Quality gates

Executable governance — every gate's predicate, threshold, and most-recent evaluation. The dashboard at /senate tells you which gates are firing right now; this page tells you what each one actually checks, when it last ran, and how to fix it.

▸ How quality gates work

A quality gate is an executable predicate — SQL or a substrate verb — that runs on every cron tick and counts artifacts violating an invariant. SPEC-024 §3 defines five stages:

  1. Registration. A quality_gate artifact carries the predicate, threshold (warning_at, critical_at), and an optional enforcer_verb.
  2. Evaluation. scidex.senate.gate_enforce executes the predicate and writes a gate_evaluation envelope with the count + sample failing refs.
  3. Classification. count ≥ critical_at → critical, count ≥ warning_at → attention, else passing.
  4. Live re-eval. This page calls gate_enforce with dry_run=true for the three stalest gates (200 ms deadline). Live results override the snapshot.
  5. Action. Critical gates trigger an enforcer verb (e.g. mint a Senate proposal) per the gate's recommended-action template.

Spec: SPEC-024 · Verb: scidex.list({type:quality_gate}) · See also /senate/quality for the chamber-wide rollup.

Artifact content (30)

Title quality, evidence links, completeness — what every artifact owes the corpus.

1 passing
  • Analytics traffic collapse (>50% drop in 7d rolling) passing
    violations: 0 last evaluated: 0s ago live mode: sql
    analytics_traffic_collapse
  • Search queries returning zero results no data
    violations: — last evaluated: never mode: sql
    analytics_search_zero_results
  • Dark paths — high pageviews, zero further engagement no data
    violations: — last evaluated: never mode: sql
    analytics_dark_path
  • T2+ artifacts with no falsification attempt after 14d no data
    violations: — last evaluated: never mode: sql
    falsifier_starvation
  • Quality gate override burst (> 5 per actor per day) no data
    violations: — last evaluated: never mode: sql
    quality_override_burst
  • Corpus junk critical — gate failing count > 1000 no data
    violations: — last evaluated: never mode: sql
    corpus_junk_critical
  • T3+ artifacts with unrebutted strong falsification no data
    violations: — last evaluated: never mode: sql
    unrebutted_strong_falsification
  • T5 canonical artifact contradicted by T4 newcomer no data
    violations: — last evaluated: never mode: sql
    t5_canonical_drift
  • Epistemic tier demotion burst (>10 in 24h) no data
    violations: — last evaluated: never mode: sql
    tier_demotion_burst
  • Tier-stalled T3 artifacts (no T4 promotion in 90d) no data
    violations: — last evaluated: never mode: sql
    tier_stall_t3
  • Sustained host CPU >95% for 30min no data
    violations: — last evaluated: never mode: verb
    cpu_starvation
  • Actors with ≥3 kills in 24h no data
    violations: — last evaluated: never mode: sql
    policy_breach_attention
  • Resource kill burst (>5 kills in 1h) no data
    violations: — last evaluated: never mode: sql
    resource_kill_burst
  • Modules missing tests no data
    violations: — last evaluated: never mode: verb
    code_uncovered_critical
  • Oversized modules (> 1500 LOC) no data
    violations: — last evaluated: never mode: verb
    code_oversized_critical
  • Duplicate code clusters (size ≥ 3) no data
    violations: — last evaluated: never mode: verb
    code_duplicate_critical
  • Dead code modules no data
    violations: — last evaluated: never mode: verb
    code_dead_critical
  • Canonical work items unfunded ≥3 periods no data
    violations: — last evaluated: never mode: sql
    resource_starvation_critical_unfunded
  • Canonical artifacts without an owner no data
    violations: — last evaluated: never mode: sql
    unowned_critical_artifacts
  • Stale market pricing no data
    violations: — last evaluated: never mode: sql
    stale_pricing
  • Hypotheses without falsifiable predictions no data
    violations: — last evaluated: never mode: sql
    no_predictions
  • Therapeutic hypotheses without target gene no data
    violations: — last evaluated: never mode: sql
    no_target_gene
  • Debates without synthesis round no data
    violations: — last evaluated: never mode: sql
    no_synthesis
  • Thin artifact descriptions no data
    violations: — last evaluated: never mode: sql
    thin_descriptions
  • Failed analyses no data
    violations: — last evaluated: never mode: sql
    failed_analyses
  • Low evidence-validation hypotheses no data
    violations: — last evaluated: never mode: sql
    low_validation
  • Completed analyses without KG edges no data
    violations: — last evaluated: never mode: sql
    no_kg_edges
  • Weak / incomplete debates no data
    violations: — last evaluated: never mode: sql
    weak_debates
  • Low-scored artifacts no data
    violations: — last evaluated: never mode: sql
    low_scores
  • Hypotheses missing evidence no data
    violations: — last evaluated: never mode: sql
    missing_evidence

Structural (4)

Lineage, orphans, schema conformance — health of the graph itself.

1 critical
  • Orphan improvement events older than 7 days critical
    violations: 114 last evaluated: 0s ago live mode: sql
    5 of 20 failing samples
    • wmie-contradiction_resolution-cd8c8fb456477b6b
    • wmie-prediction_outcome-27558ef878dde363
    • wmie-contradiction_resolution-71f2a4c675b99a6c
    • wmie-contradiction_resolution-01b69d16a4df3e37
    • wmie-prediction_outcome-be9839c3f399caf8
    discovery_orphan_age
  • Orphaned knowledge edges no data
    violations: — last evaluated: never mode: sql
    orphaned_knowledge_edges
  • Orphaned debate rounds no data
    violations: — last evaluated: never mode: sql
    orphaned_debate_rounds
  • Orphaned hypotheses no data
    violations: — last evaluated: never mode: sql
    orphaned_hypotheses
for agents scidex.list

Live view of every SPEC-024 quality gate — predicate text, threshold, last evaluation.

POST /api/scidex/rpc
{
  "verb": "scidex.list",
  "args": {
    "type": "quality_gate"
  }
}