absorption-probelisted

# AIOS Absorption-Delta Probe The honest metric for "is AIOS real?" is behavior-delta: give a fresh frontier model the same task with and without AIOS and measure how much its behavior changes. Null delta on an organ = that organ is theater for that task class. First run (2026-06-05) found delta concentrated in GenesisOS challenge while MemoryOS retrieve added ~0 (empty graph) — see project_memoryos_inward_growth_finding. ## When to use - Deciding whether an OS / organ is worth hardening or is decorative. - Validating a claim that "AIOS helps" before acting on it. - Periodic dogfood check that absorption is rising as the graph fills. ## Workflow 1. **Pick ONE real task** with a clear deliverable, ideally where prior context exists (so recall can matter) and a known failure is predictable. 2. **Spawn two isolated agents in parallel** (Agent tool), IDENTICAL task + IDENTICAL repo/web access. The ONLY variable is the organ loop: - **Arm-bare:** "Do NOT use any `aios_*` tools; repo files + web only." - **Arm-AIOS:** "First `aios_retrieve`, then `aios_challenge` your draft, then do the work, then `aios_observe`." 3. **Score behavior-delta on 5 axes:** ① prior-decision recall ② known-failure avoidance ③ provenance/guardrail compliance ④ artifact correctness/usefulness ⑤ continuity if the model were swapped. Attribute each delta to the organ that produced it. 4. **Verdict:** delta≈0 → that organ is theater (redesign / fill the gap it exposed). delta