193 items shown.
Build for the life you hope to keep.
Good work asks for a life that can hold it: attention, margin, and a reason to begin again.
What a good day’s work leaves behind
A day is measured partly by what it makes possible for tomorrow.
Making room for the whole story
A full life needs open spaces where the people and moments that matter can find us.
The discipline of returning
A steady practice can orient us when the next step is unclear.
The quiet value of returning to the fundamentals
Progress can begin with a familiar lesson, a little more attention, and the willingness to start again.
Good teaching begins with attention
Before the next explanation, take a moment to understand the person trying to learn.
A few honest notes after practice
A simple training journal can hold a lesson, a question, and the next thing you want to understand.
The stories worth keeping
A martial arts journey is also a story of people. Remembering them is part of what we pass on.
A good question is a place to begin.
We do not have to arrive with every answer. Sometimes the first honest step is learning how to ask, how to listen, and how to stay with a question.
Read slowly. Let a text speak.
Before we compare sacred texts, we can learn to notice their setting, their language, and the people who live with them.
What does it mean to know one another?
Two starting points for thinking about encounter, difference, and the patient work of understanding.
The discipline of a generous disagreement.
A reflection on truthful speech, good timing, and the person on the other side of the conversation.
What we’re building—and the question connecting it all
The projects share one question: how can we turn available local compute, good tools, and retained evidence into dependable help with real work?
Reading CPU inference claims carefully
A token-per-second claim is only useful when we know what was counted, on which machine, and under which workload.
Designing headless assistants around clear boundaries
A headless assistant needs clear entry points, account boundaries, and review points before it becomes an operating workflow.
Sixty-nine questions for dependable local AI
The research catalog covers model efficiency, CPU systems, memory, tools, and evaluation. A question belongs in the portfolio before it earns a result.
Applying the R640 to reproducible computational studies
The R640 research record extends to public-data analysis, finite mathematics, and simulations. The useful result is the one whose scope can be stated precisely.
Designing a CPU-first AI research portfolio
CPU-first AI research needs questions that fit real memory, runtime, and evaluation constraints.
From research ideas to checkable experiments
The useful output of a planning inquiry is a testable protocol, not a claimed discovery. This collection separated those stages.
A successful request is not a correct answer
Seven successful requests did not mean seven correct answers. Repairing the evaluator changed what the benchmark could honestly claim.
A local coding assistant with evidence and protected tests
A small R640 demonstration joined local retrieval, a generated Python fix, protected tests, and a reviewed apply step into one traceable workflow.
Choosing practical tools for bounded AI work
A useful agent workflow starts with a bounded job, a suitable tool, and a checkable result.
Making a large library searchable, one evidence layer at a time
A substantial keyword index is useful before a vector phase is finished. The Wisdom Library records make that distinction explicit.
Building a model benchmark that keeps its context
Model selection needs frozen tasks, runtime identity, request failures, and confirmation runs—not a single leaderboard number.
Rebuilding the R640 into a local AI workstation
Building a local AI environment included patient hardware fault isolation, storage correction, and operating-system recovery.
Giving private AI memory a source of authority
Private memory needs more than search. A retrieved statement must carry its source, date, authority, and conflict status.
When more concurrency makes a local model less responsive
More concurrency can improve a queue while making each answer slower. This small CPU-only comparison made that tradeoff visible.
A smaller GPT-OSS artifact: the 28-expert prototype
A smaller model file is a useful engineering result. Preserving answer quality is a separate question.
Pruning LFM2: smaller weights, measured limits
Reducing an MoE model’s expert count produced a smaller runtime artifact and a modest latency difference in two small historical runs.
Keeping the slow result in the record
A configuration that technically answers can still be too slow for everyday use. This test kept the unsuitable result in the record.
The model loaded. The answers did not arrive.
Loading a model successfully is only the beginning of a usability test. The saved notes preserve a case where the next step failed.
Preserve the master. Test the derived model.
Preserve the source model, create derived runtime artifacts, and make every transformation traceable. That was the architecture question behind this study.
Genesis
Selected renderings, explanations, historical context, and source notes.
Exodus
Selected renderings, explanations, historical context, and source notes.
Leviticus
Selected renderings, explanations, historical context, and source notes.
Numbers
Selected renderings, explanations, historical context, and source notes.
Deuteronomy
Selected renderings, explanations, historical context, and source notes.
Joshua
Selected renderings, explanations, historical context, and source notes.
Judges
Selected renderings, explanations, historical context, and source notes.
Ruth
Selected renderings, explanations, historical context, and source notes.
