Card guide

Historical review

agent.history_review runs a finite research task over the assigned laser’s recorded evidence.

Ask a bounded research question

Use normal and failed periods, provenance, counterexamples and explicit uncertainty to examine performance changes or propose a better method. The card uses the same pinned execution profiles and capability checks as agent.task, with the research instruction profile.

id: historical-review
operation: agent.history_review
uses: []
agent_execution:
  mode: required
  task_profile: analysis
parameters:
  agent_card_id: gateway-agent
  execution_profile: analysis
  instruction_profile: research
  lookback_s: 86400
  timeout_s: 1800
  prompt: "Compare normal and degraded periods. Identify competing explanations and a validation plan."

gateway-agent is the example gateway card ID. Use the ID configured in your laser deck; the UI shows its role as Gateway agent. Display names do not change routing IDs.

The profile ID must already exist in Agent settings. A monitor can select this card through history_task_card_id, with both an interval and a minimum amount of new recorded evidence. Manual task admission uses the same registered-card and profile checks.

Record	Purpose
Instruction and evidence bundle	Pins the task profile, objective, provenance and bounded input without silently copying an entire resident conversation.
History and knowledge tools	Read the assigned laser’s records and propose evidence-backed knowledge with limitations. A proposal does not change physical policy.
Task status and result	Record effective model/effort, capability evidence, completed analysis and unresolved questions.

Interpret the result

Historical correlation is not proof of causation. Distinguish measurement validity, configuration changes, path dependence, selection effects and missing coverage. A useful proposal specifies the supporting records, counterevidence and a bounded validation experiment.

This card cannot select the adjustment instruction profile. Historical review does not start a lock, grant a new safety envelope or adopt an unverified threshold change. Proposed methods remain proposals until the appropriate reviewed operation validates them.

Read the recorded result and local report before applying its conclusions. Automatic notifications provide a short outcome and next action; complete evidence stays with the task. Unknown interrupted work is retained for inspection and is never automatically replayed.