic.quality_gate.sample_size	—	int	50	≥ 1, ≤ 10000	Number of eval items to sample for quality gate evaluation
synthetic.quality_gate.top1_min	—	float	0.4	≥ 0.0, ≤ 1.0	Minimum top-1 retrieval accuracy to pass quality gate