Learning content — teaches MindsEye architecture through structured examples
Evaluation suite — validates state machine constraints
Retrieval store — searchable knowledge base for RAG + agent systems
Demo backend — powers live portfolio queries using the HF datasets-server API

Dataset Structure

The dataset uses a single-table design: multiple logical tables are stored together, separated by a required "table" field.

Logical Table	Purpose
`ledger_events`	Append-only event log of state transitions & interactions
`cards`	Template cards (CONCEPT, FLOW, SPEC, TEST, BUILD)
`repo_registry`	Repository → State mappings
`tests`	State constraint validation tests

Why single-table?

Because it makes the dataset easy to query using the Hugging Face datasets-server endpoints:

/filter (SQL-like filtering)
/search (full-text search)
/rows (pagination)

Core Concepts

1) Append-only ledger

ledger_events is meant to model an immutable history. Nothing “updates” the past — new events extend the timeline.

2) State machine learning

States like CORE_OS, LEDGER_CORE, DATA_SPLITTER, SQL_CORE, LAW_N, etc. are used to anchor knowledge and tests.

3) Cards as teachable units

Cards are compact “knowledge objects” that can be retrieved by state, tag, or keywords. They can also serve as system prompts / templates.

4) Tests as constraint enforcement

Tests encode the “laws” of the system (determinism, append-only, reversibility, routing constraints).

Schema Overview

`ledger_events` row example

{
  "event_id": "evt_2026_01_10_000001",
  "ts": "2026-01-10T01:12:33+02:00",
  "turn": 1,
  "actor": "user",
  "event_type": "STATE_INIT",
  "state": "CORE_OS",
  "repo_context": ["minds-eye-core"],
  "command": null,
  "input_text": "Initialize MindsEye Lab",
  "output_ref": null,
  "tags": ["init", "core"],
  "table": "ledger_events"
}

Downloads last month: 28

You need to agree to share your contact information to access this dataset

MindsEye Lab Memory Ledger