How a governed outbound decision moves from source evidence to an approved action

The decision is a chain, not a prompt

A generated message is only the visible end of a longer decision. Before wording is drafted, a source has been selected, an account has been interpreted, a playbook has supplied constraints, and an action has been proposed. Governance means keeping those stages distinct enough to inspect. A fluent draft cannot repair weak evidence, and an approval cannot be inferred from the existence of a draft.

1. Capture evidence with its boundary

The trail starts with retained source material rather than a free-floating summary. Each material fact should point back to a source reference and record when it was captured. The record should also say what the source does not establish. This prevents a later reader from mistaking an inference, an old observation, or a missing field for a verified fact.

• Keep a source URL, document reference, or system record that a reviewer can retrieve.
• Record capture time and any freshness limit that affects the decision.
• Preserve the relevant statement in context instead of storing only a generated summary.
• Mark missing, conflicting, or inaccessible evidence explicitly.

2. Turn evidence into a reviewable proposal

Qualification applies declared rules to the retained evidence. A useful review record separates three things: what the source says, what the operating rules infer, and what action the system proposes. The proposal should name the intended recipient, channel, timing, playbook version, relevant exclusions, and unresolved uncertainty. A reviewer can then challenge the reasoning without reconstructing it from the final message.

3. Keep approval separate from execution

A proposed action is not an approved action. The approval record should identify the reviewer, timestamp, decision, material shown at review, and any edits or exceptions. If the evidence changes after review, the earlier approval should not silently cover the new state. The proposal returns to review or remains blocked according to the configured policy.

4. Preserve the trail after the action

Governance continues after execution. The handback should retain what was actually sent or recorded, the execution state, any failure, and later response context. Non-response stays in the record. If the evidence suggests a playbook revision, the proposed change, reviewer decision, and prior version remain linked instead of overwriting history.

Where this method stops

This is a learning artifact and operating-method description, not verified end-to-end product proof. Castia does not yet expose a complete buyer-facing trace and export for this full chain. The method does not establish delivery, reply, meeting, pipeline, revenue, or customer-performance effects. Its usefulness depends on source quality, configured rules, reviewer judgment, and complete event capture. If a source or decision record is missing, the trail should be treated as incomplete rather than filled with an assumption.