PortAbleRI · public POC
Erase What? Host NOT · proof of concept
Tier A = measured on device · B = reproducible · C = scenario / HOLLOW until benchmark
Erase What? Host NOT
They cannot erase what they do not host. Verify locally. Export always.
Tier A · demonstrated (M2 8 GB · Sep 22–23, 2026)
| Claim | Evidence |
|---|---|
| Full stack RAM stress | qwen3:8b ~6 GB · swap pressure |
| Lean operability | llama3.2:3b ~2.5 GB · lean mode |
| Pipeline slimming (shipped code) | lean + enabled_cells · kernel test passes |
| Calls per turn | 7+ → 2 (cascade + lux) |
| Lean contact turn | ~129 s wall-clock (trial) |
| Local activation | env + venom run — no datacenter step |
Verify on your bench
| Claim | How |
|---|---|
| Continuity without provider memory | venom export · venom verify |
| Publish doc integrity | publish/*.md + sidecar · this site’s Verify page |
| Public provenance | node processes/provenance/provenance.mjs → exit 0 |
Tier C · not claimed as fact on this page
Universal 50–60× speed · billions saved instantly · vendor intent — require labeled benchmarks (master white paper Part II–III).
We’re not 60× smarter. We’re 60× less trapped — fewer rented calls, local custody, whole-line execution.
Full Crystal proof pack · family correspondence · recovery ledgers — private ring only (processes/PRIVATE-RING.md).