aletheia-design-evidence

Aletheia Design Evidence

Primitive Capabilities

Use these primitives. Do not add orchestration to runtime scripts for evidence judgment:

• system_context.py or orient.py to load active node, parent constraints, and relevant risks.
• truth_record.py list/show to inspect existing evidence, hypotheses, decisions, and risks.
• truth_record.py create/update to write evidence, hypothesis, risk, or decision records.
• truth_record.py create/list/show/update/archive orphan when a claim lacks a defensible parent node.
• validate.py to verify required evidence sections and references.
• checkpoint.py --dry-run and checkpoint.py only when the user wants the evidence update committed.

Prompt Recipe

The skill is a prompt recipe for designing falsifiable claims and evidence.

For each important claim, record:

• claim;
• method;
• evidence type;
• expected observation;
• invalidation criteria;
• interpretation rule;
• graph impact;
• follow-up decision.
• claim lifecycle impact.

Link the evidence back to affected system graph or skeleton nodes. Do not treat implementation success, benchmark output, or model confidence as self-explanatory evidence. If the claim does not yet have a defensible parent node, keep it in the orphan incubator with truth_record.py create orphan, candidate parents, and required evidence instead of forcing it into the main tree. When evidence weakens or falsifies a hypothesis, update that hypothesis Lifecycle and review whether dependent decisions still stand.