Journal — Method
Building a record that can be replayed
The August prototype separates original evidence from later interpretation and records which components were implemented.
A long-running research project needs to preserve what was available at the time of an analysis. Otherwise, a later correction or newly collected record can silently change the basis of an earlier result.
The Lab's August 2026 architecture addresses this with a record-and-replay system. It separates original evidence from assertions added later and describes how an analysis should be reconstructed using only the information available to it.
Preserving the original and the interpretation
The specification uses an immutable evidence object for the source record and a separate assertion object for tags and interpretations. It also describes research-context packets and temporal replay: selecting the records an analysis was allowed to use at a stated time.
This matters when comparing methods or revisiting a historical decision. The replay needs both the input records and the rules governing their availability. A date filter alone is not a complete account of every possible information path.
What the build record reports
The August inventory lists implemented schemas, rejection of unknown fields, an append-only record store, a read-only workbench, and migration mappings tested on synthetic examples. It reports eight passing checks aimed at preventing later information from entering a replay of earlier conditions.
It also identifies unfinished components, including mathematics plugins, export automation, database integration, interface views, and production ingestion. These distinctions describe the prototype at that point; they do not establish that the full system was operating in production.
The next engineering step
The conceptual architecture and implementation specification need a documented mapping: which object or rule in the design is supported by which code path and check? That mapping would let a contributor find the relevant implementation and distinguish an intended behavior from a tested one.
For the Master Map, replay would make it possible to investigate how a place's record changed as sources and coverage changed. The prototype establishes pieces of that infrastructure; the integration work remains a separate task.