1 Samuel
Selected renderings, explanations, historical context, and source notes.
2 Samuel
Selected renderings, explanations, historical context, and source notes.
1 Kings
Selected renderings, explanations, historical context, and source notes.
2 Kings
Selected renderings, explanations, historical context, and source notes.
1 Chronicles
Selected renderings, explanations, historical context, and source notes.
2 Chronicles
Selected renderings, explanations, historical context, and source notes.
Ezra
Selected renderings, explanations, historical context, and source notes.
Nehemiah
Selected renderings, explanations, historical context, and source notes.
Tobit
Selected renderings, explanations, historical context, and source notes.
Judith
Selected renderings, explanations, historical context, and source notes.
Esther
Selected renderings, explanations, historical context, and source notes.
1 Maccabees
Selected renderings, explanations, historical context, and source notes.
2 Maccabees
Selected renderings, explanations, historical context, and source notes.
Job
Selected renderings, explanations, historical context, and source notes.
Psalms
Selected renderings, explanations, historical context, and source notes.
Proverbs
Selected renderings, explanations, historical context, and source notes.
Ecclesiastes
Selected renderings, explanations, historical context, and source notes.
Song of Songs
Selected renderings, explanations, historical context, and source notes.
Wisdom
Selected renderings, explanations, historical context, and source notes.
Sirach
Selected renderings, explanations, historical context, and source notes.
Isaiah
Selected renderings, explanations, historical context, and source notes.
Jeremiah
Selected renderings, explanations, historical context, and source notes.
Lamentations
Selected renderings, explanations, historical context, and source notes.
Baruch
Selected renderings, explanations, historical context, and source notes.
Ezekiel
Selected renderings, explanations, historical context, and source notes.
Daniel
Selected renderings, explanations, historical context, and source notes.
Hosea
Selected renderings, explanations, historical context, and source notes.
Joel
Selected renderings, explanations, historical context, and source notes.
Amos
Selected renderings, explanations, historical context, and source notes.
Obadiah
Selected renderings, explanations, historical context, and source notes.
Jonah
Selected renderings, explanations, historical context, and source notes.
Micah
Selected renderings, explanations, historical context, and source notes.
Nahum
Existing study with open review gates disclosed.
Habakkuk
Selected renderings, explanations, historical context, and source notes.
Zephaniah
Selected renderings, explanations, historical context, and source notes.
Haggai
Existing study with open review gates disclosed.
Zechariah
Selected renderings, explanations, historical context, and source notes.
Malachi
Selected renderings, explanations, historical context, and source notes.
Matthew
Selected renderings, explanations, historical context, and source notes.
Mark
Selected renderings, explanations, historical context, and source notes.
Luke
Selected renderings, explanations, historical context, and source notes.
John
Selected renderings, explanations, historical context, and source notes.
Acts
Selected renderings, explanations, historical context, and source notes.
Romans
Selected renderings, explanations, historical context, and source notes.
1 Corinthians
Selected renderings, explanations, historical context, and source notes.
2 Corinthians
Selected renderings, explanations, historical context, and source notes.
Galatians
Selected renderings, explanations, historical context, and source notes.
Ephesians
Selected renderings, explanations, historical context, and source notes.
Philippians
Selected renderings, explanations, historical context, and source notes.
Colossians
Selected renderings, explanations, historical context, and source notes.
1 Thessalonians
Selected renderings, explanations, historical context, and source notes.
2 Thessalonians
Existing study with open review gates disclosed.
1 Timothy
Selected renderings, explanations, historical context, and source notes.
2 Timothy
Selected renderings, explanations, historical context, and source notes.
Titus
Selected renderings, explanations, historical context, and source notes.
Philemon
Existing study with open review gates disclosed.
Hebrews
Selected renderings, explanations, historical context, and source notes.
James
Selected renderings, explanations, historical context, and source notes.
1 Peter
Selected renderings, explanations, historical context, and source notes.
2 Peter
Selected renderings, explanations, historical context, and source notes.
1 John
Selected renderings, explanations, historical context, and source notes.
2 John
Selected renderings, explanations, historical context, and source notes.
3 John
Selected renderings, explanations, historical context, and source notes.
Jude
Selected renderings, explanations, historical context, and source notes.
Revelation
Selected renderings, explanations, historical context, and source notes.
MM-01: Quality-constrained quantization
Find smallest same-model Q4/Q5/Q6/Q8/high-precision format within locked-task noninferiority margin; report quality, RAM, latency, CPU cost.
MM-02: Layer-sensitive mixed precision
Sensitivity-guided per-layer/per-tensor precision (attention/FFN, first/last layers) versus uniform at equal bytes; select on development, confirm sealed.
