Release line

Current: v0.1.4. v0.2.0 is reserved for the coordinated release milestone across runtime-contracts and this package, and nothing before it may claim a 0.2.x or higher number.

Superseded tags

v0.3.0 and v0.4.0 were published in error. This package had never released a tag — its pyproject.toml said 0.2.0 and nothing was ever pushed — so those two took numbers off a line that had not been opened.

They point at the same commits as the correct tags and are left in place rather than deleted, because a published tag is immutable and deleting one breaks anybody who already pinned it:

v0.3.0  ==  v0.1.3   (superseded — do not pin)
v0.4.0  ==  v0.1.4   (superseded — do not pin)

Pin v0.1.4.

discovery-runtime

The generic Intent-Discovery runtime: turn a requester's words into a sealed VerifiedIntent.

words → [readers] → fuse (agree kept / disagree → open question) → clarify → seal
                                                                      │
                              refuses while anything result-changing is still open

It is domain-agnostic — parameterized by a schema and a set of readers, it never knows what a dimension means:

from discovery_runtime import DiscoveryRuntime

rt = DiscoveryRuntime(schema=my_schema, readers=[MyRuleReader(), MyModelReader(...)])
vi = rt.draft("…the request…")
for q in rt.clarifications(vi):          # the open "what did you mean?" questions
    vi = rt.resolve(vi, q.field, answer_for(q))
vi = rt.seal(vi)                         # raises SealError if anything result-changing is open

Where it came from, who owns what

Discovery was proved out inside RAAAL/Quantify (the wealth-manager application). This repo is the extraction of the generic mechanism into its own owner, so it can be reused without forking:

Layer	Repo	Holds
Canonical meaning	runtime-contracts	VerifiedIntent, DecisionEvidence, Unresolved, Amendment, SealError
Generic mechanism (this repo)	discovery-runtime	Reader/Reading, merge_readings (fusion), draft_intent/clarifications/resolve, seal, DiscoveryRuntime
Domain semantics (consumers)	wealth-manager (Quantify, 1st), GRC (2nd)	the schema, the reader bodies, the typed policy objects

Functional core / imperative shell. Fusion, canonicalization, hashing, and seal enforcement are pure functions (same input → same output, replay-safe). DiscoveryRuntime is the only object — the shell that owns the schema + readers + policy. No domain verbs (discover_assets, …) ever appear here; domains inject readers and a canonicalize.

Invariants

1. No reader is privileged. Two readers disagreeing on a material field → that field becomes an open question, never a silent pick.
2. Silence is not agreement. Seal refuses while any result-changing dimension is open.
3. Identity is content. VerifiedIntent is frozen; its content_hash is a stable digest of the canonical meaning + evidence + unresolved. A plan re-runs from the hash, never from the sentence.
4. The runtime is domain-free. Meaning lives in runtime-contracts; semantics live in the consumer. This runtime only knows how to discover, contest, clarify, and seal a dimension.

Test

PYTHONPATH=../runtime-contracts python -m pytest tests/ -q

The conformance suite uses a deliberately non-finance toy domain ("book a meeting") to prove the runtime is domain-agnostic.