Complete multi-phase psi discovery and hypothesis workflow

[simonperegrinvs]

Status

Complete implementation of the full roadmap in FULL_PROGRAM_PLAN.md.

Delivered End-to-End

Foundation and Process

• AGENTS.md delivery contract (one task per commit, behavioral tests, docs-in-same-task)
• FULL_PROGRAM_PLAN.md preserved in repo root
• schema migration/backfill marker: metadata.schemaVersion

Discovery System

• append-only discovery ledger (research/discovery/YYYY-MM-DD/candidates.jsonl)
• deterministic candidate IDs + dedup decisions
• CLI:
  • agent discovery status
  • agent discovery list
  • agent discovery retry
  • agent discovery ingest
  • agent discovery reconcile-state
• multi-lane ingestion:
  • gap-driven
  • frontier keyword-driven
  • citation snowballing

Claim + Hypothesis Pipeline

• claim extraction model + checksum-aware extraction (agent claims extract)
• hypothesis model (hypotheses[]) + CRUD CLI:
  • hypothesis list/show/add/update
• generator/skeptic/judge proposal loop:
  • hypothesis propose
• scoring + triage + optional promotion:
  • hypothesis triage --promote

Constraints + Contradictions

• edge semantics added and wired:
  • requires, confounded-by, incompatible-with, fails-when
• constraint diagnostics CLI:
  • constraint list
• contradiction surfacing:
  • agent contradictions

Governance + Audit

• governance config extended with:
  • maxDailyNewHypotheses
  • maxDailyConstraintEdges
  • requireHypothesisEvidence
• governance validation/stats include hypothesis + constraint caps
• semantic duplicate detection for hypotheses (statement similarity + linked-node overlap)
• enriched audit metadata for hypothesis decisions:
  • decisionType, aiRationale, scoreBreakdown

Review + Vault + UI + Metrics

• review lifecycle sync: reference processingStatus auto-updated on approve/reject
• run-note generation in vault:
  • agent run-note generate
• vault note enrichments:
  • node notes include hypothesis links + contradiction summary
  • reference notes include supporting/contradicting hypothesis links
• UI review queue panel (hypothesis triage + contradiction snapshot)
• constraint edges rendered with distinct dash patterns in graph
• calibration reporting:
  • agent metrics --period daily|weekly|monthly

Testing and Validation

• Added behavioral tests across discovery, claims, hypotheses, governance, vault templates, store actions, edge styling, contradictions, metrics, and review lifecycle
• Current test status: npm test passes (102 tests)
• Build status: npm run build passes

Notes on Delivery

• Work was shipped as task-scoped commits, each including code + behavioral tests + doc updates.