Event Block Reference

The reference for the Event Block: request and response fields, the enum vocabularies, the status transitions, the handler registry and template grammar, the ledger records a handler writes, and the graph edges they materialize to. For the concepts, read Event-Driven Ledger first; for a worked integration, Build a Ledger Integration.

All operations are POST /extensions/roboledger/{graph_id}/operations/{name}, return an OperationEnvelope whose result carries the payload below, and accept an Idempotency-Key header. See Operations Contract. Exact types are in the live spec at robosystems.ai/docs/api.

Table of Contents

• create-event-block Request
• The Event Block Envelope
• Vocabularies
• Statuses and Transitions
• update-event-block
• preview-event-block
• execute-event-block
• Validation and Idempotency
• Reconciling Items
• Handlers
• Ledger Records
• Graph Projection
• Reads
• Errors
• Related Documentation
• Support

create-event-block Request

The Event Block Envelope

EventBlockEnvelope is what every event write and read returns. It echoes the request fields (event_type, event_category, event_class, event_action, agent_id, resource_type, resource_element_id, occurred_at, effective_at, source, external_id, external_url, amount, currency, description, metadata, dimension_ids, obligated_by_event_id, discharges_event_id) and adds:

Vocabularies

event_class and event_category are paired, and the database enforces the pairing:

Support events record value-chain and audit-trail steps and have no ledger effect.

resource_type: goods, services, money, right, obligation, information, labor, or null.

event_action, 19 verbs aligned with Valueflows: produce, raise, consume, lower, use, cite, work, deliverService, pickup, dropoff, accept, transferCustody, transferAllRights, transfer, move, modify, combine, separate, copy. Optional; a capture-first event usually leaves it null.

source: manual, system, and schedule are always accepted. Any other value must name a provider or source registered as a connection on the graph; registering a connection is what opens a new source name.

Statuses and Transitions

• Landing status on create. Capture-only: captured. With apply_handlers: true, the handler sets it (see Handlers); the journal-entry handler lands classified for a draft and fulfilled for an entry posted immediately, and a DSL handler lands classified.
• Retractions are guarded. voided and superseded are refused, whatever the current status, once any of the event's entries or transactions have posted, or the event has published to QuickBooks. Reverse the posted entry instead.
• superseded needs superseded_by_id, naming an existing event other than itself. Both sides of the chain are set together.
• captured or classified → committed fires the event's Python handler unless the event already has an entry linked to it.
• captured → classified gives the handler a chance to refuse a classification it could not post.
• Scheduled obligations (schedule_entry_due) are created pending and moved to classified by the promotion sweep (promote-obligations, or its background sensor) when their month matures.

update-event-block

The metadata patch is applied before the handler runs, so a commit sees the final shape. A handler error rolls back the whole update, status change included. Returns the updated envelope.

preview-event-block

Takes a create-event-block body and writes nothing. Returns:

For multi-leg entries the plan lists the first debit and first credit of each entry.

execute-event-block

Publishes an event to its connection's system of record. Request: event_id, and optionally connection_id to override the event's metadata.connection_id.

Response: event_id, status, qb_external_id, qb_error ({code, message, qb_response_at} on rejection). An event already fulfilled or already carrying a QuickBooks id is returned as-is; a voided or superseded event is refused. Write policies are covered in QuickBooks Sync and Write Policy.

Validation and Idempotency

A create is validated in three phases:

1. Schema. Types and enums on the request.
2. Category and class. The pairing in Vocabularies, enforced by a database constraint.
3. Handler. Only with apply_handlers: true: the handler validates metadata, and the entries it builds must balance.

A capture-only event clears the first two phases and is stored without handler logic.

Idempotency is (source, external_id). A second create with the same pair is refused as a duplicate. Omit external_id and there is no dedup; supply it for anything a sender might retry. This is separate from the Idempotency-Key header, which replays the response to a retried request.

Reconciling Items

When a re-sync brings a changed payload for an event whose entry has already posted, the event is flagged rather than rewritten: is_reconciling_item is true, the incoming payload is held on metadata.drift_payload with metadata.drift_detected_at, and the posted entry and stored payload are left as they were. List them with eventBlocks(isReconcilingItem: true).

preview-reconciling-item shows the change and what each disposition would do; resolve-reconciling-item applies one:

Writes from a disposition are subject to the closed-period gate. Detail is in QuickBooks Sync and Write Policy.

