Rule signal promotion criteria

Rule

A candidate signal must only be promoted to a pattern when it satisfies all mandatory criteria and is supported by observable evidence across the system.

Promotion Criteria

A signal is eligible for promotion only if ALL of the following are satisfied:

Recurrence (mandatory)
The signal appears multiple times across:
architecture_work entries, OR
alignment findings, OR
repeated conversational emergence
System Impact (mandatory)
The signal affects at least one of:
system behaviour
governance
execution flow
contracts
consistency / drift
Reusability (mandatory)
The insight applies beyond a single instance or local context
Stability (mandatory)
The signal is not transient, tool-specific, or one-off

Evidence Requirement

At least one of the following must support the signal:

Referenced in alignment findings (e.g. related_signals)
Captured in architecture_work
Repeated across multiple runs or interactions

Non-Promotion Rule

If the above criteria are not fully satisfied:

The signal MUST remain a candidate signal
It MUST NOT be promoted to a pattern
It MAY be refined or re-evaluated in future iterations

Rationale

To prevent premature or noisy pattern creation, promotion must be based on consistent, system-level signals rather than isolated observations, ensuring patterns represent stable and reusable architectural knowledge.

Applies To

signal → pattern promotion workflow
architecture_work decision-making
alignment-engine (candidate signal evaluation)
architecture-work-to-adr validation

Example

A signal like "generators behave inconsistently without explicit contracts" is promoted because it:

appears across multiple runs (recurrence)
impacts execution and governance (impact)
applies to all generators (reusability)
remains valid across tools (stability)
is supported by alignment findings and architecture work (evidence)