Jim Butcher, PhD MBA

MBA, University of Phoenix. PhD: honorary distinction in Martial Sciences, International University of San Diego (2013), not an earned research doctorate.

Public professional and creative portfolio

Master Portfolio and Application Reference / Combined edition 2 / 2026-10-19

I bring a commercial-finance and business-development background to practical software, research, education, and creative projects. I direct the work, use AI-assisted implementation and review, and retain human decisions over content, rights, spending, and release. The aim is useful work with evidence behind it, not a claim that every prototype is a finished product.

This public summary follows the master portfolio dated October 19, 2026. Underlying records have different dates. The R640/HP results are attributed to the technical review summarized in that document, not freshly rerun or independently accepted in this website update. Documented features, author tests, reported independent checks, and proposed work remain distinct.

MARTIAL ARTS BACKGROUND

Jim Butcher began training at Bushi Ban Clear Lake in September 2002. His background connects martial-arts practice, teaching, school leadership, strength breaking and service to the Sport Karate Museum. He reports seventh-degree recognition in 2025 through Sport Karate Museum and American Fitness and Martial Art (AFMA).

Early progression: September 2002 training start; provisional black belt in 2006; First Degree in 2007; Second Degree in 2009. AFMA Third Dan certificate: September 8, 2012. Fourth-degree correspondence: 2014-2015. Jim reports Fifth Degree in 2018 (Sport Karate Museum, AFMA, WMARA), Sixth Degree in 2021 and Seventh Degree in 2025 (Sport Karate Museum and AFMA).

2013 USA Martial Arts Hall of Fame: Breaking Master of the Year, August 10, 2013, supported by a certificate reproduction and contemporary AFMA announcement. Jim separately reports Celebrity Martial Arts Hall of Fame recognition in 2025 and 2026; ceremony and category records remain to be collected.

Teaching emphasizes confidence, discipline, achievable goals and helping children find their voice. Historical Santa Fe school leadership and museum service are not a claim of uninterrupted current tenure. The sport-karate history collaboration with Professor Gary Lee remains in review, not a published book.

Updated from the October 23, 2026 career profile, revision R3. Early dates and the 2018, 2021 and 2025 promotions are Jim-reported autobiographical information. Historical certificate reproductions and correspondence support the earlier milestones in this summary; they are not independent issuer authentication. The private dossier and its email links are not published here.

Full public background: https://jimbutcherstudio.com/martial-arts/

Business operations and public intelligence

Documented workflows / coverage and qualification limits retained

My commercial-finance background informs practical work on business assessment, receivables financing, relationship development, documentation, and communication. The professional roles and education are listed separately in the CV, with their source dates and qualifications.

Business-intelligence workflows combine public-source research, contact verification, approval history, and provider-result reconciliation. Research, a draft message, an approved send, a reply, and a customer outcome are different states. This portfolio does not offer employer-owned records, customer information, or contact lists as personal products.

A bounded public-information collection project covers procurement, company information, economic indicators, property-related public records, and publishing-platform notices. Coverage remains partial. Historical awards are not automatically current opportunities, GIS records do not establish title, and economic series need dated metadata. The contribution is repeatable source handling, not a claim of exhaustive intelligence coverage.

Learning software and everyday tools

Local prototypes and documented features / user and domain review pending

The software collection explores practical learning, teaching, and everyday tasks. The master document reviewed project documentation and selected evidence; it did not relaunch or retest every application. No download, adoption, learning-gain, or paid-customer total is claimed.

Study tools are not exam certification or authority to provide regulated services. Memory, listening, exercise, and visual-practice tools are not clinically validated treatments, diagnoses, or demonstrated intelligence improvements. Financial and health calculators retain their assumptions and are not professional advice.

Codecraft
Beginner JavaScript and Python lessons, exercises, hints, concept checks, and practical projects.

Wordwise
Spelling and grammar practice for grades 1-12; documentation describes 48 lessons and 240 original questions. Educator and learner review remain pending.

History Trail and World History Quest
Selected history topics, explanations, source links, and grade-band paths. World History Quest documents 12 selected chapters and 96 activities; neither is a complete twelve-grade curriculum.

Hola Diario and Undiscovered Country
Spanish lessons, phrases, grammar, dialogue, and story-based language practice.

Wordbridge
A small prototype exploring related spellings and meanings across English, Spanish, French, Italian, and Portuguese; not a complete language course.

Series Ready
Life-insurance and securities-study lessons, practice, and scenarios; no regulator endorsement or exam-pass claim.

Calculator Library
Documentation describes 42 calculators spanning finance, business, mathematics, everyday use, health, and conversions, with explanations and local saved results.

Pantry and Plate
Recipe and ingredient matching with household preferences. Ranking signals are not verified nutrition totals or store-price budgets.

Fieldwork and Little Dojo
Equipment-based exercise sessions and editable children's martial-arts class routines and plans.

