Skip to content

P0: PostgreSQL Knowledge Substrate + Graph Projection + Pandora Adapter #113

Open

Description

@bitmaster162

Objective

Build the durable relational/graph substrate that can consolidate historical memory while preserving existing responsibility boundaries.

Baseline

Planning baseline observed on protected master: 021e2d521efc4df0ce390b38a919bc2f0b675460 / tree 803d89c3e81ea3a83147cd63879a32f8a88c9afe.
Re-check before any source write.

Proposed stack

  • PostgreSQL as canonical relational query substrate for large-scale knowledge/history
  • pgvector as derivative semantic index
  • PostgreSQL lexical search first
  • ArchiveOS/content-addressed storage for raw bytes
  • SQLite retained for local/edge single-backend operation where appropriate
  • graph is a projection, not independent authority

Required domains/tables

sources, source_objects, artifacts, artifact_versions, artifact_locations, segments, entities, entity_aliases, events, event_links, causal_spines, causal_frontiers, claims, claim_evidence, claim_counterevidence, contradictions, decisions, decision_inputs, decision_outcomes, projects, project_states, state_observations, current_state_projection, agent_runs, agent_events, agent_outputs, agent_relationships, skills, skill_versions, skill_runs, effects, effect_receipts, commercial_accounts, commercial_opportunities, commercial_events, embeddings, retrieval_documents.

Temporal contract

Support both:

  • valid time (when true in the world)
  • transaction/ingestion time (when learned/recorded)

Graph projections

G0 provenance
G1 events/supersession
G2 claim/evidence/contradiction
G3 causal spine
G4 decision/outcome/calibration
G5 agent trajectory/tool/artifact
G6 product/project/dependency/current-state
G7 Person Twin

Pandora

Preserve existing Pandora as visualization/simulation projection. Add a stable read-only graph projection contract that can feed Epoch DAG / 3D / causal views. Pandora must not own canonical truth or approval.

Technology evaluation

Adapt useful open-source patterns from Graphiti/Mem0/Letta/OpenAI Agents SDK/Microsoft Agent Framework/Google ADK subject to license attribution and without importing their authority semantics.

Acceptance

  • versioned SQL migrations
  • read/write/readback transaction tests
  • bitemporal fixtures
  • supersession/correction fixtures
  • relational -> graph projection reproducibility
  • pgvector index rebuild must never change canonical state
  • no dual canonical DB sync
  • one active EvidenceStore backend per runtime

No-effect boundary

No Drive mutation, no deployment, no runtime cutover, no trading/capital effect from this issue.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions