Use this issue to track InQL RFC 032, proposed in PR #60 at docs/rfcs/032_execution_observations.md.
Area
- Specification (RFCs)
- Documentation
Summary
RFC 032 defines execution observations as structured facts emitted by sessions, adapters, and execution attempts. These observations attach runtime evidence to semantic plan targets without making backend logs or adapter behavior the semantic owner.
Motivation
InQL needs local, reviewable evidence for runtime facts such as execution attempts, row counts, backend diagnostics, timing, and capability observations. That evidence should be available to CI, agents, and users without requiring hosted observability or treating unstructured logs as the source of truth.
Proposal sketch
- Define execution attempts and observation records.
- Attach observations to stable semantic targets where possible.
- Capture source, severity/state, timestamps, and metrics when available.
- Link observations to plan profiles, adapter requirements, and execution context.
- Treat unavailable observations as unknown rather than fabricating evidence.
Alternatives considered
- Rely only on backend logs.
- Store telemetry only inside adapter-specific structures.
- Require an external observability platform before InQL can explain execution behavior.
Impact / compatibility
This is additive. It should not change query semantics or execution behavior. Existing sessions can continue without observations and report unknown evidence where no structured observation exists.
Implementation notes (optional)
Review PR #60 for the draft. This work attaches execution facts to the semantic evidence model rather than redefining query meaning.
Checklist
Use this issue to track InQL RFC 032, proposed in PR #60 at
docs/rfcs/032_execution_observations.md.Area
Summary
RFC 032 defines execution observations as structured facts emitted by sessions, adapters, and execution attempts. These observations attach runtime evidence to semantic plan targets without making backend logs or adapter behavior the semantic owner.
Motivation
InQL needs local, reviewable evidence for runtime facts such as execution attempts, row counts, backend diagnostics, timing, and capability observations. That evidence should be available to CI, agents, and users without requiring hosted observability or treating unstructured logs as the source of truth.
Proposal sketch
Alternatives considered
Impact / compatibility
This is additive. It should not change query semantics or execution behavior. Existing sessions can continue without observations and report unknown evidence where no structured observation exists.
Implementation notes (optional)
Review PR #60 for the draft. This work attaches execution facts to the semantic evidence model rather than redefining query meaning.
Checklist