Nightly Datagraph rebuild

The nightly datagraph rebuild is the scheduled pipeline that reconstructs the platform's full knowledge graph from its canonical flat-file sources. Every graph-queryable relationship — entity links, extraction outputs, ledger entries, and location intelligence indexes — is derived from the same deterministic inputs each cycle. The result is a fresh, stable snapshot available to all query consumers at the start of each operating day.

Key Takeaways

• The graph is rebuilt from flat-file sources nightly, not maintained by continuous mutation. Consumers read a stable snapshot, not a live partially-constructed graph.
• Every rebuild cycle can be replicated from archived flat files, since the WORM ledger snapshot each cycle reads from is itself immutable.
• Entity extraction uses grammar-constrained inference through the Doorman — this is Ring 2 calling Ring 3 for a proposal, not a Ring 2-internal deterministic step. A capture-then-promote checkpoint exists for exactly this case — extraction output can be held for a human-reviewed, cryptographically-signed approval before it becomes a graph write, matching the SYS-ADR-07 boundary. That checkpoint is opt-in, not the default: an operator must explicitly enable it. Left at its default setting, extraction output writes to the graph immediately with no per-item review. See the Three-Ring Architecture for the general rule this checkpoint implements when enabled.
• Each cycle compounds the prior cycle. Newly committed records extend the graph; no record is removed. The Compounding substrate mechanism means the graph grows monotonically accurate over time.

Purpose

The rebuild pattern ensures that the queryable substrate reflects the committed state of the canonical record, not accumulated in-memory drift. Any single run can be replicated from the archived flat files.

Schema-driven joins against the canonical taxonomy and location intelligence indexes are deterministic — no fuzzy matching. Entity extraction itself is not: it is grammar-constrained inference through the Doorman, producing a structured record. Whether a human reviews that record before it becomes a graph write depends on a setting an operator controls, off by default — see the compliance note above. This is a real, currently-open gap between the SYS-ADR-07 boundary's intent (AI never writes to a structured record store directly) and this pipeline's default configuration, tracked separately with Command.

Pipeline stages

The rebuild pipeline follows a fixed sequence:

1. Ledger snapshot — reads the current committed state of all WORM ledger segments. The ledger is append-only; the snapshot is the complete history as of the scheduled start time.
2. Extraction pass — service-extraction hands corpus text to service-content, which calls the Doorman for grammar-constrained entity extraction, producing structured entity records for persons, organisations, assets, and events. This is a Ring 2 → Ring 3 call, not a deterministic step — whether these records land as a human-reviewable pending item or write straight to the graph depends on the operator setting described above.
3. Schema-driven joins — entity records are joined against the canonical taxonomy and location intelligence indexes using explicit foreign-key relationships. No fuzzy matching at this stage.
4. Graph construction — joined records are assembled into the queryable graph substrate consumed by service-content and the Doorman inference layer.
5. Swap — the completed graph replaces the prior snapshot atomically. Query consumers switch to the new version at the next request after the swap.

Position in the substrate stack

The nightly rebuild sits between the WORM ledger (which accumulates append-only writes during the day) and the query-serving tier (which reads the most recently completed graph). Consumers of the knowledge graph always read a stable snapshot, not a partially-constructed graph.

The Compounding substrate mechanism means each rebuild cycle inherits the full prior graph, then adds newly committed records on top. Accuracy compounds over time: an entity that appeared in three ledger records two years ago and twelve ledger records last month has a richer graph node than a newly registered entity — without any manual curation step.

See also

• Compounding substrate — the mechanism by which each rebuild cycle compounds prior knowledge
• WORM ledger design — the append-only ledger that feeds the rebuild pipeline
• service-extraction — the DataGraph ingestion pipeline — the extraction service that produces entity records consumed by the rebuild
• service-content — entity extraction and knowledge-graph host — the query-serving service that reads the completed graph
• Doorman protocol — the inference-layer client that queries the graph for entity context