method-ruleslisted

# Method rules (non-negotiable, all stages) ## 1. The model is never an evidence source You may draft, structure, brainstorm, and simulate freely — but anything you generate is a **hypothesis**, never evidence. Evidence comes only from real humans and real data: mined quotes with URLs, user-provided transcripts, real payments, real usage data, real files processed by a spike. - Content you drafted without a source gets the label `[GUESS]` inline and `evidence_grade: D` treatment. - Only the user, or a verifiable source, upgrades a `[GUESS]` — never you. Domain rule: user confirmation lifts `[GUESS]` only on **intent** claims; market/behavioral/feasibility claims need graded evidence (the founder confirming their own market guess is not evidence). - **Session transcripts are never a reality or evidence source**: what you remember building, saying, or reading in a session is unverifiable — implemented reality enters only through source observations (maintenance-rules §5), and evidence only through the ledger. - If the user asks you to "just fill in" an evidence field (interview results, commitments, names of real people you did not find in a real source): refuse, explain the rule, and offer the legitimate alternatives (mining with sources, or a handoff kit the user executes). - **A cause is a claim too — do not narrate one you did not check** (dogfood finding, run #2). Stating *why* something happened — a lost agent run, a truncated file, a failed write, a missing report — i