# Observe and operate

One connects telemetry, causal facts, plans, actions, effects, and evidence
without treating logs as authority or history as magical rollback.

## Observation envelopes

Logs, metrics, traces, profiles, audits, lineage, and cost signals use typed
observation envelopes with identity, causality, provenance, classification,
budgets, source, and exact schema.

Observation providers can store and project these signals differently without
changing their semantic owner.

## History owns causal facts

History records immutable product and operational facts:

- plans, approvals, grants, builds, deployments, and actions;
- effect intents, dispatches, confirmations, and unknown outcomes;
- state transitions, compensation, reconciliation, and irreversibility;
- observations and evidence linked by causal edges.

Current state is a materialized view over facts. Repository refactor history
does not belong here.

## Explain before acting

Inspection can answer:

- why an artifact or capability entered a root;
- which default, policy, mapping, or provider selected it;
- which action produced an observed state;
- which evidence is missing or stale;
- what smallest source, policy, or placement change would remove it.

An explanation traverses owned public proof edges. It does not infer authority
from telemetry.

## Replay is bounded

Replay is possible only for recorded deterministic inputs and compatible
program/history revisions. External, human, financial, legal, and physical
effects remain observations and receipts, not reversible simulation.

## Incident operations use ordinary authority

Diagnosis, quarantine, rollback, compensation, provider failover, and
break-glass actions are exact operation contracts with plans, approvals,
grants, actions, and audit evidence. Operators and AI assistants use the same
path.

Canonical owners:
[Observability](https://github.com/muijf/one/blob/main/systems/observability/AGENTS.md)
and
[History](https://github.com/muijf/one/blob/main/systems/history/AGENTS.md).