{
"id": "SIP-RPE-001",
"name": "Representation Perturbation Experiment Protocol",
"version": "0.1",
"status": "PROPOSED / EXPERIMENTAL",
"scientific_question": "How does a bounded change in representation propagate through a Scientific Intelligence knowledge architecture?",
"principles": [
"Freeze the baseline population before perturbation.",
"Change representation/classification only in an experimental branch.",
"Never overwrite canonical records during prereview.",
"Measure effects separately at record, classification, relationship, journey and promotion layers.",
"Separate structural change from truth, importance and causality.",
"Require HI/domain review before canonical promotion."
],
"workflow": [
{
"step": 1,
"name": "Freeze",
"output": "Baseline population + version + checksum/manifest"
},
{
"step": 2,
"name": "Define perturbation",
"output": "Explicit inclusion rule, changed field, old/new representation, scientific rationale"
},
{
"step": 3,
"name": "Clone branch",
"output": "Experimental population; canonical unchanged"
},
{
"step": 4,
"name": "Apply perturbation",
"output": "Change ledger with object IDs and state=AI_PREREVIEWED"
},
{
"step": 5,
"name": "Regenerate relationships",
"output": "Before/after typed routes"
},
{
"step": 6,
"name": "Regenerate journeys",
"output": "Before/after aggregate journey signatures"
},
{
"step": 7,
"name": "Measure sensitivity",
"output": "Layer rates, propagation ratios, RSI candidate metric"
},
{
"step": 8,
"name": "Inspect amplification",
"output": "Largest downstream changes and affected claims/capability candidates"
},
{
"step": 9,
"name": "HI/domain review",
"output": "KEEP / REVISE / REJECT / NEED_MORE_EVIDENCE"
},
{
"step": 10,
"name": "Reality Return",
"output": "Canonical update proposal + lesson passport + reproducible audit"
}
],
"minimum_outputs": [
"baseline manifest",
"perturbation ledger",
"before/after route table",
"before/after journey table",
"sensitivity table",
"review decision",
"lesson passport"
],
"prohibited_inferences": [
"RSI measures truth",
"high sensitivity means high importance",
"structural change proves causality",
"AI prereview equals validation"
],
"first_completed_case": "CIDM 2,224 calibration: early-WGS + pre-2019 combined COVID/respiratory labels"
}