metaspn-gates 0.1.8

Config-driven gate and state machine evaluator for MetaSPN entity pipelines

metaspn-gates

metaspn-gates is a config-driven gate + state-machine evaluator for MetaSPN entity pipelines.

Current implementation (v0.1 seed)

• Deterministic gate evaluation from config
• Hard requirements and soft thresholds
• Per-gate per-entity cooldown checks
• Transition attempt snapshots
• Transition application + task emissions
• Config parsing and validation from mapping
• metaspn-schemas integration hooks (currently used for typed emission shaping)
• Optional metaspn-schemas emission shaping (Task + EmissionEnvelope)

Public API

• evaluate_gates(config, entity_state, features, now)
• apply_decisions(entity_state, decisions, caused_by=None)
• parse_state_machine_config(payload)
• load_state_machine_config(path)

Notes

• load_state_machine_config can use metaspn_schemas parsing/validation hooks when exposed by that package version.
• Canonical schema hooks:
  • parser: parse_state_machine_config (mapping payload)
  • validator: validate_state_machine_config
• If metaspn_schemas is unavailable, it falls back to JSON parsing only.
• Current dependency target: metaspn-schemas>=0.1.0,<0.2.0.
• apply_decisions(..., use_schema_envelopes=True) attaches schema-shaped payloads when entity_state.entity_id is present.

M0 Minimum Keys

For the sample M0 progression (SEEN -> OBSERVED -> PROFILED) the minimum keys are:

• entity_state.state: SEEN or OBSERVED
• entity_state.track: M0
• features.ingestion.resolved_entity_id: required for SEEN -> OBSERVED
• features.profile.handle: required for OBSERVED -> PROFILED
• features.profile.confidence: numeric threshold for OBSERVED -> PROFILED (sample gate uses >= 0.7)

Reference fixture:

• /Users/leoguinan/MetaSPN/metaspn-gates/tests/fixtures/m0_state_machine_config.json

M1 Required Keys

For the M1 routing progression (SEEN -> OBSERVED -> PROFILED -> QUALIFIED -> ROUTED) the minimum keys are:

• entity_state.state: one of SEEN, OBSERVED, PROFILED, QUALIFIED
• entity_state.track: M1
• entity_state.entity_id: required for emitted task metadata
• features.ingestion.resolved_entity_id: required for SEEN -> OBSERVED
• features.profile.handle: required for OBSERVED -> PROFILED
• features.profile.confidence: thresholded for OBSERVED -> PROFILED
• features.social.followers: required floor for PROFILED -> QUALIFIED
• features.scores.routing_readiness: required for PROFILED -> QUALIFIED and QUALIFIED -> ROUTED
• features.scores.profile_quality: thresholded for PROFILED -> QUALIFIED

M1 reference fixture:

• /Users/leoguinan/MetaSPN/metaspn-gates/tests/fixtures/m1_routing_state_machine_config.json

M2 Required Keys

For recommendation gating (QUALIFIED -> RECOMMENDABLE -> DRAFT_READY) the minimum keys are:

• entity_state.state: QUALIFIED or RECOMMENDABLE
• entity_state.track: M2
• entity_state.entity_id: required for downstream task metadata
• features.profile.handle: required recommendation eligibility hard requirement
• features.scores.recommendation_score: recommendation and draft thresholds
• features.scores.intent_score: recommendation eligibility threshold
• features.context.channel: channel-scoped cooldown key
• features.context.playbook: playbook-scoped cooldown key
• features.context.suppress_recommendations: suppression flag for recommendation gate
• features.context.suppress_drafts: suppression flag for draft gate

M2 reference fixture:

• /Users/leoguinan/MetaSPN/metaspn-gates/tests/fixtures/m2_recommendation_state_machine_config.json

M3 Learning Outputs

Learning helpers are exposed for reviewable gate analytics artifacts:

• evaluate_attempt_outcomes(...)
• classify_failure_reason(...)
• generate_calibration_proposals(...)

These produce deterministic outputs for:

• moved-too-early detection
• false-negative/false-positive classification
• calibration recommendations (threshold/cooldown direction hints)

Calibration Safety Constraints

• Calibration outputs are proposal-only artifacts.
• No threshold or cooldown values are auto-applied silently.
• Every proposal has auto_apply=False and must be explicitly approved by a human/operator workflow.

Demo Support

Demo fixture:

• /Users/leoguinan/MetaSPN/metaspn-gates/tests/fixtures/demo_state_machine_config.json

Demo progression:

• SEEN -> OBSERVED enqueues enrich_profile
• OBSERVED -> SHORTLISTED enqueues score_entity
• SHORTLISTED -> READY optionally enqueues draft_message when features.demo.draft_enabled=true

Digest explainability helper:

• format_decision_trace(decisions) returns rows with passed, blocked, and reason.

Token Health Support

Token fixture:

• /Users/leoguinan/MetaSPN/metaspn-gates/tests/fixtures/token_health_state_machine_config.json

Token minimum keys:

• entity_state.state: CANDIDATE or WATCHLISTED
• entity_state.track: TOKEN
• features.token.holder_count: hard requirement floor
• features.token.creator_behavior_risk: hard requirement ceiling
• features.scores.credibility: threshold for both gates
• features.scores.organic_volume: soft threshold
• features.scores.concentration: soft threshold
• features.token.address, features.token.chain, features.token.symbol: copied into task emission metadata for token workers

Release

• GitHub Actions workflow: /Users/leoguinan/MetaSPN/metaspn-gates/.github/workflows/publish.yml
• Publish trigger: GitHub Release published (or manual workflow_dispatch)
• Publishing method: PyPI Trusted Publishing via pypa/gh-action-pypi-publish

One-time setup

1. In PyPI, create project metaspn-gates and configure Trusted Publisher for this GitHub repo/workflow.
2. In GitHub, create environment pypi (optional protection rules supported).

Release flow

1. Bump /Users/leoguinan/MetaSPN/metaspn-gates/pyproject.toml version.
2. Tag and push a release commit.
3. Publish a GitHub Release for that tag.
4. publish.yml builds and uploads to PyPI.