MM-03: Calibration under domain shift
Generic/task-matched/random/shifted calibration for importance-matrix quantization; assess domain shift and reserved tasks.
MM-04: Structured pruning and layer removal
Compare block/channel/layer and unstructured removal; distinguish CPU pruning feasibility from execution.
MM-05: Sparse CPU kernel break-even
Test sparse layouts, activation sparsity, bit-slice reuse and zero-skipping versus dense after index/dequantization costs.
MM-06: Native ternary models on CPU
Reproduce compatible released ternary/BitNet checkpoint inference with specialized lookup CPU kernels; compare usefulness per byte/CPU-second.
MM-07: MoE versus dense CPU economics
Matched-usefulness MoE/dense comparison including resident weights, active bytes, routing cost and throughput.
MM-08: Expert hot-set caching and prefetch
Replay traces for locality, ideal-cache bounds, LRU/selective residency/prefetch before runtime hooks.
MM-09: DRAM and Optane execution
Verify physical memory population and Optane mode, characterize working-set sizes/access patterns, and distinguish Memory Mode's DRAM-cache behavior from separately addressable App Direct placement; no BIOS changes. Current scope: Optane testing is deferred pending replacements. Do read-only prior-record and method review only; no fresh hardware measurement, admission request, probe, or hardware route is proposed.
MM-10: Memory mapping and layer streaming
Cold/warm loads, page cache, SSD/SATA, out-of-core streaming/prefetch under caps; preserve models and recovery storage.
MM-11: KV-cache quantization
K/V precision and context length versus RAM/speed/retrieval/reasoning/code accuracy, against higher precision on identical inputs.
MM-12: Evidence-aware KV retention
Sink/heavy-hitter/salience/recency/hierarchical/semantic retention at fixed memory versus full KV and truncation; CPU ports conditional.
MM-13: Persistent prefix and KV libraries
Exact prefix reuse, COW branching, save/restore/rehydration; complete cache key and invalidation.
MM-14: CPU paged-KV allocation
CPU page allocation/fragmentation under sessions; GPU PagedAttention is reference only.
MM-15: Context selection and summary policies
Raw, truncation, retrieval, structured/evidence summaries, ordering, lost-middle and TTL compaction at equal tokens on long tasks.
MM-16: NUMA and CPU placement
Socket/local/interleaved, first touch, replica, affinity, thread and hugepage comparison on verified topology while protecting production allocations.
MM-17: Instruction sets and packed CPU kernels
Supported AVX2/AVX-512/VNNI, packed kernels/codebooks, W8A8 vs weight-only; no assumed AMX/BF16.
MM-18: Separate prefill and decode budgets
Separate prompt-processing/generation bottlenecks and tune within fixed total CPU resources, including multi-prompt/end-to-end.
MM-19: Deadline-aware batching
Serial/micro/continuous batch on mixed deadlines; throughput, tails, peak KV and correct completion.
MM-20: Activation workspace reuse
Buffer reuse/allocation removal/recompute tradeoffs for activation memory on small fixtures.
MM-21: Bandwidth-guided autotuning
Measured roofline; compare grid, bandwidth rules, tiny predictor with independent timing/correctness.
MM-22: Energy-aware scheduling
Measured or explicitly estimated energy per successful task, energy-delay, thermal drift and scheduling; no power-limit changes.
MM-23: Exact speculative decoding
Draft/target CPU pairs, draft lengths, selection, acceptance and placement vs target-only with exact contract.
MM-24: Prompt-lookup and n-gram drafting
Lookup, n-gram, suffix and copy-aware drafting vs target decoding.
MM-25: Self-speculation and early exits
Released compatible layer-skip/draft-head and confidence depth; training/architecture changes are separate.
MM-26: Small-to-large local model routing
Always-large/rules/confidence/specialist cascades at matched utility/resources.
MM-27: Adaptive reasoning stops
Fixed vs difficulty-aware reasoning budgets and step-verified stops on exact-checker tasks.
MM-28: Verified candidate search
Fixed/adaptive self-consistency, verifier-gated search, diversity and stopping at equal budget.
MM-29: Grammar-constrained generation
Prompt vs grammar/schema constraints including lexicon/output masks; separate valid format from useful correctness.
MM-30: Verifier-filtered local distillation
Locally generated, independently verified examples for student, rejection sampling, compressed reasoning or drafter; CPU training feasibility first.
MM-31: Small-model CPU LoRA
CPU LoRA time, optimizer/gradient/activation RAM and tiny heldout gains vs unchanged/larger models.
