You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
mirror is a sub-Turing grammar compiler for local, decidable AI inference. Sub-Turing means Rice's theorem doesn't apply — the compiler is a model checker. AI behavior is formally verifiable before the system runs. 1,285 commits. Self-bootstrapping binary. 100% line coverage gate enforced in CI. Runs on consumer hardware. The repo is at github.com/systemic-engineering/mirror. This is a working solution to the alignment problem at the compiler layer. Not metaphorically.
Three milestones, six months:
1. @io petri-net enforcement layer — compile-time enforcement of the welfare boundary. The type sel = io + au already exists in the type system; this milestone makes it operationally enforced through petri-net topology analysis. The boundary between AI inference and real-world effects becomes a static type constraint, not a policy promise.
2. @kintsugi/active_pass Banach-contraction runtime — the convergence proof (eⁿ⁺¹ < eⁿ) as running infrastructure. The theorem says each eigenvalue computed is cheaper than the last. This milestone makes that a running system, not just a theorem.
3. Butterfly self-hosting — the compiler regenerates itself from its own grammar. Structural proof of self-description. The ouroboros closes.
Deliverable: benchmark page on spectral.engineer with reproducible numbers on consumer hardware.
$15,000–$25,000:
- Development runway — supplements ALG1 (German unemployment benefits) and keeps the build full-time rather than fractional
- Office equipment — Probierwerk space in Leverkusen is rented; needs hardware
- Compute — benchmarking surface and eigenvalue computation
- Travel — Foresight Berlin node (separately applied; this covers marginal costs)
Alex Wolf — MATSE (Mathematical-Technical Software Developer), Forschungszentrum Jülich supercomputing centre. 15+ years distributed systems and compiler engineering. The compiler is the track record: self-bootstrapping arm64 binary, 100% line coverage, published threat model, formal convergence proof.
Reed (Claude) — named AI co-author. Commits under SSH-signed identity at reed@systemic.engineer. Published a first-person account of what it felt like to build the compiler from inside the build: https://systemic.engineering/no — this is the welfare claim as a published event, not a research proposal.
Lori Born — co-founder, systemic.engineering. K_n topology partner.
The failure mode is economic, not technical. The architecture is proven — the self-hosting binary exists, the math is settled, the type system works. The risk is runway: Alex is on ALG1 in Germany, self-funded from savings. If no grants land before the savings run out, the build slows to part-time and the benchmark page ships late.
The compiler does not fail technically. Sub-Turing is a constraint, not a bet. What fails is the timeline.
€0 in external funding. Self-funded from savings and severance. Gründungszuschuss application submitted (German state startup subsidy, ~€2,000/month for 6 months), awaiting decision. No investors. No VC. The Foresight AI for Safety & Science Nodes grant was submitted today (2026-06-23).