infer.query.v1 Schema

Canonical field reference for the infer.query.v1 body shape — mirrors the JSON Schema shipped in @syncropel/config@0.3.x.

Overview

This page is the canonical field-level reference for the infer.query.v1 body shape. It mirrors the JSON Schema published in @syncropel/config@0.3.0 at src/schemas/infer-query.schema.json.

Schema id: https://syncropel.com/schemas/config/infer-query.v1.schema.json
JSON Schema draft: 2020-12
TypeScript types: import { InferQueryV1 } from "@syncropel/config"
Validator: import { validateInferQuery } from "@syncropel/config"

For the conceptual walkthrough with worked examples, see the Query Anatomy guide.

Top level

{
  "kind": "infer.query.v1",
  "input": { "inline": "..." },
  "answer_shape": { "kind": "core.text.v1" },
  "responders": [{ "kind": "llm", "model": "~sonnet" }],
  "fold": { "function": "consensus" },
  "dial": 0.5,
  "orchestration": { "pattern": "single_shot" },
  "side_effects": { "reversible": true, "max_cost_usd": 0.40 },
  "relevance": { "model": "current", "threshold": 0.5, "top_k": 3 },
  "metadata": { "trace_id": "trace_abc123" }
}

Field	Type	Required	Default
`kind`	const `"infer.query.v1"`	yes	—
`input`	Input — oneOf three shapes	yes	—
`answer_shape`	AnswerShape	yes	—
`responders`	ResponderPredicate[] (minItems 1)	yes	—
`fold`	Fold	yes	—
`dial`	number OR AdaptiveDial	no	`0.5`
`orchestration`	Orchestration	no	`{ pattern: "single_shot" }`
`side_effects`	SideEffects	no	namespace defaults
`relevance`	Relevance	no	`{ model: "current", threshold: 0.5, top_k: 3 }`
`metadata`	map of string to string	no	`{}`

Input

Exactly one of:

{ "inline": <any JSON>, "inline_kind": "core.text.v1" }

{ "record_id": "3fa8...e2c1" }

{ "ref": { "kind": "@music.track", "id": "spotify:track:4iV5W9uYEdYUVa79Axb7Rh" } }

Field	Type	Notes
`inline`	any	Payload hashed into the query record id
`inline_kind`	string	Optional body.kind
`record_id`	string	64-char hex SHA-256 record id
`ref`	object	Canonical ref per the 11 Ref.* constructors

AnswerShape

Validates the folded output (not per-response bodies).

{
  "kind": "core.translation.v1",
  "required_fields": ["body.text", "body.target_language"],
  "schema_ref": "https://example.com/translation.schema.json"
}

Field	Type	Required
`kind`	string (body.kind)	yes
`required_fields`	string[] (dot-paths)	no
`schema_ref`	string (URI)	no

ResponderPredicate

An OR-of-classes predicate. Each entry adds AND constraints within its class.

{
  "kind": "llm",
  "model": "~sonnet",
  "did": "did:sync:agent:translate-bot",
  "domain": "translation",
  "capability": "en-es",
  "trust_gte": 0.6,
  "trust_lt": 0.99,
  "budget_usd": 0.05,
  "latency_secs": 30,
  "match_level_gte": "L2",
  "age_days_lt": 7,
  "available": true,
  "expression": "candidate.last_seen_at > now() - 86400"
}

Field	Type	Notes
`kind`	enum	`actor` \| `llm` \| `pattern` \| `system` \| `any`
`model`	string	LLM model selector (e.g., `~sonnet`, `sonnet-4-6`)
`did`	string (DID)	Specific responder DID
`domain`	string	Capability domain
`capability`	string	Capability tag
`trust_gte`	number	Minimum trust score
`trust_lt`	number	Maximum trust score
`budget_usd`	number	Per-call cost ceiling
`latency_secs`	integer	Per-call latency ceiling
`match_level_gte`	enum	`L0` \| `L1` \| `L2` \| `L3`
`age_days_lt`	integer	Record age bound
`available`	boolean	Availability requirement
`expression`	string (CEL)	Arbitrary predicate over candidate metadata

