Current value · September 2026
I curate capabilities by connecting human judgment, AI, formal methods, and custom tools. I choose the question, bring together the right tools, and check what the results support.
Engineer, philosopher, and futurist. My work connects research, systems design, and intelligence augmentation: expanding what I can understand, build, and verify.
The value in the combination
I bring together the models, tools, and checks a problem needs. My projects show some of these combinations in use. The same capabilities can help me explore new questions, learn, create, and make better decisions.
I decide which question is worth asking, how to represent the problem, which tool can check the claim, and when there is enough evidence to proceed.
Where I can contribute
Here are some ways I put these capabilities to use, with examples you can inspect.
Research & discovery
Develop competing hypotheses, useful abstractions, predictions, and counterexamples with LLM assistance. Distinguish a checked mathematical result from provisional support for a scientific model.
Useful output: a precise research question, a bounded experiment or formal model, and a record of what survived testing.
Systems & verification
Design explicit state transitions, policy boundaries, and independent checks. Use Lean, Z3, Tau, tests, and replay according to what the particular claim requires.
Useful output: a scoped implementation or review with reproducible checks, failure cases, assumptions, and remaining obligations.
Research infrastructure
Connect models and symbolic tools to durable research memory. My custom Research Kernel Protocol MCP preserves questions, evidence, dependencies, and refutation attempts so later work can build on the record.
Useful output: a workflow and research record that can be inspected, continued, and adapted to another problem.
Communication & decision support
Translate difficult concepts into explanations, diagrams, interactive examples, and maps connecting claims to evidence. Help a reader understand the result, question its assumptions, and decide what to do next.
Useful output: a technical explanation or decision brief with an understandable model and evidence you can inspect.
My contribution
I guide the workflow today: choose the problem, connect ideas across domains, decide which capabilities to assemble, and review the implications of the result. LLMs help propose, predict, analyze, recombine, and implement.
Deterministic tools establish specified formal properties. Empirical observations and demanding tests provide evidence about scientific models. I keep those forms of evidence distinct when deciding what can be claimed.
The practical value is a clearer decision, an artifact that can be checked, and a capability that can be used again.
Connect a worthwhile objective to a question that can be investigated.
Select and integrate the expertise, models, tools, and representations that fit.
Assess what a proof, test, prediction, or counterexample actually establishes.
Preserve useful results and failed approaches so the next investigation starts with more knowledge.
How the value can grow
A useful model, checker, or research tool can help with many future questions. Each improvement gives me something to build on next time.
I guide research using LLMs and symbolic tools. My public repositories, proofs, software, tutorials, and diagrams show that process in use.
Transfer more of my workflow to coordinated agents through prompting, tools, and training. The constitutional-training study and full swarm transfer remain research goals. Their gains need to be demonstrated.
I would measure successful scaling by independently accepted outputs, useful task completion, defect rates, review effort, and compute cost. I judge progress by the usefulness and reliability of the results.
Cost model · Updated September 6, 2026
A refreshed version of the earlier code-volume model, covering MPRD, ZenoDEX, and Formal Methods Philosophy. Other portfolio projects and unpublished work are outside this estimate.
Legacy pooled model · USD
$52.1M–$152.5M
Illustrative conventional rebuild-cost scenarios. These are model outputs, not a quote, measured savings, or my market value. The range is not a confidence interval.
The latest public commits contain 909,096 selected nonblank lines across 4,175 files, up 14.0% from 797,416 in June. The count includes comments, tests, proofs, and examples within selected source files; it is a size proxy rather than physical source lines of code or verified authorship.
Scroll the table horizontally to see each estimate.
| Repository / pinned commit | Nonblank lines | Cost scenarios |
|---|---|---|
| MPRDf1a7e9d02e36 | 149,110 | $7.8M–$20.1M |
| ZenoDEXee9f81a3164b | 702,843 | $39.8M–$114.3M |
| Formal Methods Philosophyebe66495c746 | 57,143 | $2.9M–$6.9M |
| Sum of separate estimates | 909,096 | $50.5M–$141.3M |
The headline retains the earlier method of pooling all three repositories before applying the formula. That produces a larger number than adding three separate estimates because effort scales nonlinearly. The original June pooled result was $44.5M–$129.7M under the older wage inputs.
Let K be selected nonblank lines divided by 1,000. The retained Basic COCOMO equations produce person-months: 2.4 × K1.05 for the lower scenario and 3.0 × K1.12 for the higher scenario. Cost equals person-months ÷ 12 × annual wage × labor multiplier. Formula reference.
The BLS May 2025 wage data, checked September 6, 2026, supplies $135,980 at the median and $214,670 at the 90th-percentile threshold. The respective labor multipliers remain 1.50 and 1.38, giving assumed annual labor costs of $203,970 and $296,245. These multipliers are assumptions, not measured employer costs. No inflation forecast is added.
The smaller-size cases are arbitrary sensitivity checks. They are not measured AI productivity discounts or estimates of how much work a rebuild would actually require.
These older equations are extrapolated from a nonblank-line proxy and have not been calibrated for my repositories, AI-assisted development, formal proofs, or research artifacts. They do not establish equivalent human engineering years, production readiness, or the cost of reproducing the same functionality with a different design.
The collector retains the June source extensions and path exclusions for comparability. Known vendor, generated, data, internal, and specified derived-runtime paths are excluded; path filters cannot establish that every third-party or generated file is absent. Exact duplicate files remain in this comparison and are reported in the data. There is no semantic deduplication or authorship audit. Tests and proofs are counted within the footprint, without adding a second labor charge.
External audits, legal fees, deployment, and ongoing operations are not separately priced. A useful project quote would require a defined scope, acceptance criteria, a reuse plan, and measured delivery rates.
How to assess the value
Start with the problem you need solved. Inspect a related artifact, identify my role and the tools involved, then agree on a small project and decide how to assess its results.
Published work provides evidence of particular capabilities. A proof has stated assumptions; a benchmark has a defined scope; a prototype needs further work before deployment. I make those boundaries part of the assessment.
Does the output answer the question or enable the intended task?
Can the relevant evidence be inspected or replayed, with its assumptions visible?
How much time, compute, integration work, and review did the outcome require?
Does the work leave behind knowledge, tooling, or a method that makes subsequent work easier?
Working together
A difficult research question, a system that needs clearer guarantees, an agent workflow to improve, or a technical idea that needs an understandable demonstration. Start with a defined question and an output we can evaluate.
The earlier quantitative page used a June 17, 2026 data snapshot and conventional software-effort models. Those dated calculations remain available for inspection. They describe model assumptions and artifact volume; they do not establish my current market value, realized savings, or an equivalent number of human engineering years.
