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, simulated expert review, realized-gain learning, and reprioritization | series/ai-methodology.md |
| Games Design | puzzle hunts, board-game research, verifiable classic games, sports analysis, and shared simulation and UX infrastructure | series/games-design.md |
| Knowledge Systems | civilization, reference, decomposition, 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 transportation, connectivity, public access, civic boundaries, and applied analytics | series/applied-systems.md |
The portfolio also adds STACK-AND-TRACK, a closed-loop methodology for comparing expected impact with realized gains and reprioritizing the next sufficient stack, and LEXICON, a reviewed reference for separating a concept's senses, comparing its possible decompositions, and preserving the evidence and constraints that make those distinctions useful.
OSW — Ocean States of the World is the portfolio's ocean-science atlas: 56 approximate surface provinces form a state-like base map while waters, currents, fronts, depth layers, seafloor structures, living systems, and events overlap them. The public site currently serves reviewed Atlas 07; later atlases remain separately governed previews.
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 → PROOF → 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. |
| PROOF | 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; WITNESS 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. |
| WITNESS | 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, 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.
pwsh -NoProfile -File tests\check-proof.ps1
pwsh -NoProfile -File tools\check-local-links.ps1 README.mdThe profile's local series links are accepted. The retained invalid profile is
rejected with a structured missing local link target failure.
- Email: [email protected]
- GitHub: @giodl73-repo
giodl73-repo uses separate licenses for software and content. Source code, executable scripts, tests, configuration, and ordinary software documentation are MIT-licensed (copyright Gio Della-Libera). Original non-software content is licensed CC BY-NC 4.0 (copyright Gio Della-Libera); commercial use of that content requires separate written permission. Third-party material remains under its own terms. See LICENSE for the complete notice.