MindTrail, MindForge, Attune, and Sightline
Memory, focus, reasoning, listening, and visual-response practice. These are practice prototypes, not clinical products.

Family math-learning prototype
Grade and course paths, story progression, and saved state. Third-party characters, music, and other assets require separate rights review before commercial use.

Games and interactive storytelling

Author-side implementation / editorial, playtest and rights gates remain

The Covenant of Seven Roads game adapts an original creative world into a local browser prototype. Documented work includes story state, character art integration, movement, save compatibility, pause and reduced-motion behavior, and a related strategy preview with queued unit orders. Author checks are not independent final acceptance or a commercial game release.

Nexus of Echoes is documented as a collection of 100 revised adventures with branching narratives. Its development record reports 1,800 route, decision, and ending paths. Those are content structures, not 1,800 products or human playtests. Intended play duration and reader quality still need human evaluation.

The Shattered Compass and language-adventure work explore narrative interfaces and saved progress. Review-only scene and artwork proposals remain proposals; this website does not activate unreleased game content, publish manuscripts, or offer downloads of rights-uncleared assets.

Books, publishing, and knowledge assets

Manuscripts and preparation / not a list of published books

The book pipeline includes The Covenant of Seven Roads, the three-book Life-Story Collection, The Infinite Sandbox, Jimmy Learns, the AI Playbook, and the Human Questions Anthology. Work includes source registers, manuscript and artwork tracking, editorial support, version comparisons, page maps, and human-review packages.

The projects retain different gates: human read-through, truth and privacy review, governing revision choices, source custody, final artwork, and rights. The AI Playbook is an assistant-authored prompt document, not a verified transcription of source photographs. The Human Questions Anthology has internal research and teaching pilots, not an approved final anthology.

Other concepts and source-recovery candidates include Psalms for Every Emotion, Defense Against the Dark Arts, The Sovereign Soul, anxiety and recovery topics, and Singers' True Stories. A named concept is not proof of a finished manuscript. The Wisdom Library and summary efforts also retain separate source identities, edition and quotation rights, review states, and permission limits.

The goal is selected, carefully reviewed releases. No publisher contract, publication date, bestseller result, sales, or royalty income is implied.

Music and creative media

Organized recording and lyric assets / release and rights review pending

The October 18 owner record describes new songbook and asset-manual editions, 205 additional distinct archived lyric texts associated with 242 recording IDs, and a manual covering 307 recording IDs. The owner reports checking all 307 original MP3 hashes and preparing exact-ID label mappings without changing the audio or its tags.

This is an organized creative asset base, not 307 commercially cleared songs or published releases. The preservation counts are source-reported production records, not a new listening, authorship, or rights audit. Cleaned listener-facing lyrics, original recordings, and proposed release metadata remain separate.

Outstanding review includes incomplete lyric sources, instrumental labels, provisional language and genre information, source and theological holds, sung wording, generation-plan terms, artwork, performer attribution, and release selection. Possible future uses include selected original songs, rights-cleared lyric collections, and music for owned creative work. No income or adopted release plan is promised.

Applied AI engineering and reliability

Bounded engineering records / not autonomous production acceptance

Guarded task and retrieval controls include frozen task packets, a task state machine, source-aware prompt compilation, audit records, review gates, and an offline retrieval-handoff validator. The original implementation status records 17 passing Node tests but an unsuccessful live model-endpoint probe. Passing software tests did not establish an accepted live retrieval-and-generation service.

An evaluator reassessed seven saved successful API requests and found two quality failures, with five answers still needing semantic or completeness review. It did not run a new model. Transport success, format, completeness, and actual correctness were deliberately kept separate.

Checkpoint and recovery work records 165 author checks: 162 passed, no failures or errors, and three Windows-privilege skips. Those tests do not establish measured token, cost, or energy savings. A separate seven-component synthetic tool review accepted only three for limited helper or advisory use; four were not accepted for live reliance.

Source-preserving document extraction and operational-receipt tools extend the same approach. Controller-side checks do not certify deployment on another computer, a dated receipt does not prove continuous security coverage, and an unfinished context-reset check is not a reliability pass.

R640 and HP local-AI work

Source-reported historical pilots / no current-operation claim

The combined document describes Jim's direction of an agent-assisted Linux research environment spanning three Dell PowerEdge R640 servers and an HP worker pool. The public scope is project design, bounded work, resource limits, source preservation, and recoverable procedures, not machine-access details or a certified combined resource pool.

A local retrieval and coding demonstration used llama.cpp, Qwen models, and Qdrant. The technical report describes a small approved-code fixture, retrieval of the intended source, and a proposed one-file correction. It reports eight protected post-fix tests passing after seven failures in the intentionally defective baseline, plus a separate 31-test package suite. This is a bounded pilot, not a production-repository benchmark or cold-reboot acceptance.

