RunRecord

A RunRecord bundles everything about one execution — the chain snapshot, the input, the (lossless) result, and timing — into a single JSON-serialisable record. It’s what MAESTRO stores as run history and what care replay reads back.

Capture a run

from datetime import datetime, timezone
from mmar_carl import RunRecord

started = datetime.now(timezone.utc)
result = await chain.execute_async(context)

record = RunRecord.from_run(
    chain=chain,
    context=context,
    result=result,
    started_at=started,
    chain_id="my-chain",        # optional
    chain_version="v3",         # optional
)

from_run snapshots chain.to_dict(), captures outer_context + memory + context metadata (dropping framework-internal __-prefixed keys), and stamps finished_at = now() if you don’t pass one.

Fields

Field	Meaning
chain_id / chain_version	Identity of the executed chain (optional).
chain_dict	The full serialized chain (durable snapshot).
input	{outer_context, memory} the run started from.
result	The lossless ReasoningResult.
started_at / finished_at	Timestamps.
runtime_info	Library version + environment info.

Persist & reload

record.save("runs/run-001.json")
loaded = RunRecord.load("runs/run-001.json")

d = record.to_dict();  RunRecord.from_dict(d)

Then step through it with care replay run-001.json.

See also

• Result serialization — the ReasoningResult inside a record.
• MAESTRO integration overview