Trust layer
EvidencePacket quality drives card quality.
The model or deterministic generator should never receive raw unbounded data. It should receive a clean EvidencePacket that defines known facts, plausible context, blocked claims, and uncertainty.
Packet contract
{
"packet_id": "2026_04_25_indigo_packet",
"date": "2026-04-25",
"asset": {},
"session": "post_market",
"move": {},
"peer_moves": [],
"sector_context": {},
"market_context": {},
"events": [],
"evidence_items": [],
"hypotheses": [],
"allowed_claims": [],
"blocked_claims": [],
"confidence_recommendation": "medium",
"card_type_recommendation": "move_anatomy",
"created_at": "2026-04-25T13:00:00.000Z"
}Evidence item rules
- One evidence item should support one claim-like fact.
- Every numeric value used in a card must come from the packet.
- Every item needs source name, source timestamp, and confidence.
- Negative evidence is valid. “No clear public catalyst was found” is a useful evidence item when properly scoped.
Allowed vs blocked claims
Allowed
“The move lines up with a material filing.”
“Volume was 2.3x the 20-day average.”
“Margin impact was not disclosed.”
Blocked
“The stock rose because of the filing.”
“The stock will continue rising.”
“Buy, sell, hold, exit, or accumulate.”