Fold

{
  "function": "consensus",
  "weight_expression": "response.trust * response.recency",
  "tie_break": "highest_trust",
  "min_quorum": 2,
  "expression": "..."
}

Field	Type	Default	Notes
`function`	enum	— (required)	`consensus` \| `best_of` \| `waterfall_first` \| `ensemble_weighted` \| `expression`
`weight_expression`	CEL string	`"trust * recency * pattern_confidence"`	Applies to `consensus` and `ensemble_weighted`
`tie_break`	enum	`"highest_trust"`	`highest_trust` \| `highest_confidence` \| `most_recent` \| `lexicographic`
`min_quorum`	integer ≥ 1	`1`	Minimum non-error responses before fold runs
`expression`	CEL string	—	Required when `function == "expression"`

Orchestration

{
  "pattern": "verify",
  "primary": [{ "kind": "llm", "model": "~sonnet" }],
  "verifier": { "kind": "llm", "model": "~gpt-4" },
  "tiebreaker": { "kind": "actor", "did": "did:example:alice" },
  "agreement_threshold": 0.85
}

Field	Type	Notes
`pattern`	enum	`single_shot` \| `verify` \| `waterfall` \| `retry_on_low_confidence` \| `escalate` \| `ensemble_with_audit`

Per-pattern fields:

verify

Field	Type
`primary`	ResponderPredicate[]
`verifier`	ResponderPredicate
`tiebreaker`	ResponderPredicate
`agreement_threshold`	number or CEL

waterfall

Field	Type
`stages[]`	`{ responders: ResponderPredicate[] }`
`accept_expression`	CEL string

retry_on_low_confidence

Field	Type
`threshold_expression`	CEL string
`max_attempts`	integer
`retry_dial`	number 0..1

escalate

Field	Type
`tiers[]`	`{ responders: ResponderPredicate[] }`
`escalation_expression`	CEL string

ensemble_with_audit

Field	Type
`primary_responders`	ResponderPredicate[]
`auditor`	ResponderPredicate
`audit_expression`	CEL string

SideEffects

{
  "reversible": true,
  "idempotent": true,
  "max_cost_usd": 0.40,
  "max_latency_secs": 1800,
  "obligation_resolution": "last_writer_wins"
}

Field	Type	Default
`reversible`	boolean	`true`
`idempotent`	boolean	`true`
`max_cost_usd`	number	namespace cap
`max_latency_secs`	integer	`300`
`obligation_resolution`	enum	`last_writer_wins` — also `fulfills_wins`, `validation_error`

Relevance

{
  "model": "current",
  "threshold": 0.5,
  "top_k": 3
}

Field	Type	Default
`model`	string	`"current"` — also `"heuristic"` to force the additive-with-floor scorer
`threshold`	number 0..1	`0.5`
`top_k`	integer	`3`

Dial

Scalar form:

{ "dial": 0.3 }

AdaptiveDial form:

{
  "dial": {
    "replay_if": "pattern_match >= 'L1' && trust >= 0.92",
    "create_if": "pattern_match == null",
    "else": 0.4
  }
}

Field	Type
`replay_if`	CEL string — if matches, dial = 0.1
`create_if`	CEL string — if matches, dial = 0.9
`else`	number 0..1

Metadata

{ "metadata": { "trace_id": "trace_abc123", "customer_id": "cust_42" } }

Opaque string-to-string map. Carried through into the KNOW record unchanged.

Shipped in @syncropel/config@0.3.0 alongside infer-query.schema.json:

infer.call.v1 — CALL record emitted per dispatched responder (executor step 4).
infer.error.v1 — KNOW with an error code instead of an answer. 10 canonical codes: invalid_query, no_candidates, no_relevant_candidates, cost_budget_exceeded, latency_timeout, responder_unavailable, quorum_not_met, fold_failed, answer_shape_mismatch, obligation_conflict.