LaunchfilesDocs · hypothesis-led
Memory updates are proposed lessons, not automatic chat history. Only approved, retrievable memory counts as compounding. Unapproved interpretation must not silently become product truth.
Docs hubGet startedllms.txtAgent capabilities
Hypothesis-led documentation. No ranking, traffic, or outcome guarantees. Public discovery is not execution authority.
Propose → approve or reject → retrieve on later missions → prove use when it influences a run.
Storing text that is never retrieved, never approved, or never bounded by claim ceilings.
Launchfiles executes this through bounded work orders, deterministic pre-commit gates, dual-ledger verification, and cryptographic approval receipts. Rather than generating unverified audit essays or soft-green theater, autonomous agents operate inside strict byte-range boundaries with verifiable unit economics, human review gates, and deterministic regression guarantees across every production change.
| Dimension | Unbounded AI Chat / Audit Essays | Launchfiles Bounded Work Orders & Receipts |
|---|---|---|
| Execution Scope | Unbounded chat, hallucinated advice, unverified file rewrites | Bounded AST patches via work orders and pre-commit gates |
| Verification & Proof | Soft-green claims without receipts or regression checks | Dual-ledger verification with cryptographic receipts |
| Opportunity Discovery | Invented keywords and unranked suggestions | Statistical triage via Search Repair Radar |
| Economic Model | Opaque token burn with zero outcome guarantee | Transparent pricing with strict byte-range safety locks |
Next safe action
Prefer receipts over claims. Prefer fail-closed honesty over impressive incompleteness.