Software-only verification tooling and red-team harnesses for the AI Futures Project's AI 2040: Plan A verification agenda. Each repo states what it does, what it does not do, and how to check its claims yourself.
| Repo | What it is | Check it |
|---|---|---|
| fides | Reference implementation of a packet-hashing, inference-only verification tap. Each module maps to a row of Amodo Design's Plan A SITREP: 7 of 17 workstreams addressed in software (2 fully, 5 as narrow slices); the 6 physical ones are explicitly out of scope. | pip install -e ".[dev]" && pytest && python verify_claims.py |
| ai-2040-plan-a-toolkit | RVP-1 protocol package: trust-root composition checker, receipt schema and validator, compute accounting, workload discrimination, red-team benchmarks. | pip install -r requirements-rvp1-lock.txt && pip install -e . && pytest -q (282 tests) |
| frontier-verify | Tamper-evident evidence and receipt pipeline plus a recomputation prototype. Pre-alpha; says so in its README. | make install && make test |
| ai-2040-verification | 30-section experimental notebook: trained toy language model, DiFR-style server, LoRA backdoors, red-team runs, Rust tap reader. | make test |
| continental-load-registry | Track 3: public large-load interconnection-queue registry and a conservative residual-compute bound. | pip install pytest && pytest tests/test_track3_bound.py tests/test_track3_tail_audit.py |
| lumen-align-audit | Audit toolkit built against Plan A's Alignment Roadmap: faithful CoT, deception probes, model organisms, AI control. | pip install -e ".[dev]" && pytest (about 5 minutes) |
Feedback, red-team results and corrections are welcome. Contact: AnonRish@protonmail.com