In a separate 12-task synthetic comparison, the raw reported scores were GPT-OSS 9/12 and Qwen3-Coder 6/12. A selected GPT-OSS profile using a fixed deployment prompt and explicit JSON mode later passed 12/12 with a reported median response of 20.98 seconds on four CPU threads. Different prompts and evidence sets must not be merged into a general model ranking or frontier-model superiority claim.

Synthetic LoRA experiments checked adapter updates, a frozen base, checkpoint integrity, and exact optimizer/RNG resume. They did not train a production model. A separate historical lyrics-training ledger ended in failed-test and quality-false states. Production LoRA quality and completed QLoRA training are not established.

A cooperative invoice-factoring experiment retained negative results: a solo model scored 2/9 checks, a first multi-model handoff 5/9, and a compact retry 2/9. Model-to-model review did not reliably fix numerical and eligibility errors. Deterministic checks remain important before drafting.

Library work retained a successful small canary, an input-cap stop, and a report of 364 original files rehashed without drift, while acceptance and quality stayed false. The illustration workflow reported 157 image/receipt pairs technically verified against a 400-asset target, leaving 243 uncreated. Technical file identity is not creative approval. These records do not establish a finished library, fully autonomous production, current worker utilization, or a fully protected fleet.

Applied computational studies

Reported descriptive and finite results / assumptions and negative findings retained

Travel and disability-related difficulty
The technical report describes analysis of preserved 2022 NextGen National Household Travel Survey data. The weighted annualized contrast was -278.5636 trips per person, approximate standard error 34.9508, and approximate 95% interval -347.0764 to -210.0508. This is a descriptive contrast with approximate uncertainty, not a causal effect, official NHTS estimate, or full complex-survey inference. A separate operational refresh failed qualification; the result does not certify unattended operation.

Finite Goldbach enumeration
Exact methods reportedly agreed on minimal prime-pair witnesses for 49,999 even integers through 100,000. Complete pair sets for 9,999 integers through 20,000 totaled 1,437,831 unordered partitions, with 12 reported final integration checks passing. These finite computations and exploratory density observations are not a proof of Goldbach's conjecture, a mathematical law, a world record, or a novelty claim.

Food-loss scenarios
The reported finite design covered 114 USDA food forms and 465 scenarios, producing 53,010 result rows. Its included 2019 model estimated 250.7901 pounds of modeled loss per person per year; the ten highest forms accounted for 65.456%. Multipliers of 0.8, 1, and 1.2 are illustrative assumptions, not confidence limits. Cooking and moisture losses are included; the total is not entirely avoidable household waste or a savings promise.

Household energy
The source reports 16 principal descriptive contrasts, 1,040 numerical comparisons with maximum difference zero, and 144 additional exact comparisons. Additivity failed in 166 cases with the cause unresolved, and an income-burden result was withheld because its time basis was uncertain. No settled causal or affordability conclusion is claimed.

HELIOS numerical prototype
A retained reproduction report records 28 bundle artifacts and zero hash mismatches, limited to numerical execution and copied-file integrity. Physical hardware performance, commercial advantage, customer validation, and patent novelty remain unestablished.

Research agenda and unfinished work

Planning inventory / no scientific execution for the 69 AI topics

The latest combined inventory contains 69 AI topics: 64 drafts, four preparation-complete but not admitted entries, and one preparation-blocked entry. All 69 have no scientific execution. MM-16, MM-44, MM-49, and MM-50 are prepared but not admitted or tested; MM-01 is blocked.

A separate 52-entry CPU experiment catalog and a 12-project planning portfolio have overlapping themes and different identities. Their counts must not be added to the 69-topic inventory as distinct or completed experiments. The finite applied studies above do not imply that these proposed AI studies have run.

Unfinished work includes resource accounting, target-host qualification, accepted long-context library evaluation, production adapter quality, educator and user review, manuscript read-through, rights decisions, and unmeasured commercial demand. This summary records gates; it does not launch new work or override an existing hold.

What I am working toward

Goals and potential uses / not validated demand or recurring income

The objective is to turn a broad working collection into a smaller set of credible outcomes for defined audiences. For software, that means dependable behavior and genuine user value. For writing and music, it means editorial quality, rights, and deliberate release. For research, it means bounded questions, preserved inputs, repeatable methods, and honest negative findings.

Appropriate support could help with a capped evaluation, independent review, accessibility and user testing, rights-cleared preparation, or a reproducible demonstration. These are discussion categories, not a submitted budget, a funding award, or a commitment to deliver new work.

Potential paid work or creative releases require clear ownership, an appropriate scope, actual access, and suitable review. Employer material, third-party intellectual property, client information, and private records do not become part of a personal offer merely because they appear in a working portfolio. No product adoption, earnings, recurring revenue, or funding outcome is implied.