Handlers

Resolution runs in order: the Python registry first, then the graph's DSL handlers. With apply_handlers: true, no match is an error.

Python handlers

The registry is fixed in platform code. The core handlers:

Source transaction types that book the same way share the journal-entry handler: invoice_issued, bill_received, sales_receipt_recorded, cash_expense_recorded, check_written, credit_card_charge, credit_card_refund, deposit_received, inventory_adjusted. Only their inbox label, category, and downstream filtering differ.

journal_entry_recorded metadata takes exactly one of two shapes; both or neither is rejected.

• Flat, one entry: posting_date, memo, line_items[] (each element_id, debit_amount, credit_amount in cents with exactly one non-zero, optional description and metadata), type (standard default, adjusting, closing, reversing), status (draft default, or posted), optional transaction_id.
• Nested, several entries from one event: entries[], each with posting_date, memo, line_items[] (at least two; a line may name element_external_id instead of element_id, resolved through the source connection), type, and optional external_id. status applies to all of them. This is the shape an importer uses when one source transaction produces several journals.

publish_to_source (true or false) overrides whether the entry is published back to the source system; left unset it follows the event's source.

DSL handlers

Registered with create-event-handler (and managed with update-event-handler, list-event-handlers, get-event-handler). A handler row has name, description, event_type, optional match fields (event_category, match_source, match_agent_type, match_resource_type, match_metadata_expression), a transaction_template, priority, and is_active.

Matching. Among active handlers with the event's event_type, a null match field is a wildcard and a set one must equal the event's value. The highest priority among the matches wins; two matches tied at the top priority are an error rather than a guess. A handler suggested by an AI operator is not used until someone approves it.

Template. The transaction_template lists one or more balanced entries:

{
  "transactions": [
    {
      "entry_template": {
        "debit":  { "element_id": "elem_…", "amount": "{{ event.amount }}" },
        "credit": { "element_id": "elem_…", "amount": "{{ event.amount }}" }
      }
    }
  ]
}

A missing field, division of a non-integer, or division by zero is an error. Every entry built must balance, and every amount must be non-negative.

Ledger Records

What a handler writes, level by level:

An entry needs at least two line items and must balance. A posted entry is corrected by reversing it, never by editing it, and an entry can be reversed at most once.

Graph Projection

Events and entries are written to the extensions database and then materialized into the graph, where the audit chain is queryable in Cypher. The graph edges exist only after materialization.

Agent and Event belong to the graph's base schema, since REA is not specific to RoboLedger:

The RoboLedger extension adds the edges from an event to the ledger it produced:

MATCH (e:Event)-[:EVENT_TRIGGERS_TRANSACTION]->(t:Transaction)
      -[:TRANSACTION_HAS_ENTRY]->(en:Entry)
      -[:ENTRY_HAS_LINE_ITEM]->(li:LineItem)
RETURN e.event_type, t.amount, en.status, li.debit_amount, li.credit_amount
LIMIT 10

Run it with the MCP read-graph-cypher tool or the query endpoint in Querying the Analytical Graph.

Reads

GraphQL at POST /extensions/{graph_id}/graphql:

• eventBlocks(eventType, eventCategory, status, agentId, source, isReconcilingItem, limit, offset)
• eventBlock(id)

MCP: list-event-blocks, get-event-block, and the write tools create-event-block, update-event-block, preview-event-block, execute-event-block, promote-obligations.

Errors

Related Documentation

Wiki Guides:

• Event-Driven Ledger - The concepts: events before entries, the lifecycle, capture and commit
• Build a Ledger Integration - Posting events from your own system
• Period Close - How drafts post
• QuickBooks Sync and Write Policy - Write-back and reconciling items
• Operations Contract - The envelope, idempotency, and error format

Codebase Documentation:

• API Reference - Request and response schemas
• Extensions Models - The RoboLedger OLTP models

Support

• Issues
• Wiki
• Projects
• Discussions