I build public systems in series: each repo is a concrete artifact, and each series is the story that explains why those artifacts belong together. The implementation through-line is Rust: even when the artifact is a paper, corpus, game, or design lab, the durable machinery tends to become a Rust CLI, library, or verifier.
Public repos in this portfolio use ROLES,
the .roles convention for repository-local review panels.
Day job: VP of Engineering at Microsoft on Agents 365. Everything here is built outside that.
AI-first development is not about coding faster. It is about making the parts of engineering that usually get skipped actually happen: research, simulation, review, validation, critique, and rework before the wrong design hardens.
The pattern is consistent across the portfolio:
- A repo produces artifacts: code, papers, games, chapters, scorecards, maps, or packages.
- Skills and reviewers evaluate those artifacts against an explicit rubric.
- Findings become evidence.
- Repeated evidence becomes a stronger process, a new skill, or a new repo.
The series below are the public map.
TAXLANE asks a deceptively simple question: what should income-tax rates be if government spending claims have to prove themselves first?
Instead of beginning with a desired rate or headline savings number, TAXLANE examines fifteen public-purpose and accounting tracks through one evidence spine. It tests the fiscal object, source currency, net cost, overlap, beneficiary and service-continuity effects, implementation burden, and distribution boundary before admitting a spending reduction. Payment integrity stays non-additive, interest follows the debt path, dedicated old-age solvency stays separate, and revenue finances only the scoped remainder.
The completed internal analysis admits $0.000 billion in FY2026 primary spending reductions, retains a scoped $813.727 billion ordinary-income target, and recommends 21/23/33/35/43/46/48 as its preferred analytical marginal bracket schedule, with distinct behavior-contingency and severe-stress rails. Zero is a result: unsupported savings are excluded rather than guessed.
The repository includes a 21-deliverable explanation corpus, ten research papers, three presentations, an accessible six-page local website, machine-readable closure records, twelve eight-lens review rounds, and 388 passing tests.
Start with the new TAXLANE README, the final briefing bundle, or the local website source.
TAXLANE is an internal analytical recommendation, not enacted law, personal tax advice, an official score, formal proof of balance, or an authorized public deployment.
| Series | What it contains | Start here |
|---|---|---|
| AI Methodology | decision intelligence and simulated expert review | series/ai-methodology.md |
| Games Design | puzzle hunts, D&D, board games, escape rooms, and world-building simulators — with shared simulation and UX infrastructure | series/games-design.md |
| Knowledge Systems | civilization, reference, local-production, and evidence-gated fiscal corpora | series/knowledge-systems.md |
| Design Labs | visualization, writing, and music scoring labs | series/design-labs.md |
| Election Systems | redistricting, district plans, count evidence, and audit packages | series/election-systems.md |
| Standards & Protocols | portable conventions, identifiers, context-transfer formats, and review-panel specs | series/standards-protocols.md |
| Tools & Infrastructure | markdown, context selection and algebra, harness replay, fetch/cache, graph kernels, and partitioning | series/tools-infrastructure.md |
| Applied Systems | evidence-gated public-system design: highways, rail, grid, ports, water, aviation, connectivity, and applied analytics | series/applied-systems.md |
The MD family is a four-stage Markdown toolchain. Each repository owns one boundary, so the stack is easy to adopt one layer at a time:
Markdown → MDPATH → MDCROP → MDLOOM → MDPORT
address select build transfer
| Repo | Responsibility |
|---|---|
| MDPATH | Give headings, figures, tables, and sections stable md:// addresses. |
| MDCROP | Index, graph, filter, and select bounded Markdown corpus context. |
| MDLOOM | Validate, compile, render, and publish Markdown into human-facing formats. |
| MDPORT | Carry compact mdport.v1 document and corpus-slice records between tools. |
These five repos show the portfolio's artifact-to-context through-line: TAXLANE is the new evidence-to-rate knowledge-system highlight; ROUTE is the public-systems flagship; MDCROP selects relevant source context; LATTICE closes and explains that context; FLETCHER records and replays how an AI harness used it.
| Repo | What it demonstrates |
|---|---|
| TAXLANE | Fifteen evidence-gated fiscal tracks, disciplined zero admission, one accounting spine, and an adaptive analytical rate result. |
| ROUTE | Evidence-gated Interstate 2.0 research, simulation, maps, design doctrine, and reviewed papers. |
| MDCROP | Graph-cut context selection and provenance-aware corpus tooling. |
| LATTICE | Deterministic context closure, meet/join, budgets, frontiers, and receipts. |
| FLETCHER | Replayable AI harness events, checkpoints, and LATTICE handoffs. |
ROUTE is open for reference, review, and local adaptation; start with the open adoption guide.
The public infrastructure family now extends from highways into rail, connectivity, ports, water, and aviation. See Applied Systems for the full map and evidence boundaries.
The repos are not isolated experiments. They form layers:
- Methodology repos define how decisions are researched and reviewed.
- Artifact repos produce serious work in games, knowledge, civic systems, and design.
- Infrastructure repos make those artifacts reproducible: stable markdown addresses, compact context-transfer formats, repo-local review panels, corpus compilers, context packs, context algebra, replayable harness events, finished shared fetch/cache substrate for ROUTE, BISECT, and ICELINES, graph kernels, plan packages, audit ledgers, and shared game simulation and UX engines (MUDDLE, RALLY, COURT, RACKET).
That is the through-line: build the artifact, build the evaluator, then promote the reusable Rust substrate when more than one system needs it.
- Email: giodl73@gmail.com
- GitHub: @giodl73-repo