Complete 69-topic AI planning inventory

All topics have no scientific execution.

MM-01 Quality-constrained quantization. Preparation blocked and not admitted or tested.

MM-02 Layer-sensitive mixed precision. Draft and not tested.

MM-03 Calibration under domain shift. Draft and not tested.

MM-04 Structured pruning and layer removal. Draft and not tested.

MM-05 Sparse CPU kernel break-even. Draft and not tested.

MM-06 Native ternary models on CPU. Draft and not tested.

MM-07 MoE versus dense CPU economics. Draft and not tested.

MM-08 Expert hot-set caching and prefetch. Draft and not tested.

MM-09 DRAM and Optane execution. Draft and not tested.

MM-10 Memory mapping and layer streaming. Draft and not tested.

MM-11 KV-cache quantization. Draft and not tested.

MM-12 Evidence-aware KV retention. Draft and not tested.

MM-13 Persistent prefix and KV libraries. Draft and not tested.

MM-14 CPU paged-KV allocation. Draft and not tested.

MM-15 Context selection and summary policies. Draft and not tested.

MM-16 NUMA and CPU placement. Preparation complete but not admitted or tested.

MM-17 Instruction sets and packed CPU kernels. Draft and not tested.

MM-18 Separate prefill and decode budgets. Draft and not tested.

MM-19 Deadline-aware batching. Draft and not tested.

MM-20 Activation workspace reuse. Draft and not tested.

MM-21 Bandwidth-guided autotuning. Draft and not tested.

MM-22 Energy-aware scheduling. Draft and not tested.

MM-23 Exact speculative decoding. Draft and not tested.

MM-24 Prompt-lookup and n-gram drafting. Draft and not tested.

MM-25 Self-speculation and early exits. Draft and not tested.

MM-26 Small-to-large local model routing. Draft and not tested.

MM-27 Adaptive reasoning stops. Draft and not tested.

MM-28 Verified candidate search. Draft and not tested.

MM-29 Grammar-constrained generation. Draft and not tested.

MM-30 Verifier-filtered local distillation. Draft and not tested.

MM-31 Small-model CPU LoRA. Draft and not tested.

MM-32 Memory updates versus weight updates. Draft and not tested.

MM-33 Tiny recurrent and state-space memory. Draft and not tested.

MM-34 Independent task parallelism. Draft and not tested.

MM-35 CPU layer and pipeline sharding. Draft and not tested.

MM-36 Tensor-parallel communication feasibility. Draft and not tested.

MM-37 Specialist committees and verification. Draft and not tested.

MM-38 CPU hybrid retrieval and reranking. Draft and not tested.

MM-39 Temporal source-linked evidence graphs. Draft and not tested.

MM-40 Hierarchical episodic and procedural memory. Draft and not tested.

MM-41 Incremental indexing and deletion. Draft and not tested.

MM-42 Executable database queries. Draft and not tested.

MM-43 Calibrated abstention and confidence. Draft and not tested.

MM-44 Verifier-backed local task solving. Preparation complete but not admitted or tested.

MM-45 Constrained program synthesis. Draft and not tested.

MM-46 Reproducible data-analysis agents. Draft and not tested.

MM-47 Cost-aware experiment selection. Draft and not tested.

MM-48 Bounded failure triage and code repair. Draft and not tested.

MM-49 Checkpoint recovery and result commitment. Preparation complete but not admitted or tested.

MM-50 Protected evaluation and research integrity. Preparation complete but not admitted or tested.

MM-51 Learning from negative experiments. Draft and not tested.

MM-52 Unattended endurance and drift. Draft and not tested.

MM-53 PC hybrid-core scheduling. Draft and not tested.

MM-54 CPU runtime and build comparisons. Draft and not tested.

MM-55 Distributed prefill and KV migration. Draft and not tested.

MM-56 CPU model and adapter merging. Draft and not tested.

MM-57 Continual learning and forgetting. Draft and not tested.

MM-58 CPU preference optimization. Draft and not tested.

MM-59 Formal proof verification. Draft and not tested.

MM-60 CPU document extraction and OCR. Draft and not tested.

MM-61 Embedding and index compression. Draft and not tested.

MM-62 Low-rank weight compression. Draft and not tested.

MM-63 Dense-to-sparse MoE conversion. Draft and not tested.

MM-64 Sleep-time and anticipatory computation. Draft and not tested.

MM-65 Statistical power and adaptive evaluation. Draft and not tested.

MM-66 Programmatic tasks and contamination controls. Draft and not tested.

MM-67 CPU-scale paper reproduction. Draft and not tested.

MM-68 Input-adaptive precision. Draft and not tested.

MM-69 Funding and research evidence packages. Draft and not tested.

Public summary only. Private source appendix, internal budgets, employer activity details, raw logs, source-document links, and machine access are excluded.