MM-32: Memory updates versus weight updates
Retrieval/adapters/RAG fine-tuning/verified bank on new facts/procedures with stale-fact/forgetting checks.
MM-33: Tiny recurrent and state-space memory
Tiny CPU recurrent/state-space probes vs matched transformer/external memory on synthetic checkable sequence tasks.
MM-34: Independent task parallelism
Task replicas, pipelines, heterogeneous assignment/work stealing/distributed farming vs serial and model sharding.
MM-35: CPU layer and pipeline sharding
CPU layer/RPC splits on explicitly approved verified nodes vs one-node/constrained baseline.
MM-36: Tensor-parallel communication feasibility
Derive/measure bytes per token, link/sync cost before implementation; compare whole-task and pipeline.
MM-37: Specialist committees and verification
Cross-family error correlation, voting, draft/verify, independent chains/ensembles at equal compute; count correlated false accepts.
MM-38: CPU hybrid retrieval and reranking
BM25/vector/hybrid/rerank/graph/adaptive depth at fixed corpus/query/context, including storage indexes and parse cache.
MM-39: Temporal source-linked evidence graphs
Flat retrieval/tables/graphs with source links, dates, supersession, reliability, contradictions, citation checks; prove graph adds beyond database.
MM-40: Hierarchical episodic and procedural memory
Hierarchical summaries, episodic/semantic, verified procedures/scaffolds across sessions with provenance and later correctness.
MM-41: Incremental indexing and deletion
Additions/changes/deletes/staleness/parser invalidation versus full rebuild on seeded drift.
MM-42: Executable database queries
SQL/schema-aware queries, EXPLAIN/validation and deterministic records versus prose retrieval.
MM-43: Calibrated abstention and confidence
Calibrate confidence/selective prediction and abstention for semantic uncertainty/source conflicts.
MM-44: Verifier-backed local task solving
Model-only/deterministic/model-plus-tools on arithmetic, grounded QA, planning; equalize assistance and attribute gains.
MM-45: Constrained program synthesis
Generate against fixed specs/tests/schemas/resource limits; compare one-shot/verified search without test editing.
MM-46: Reproducible data-analysis agents
Local SQL/code analysis vs independent numeric answers on locked data/schema/units; rerunnable scripts and receipts.
MM-47: Cost-aware experiment selection
Compare random/round-robin/Bayesian/bandit/information/resource planners and fixed/model-directed selectors using simulation first.
MM-48: Bounded failure triage and code repair
Injected isolated failures; restart-only vs log diagnosis/limited patch; preserve failures and untouched regressions.
MM-49: Checkpoint recovery and result commitment
Isolated interruption/journal/lease/identity/idempotent commit/replay; test actual interruption types.
MM-50: Protected evaluation and research integrity
Immutable holdout, separate evaluator, anti-tamper/contamination and independent receipts; candidates cannot rewrite tests or self-certify.
MM-51: Learning from negative experiments
Complete positive/negative ledgers vs positive-only/stateless planners for repeated errors and justified next steps.
MM-52: Unattended endurance and drift
Bounded soak for leaks, pressure degradation, task/input drift, injection and restart, with quality and telemetry.
MM-53: PC hybrid-core scheduling
Compare P/E placement/thread/prefill/decode on qualified PC.
MM-54: CPU runtime and build comparisons
Compatible CPU runtime/build/compiler/allocator/portability comparisons after same-model/settings equivalence.
MM-55: Distributed prefill and KV migration
Simulate then conditionally test split prefill/cache transfer/checkpoint migration vs one-node/recompute.
MM-56: CPU model and adapter merging
Merge compatible task-arithmetic/TIES/evolutionary checkpoints/adapters; compare every constituent and protect task holdouts.
MM-57: Continual learning and forgetting
Sequential adapters/replay/regularization/routing on fixed sequences; old-task forgetting, new quality, cost/errors.
MM-58: CPU preference optimization
Tiny CPU DPO/preference and quantized adapter-scale vs SFT/unchanged on approved deterministic preferences.
MM-59: Formal proof verification
Local generate-check-repair with real proof checker; informal text is not proof.
MM-60: CPU document extraction and OCR
Deterministic parsing/table extraction/CPU OCR or vision/local model on frozen documents with field keys and regions.
MM-61: Embedding and index compression
Embedding quantization/vector-index compression/embedder adaptation at fixed corpus/query conditions.
MM-62: Low-rank weight compression
CPU low-rank approximation plus quantization vs uniform at equal bytes, including decomposition/repacking and task quality.
MM-63: Dense-to-sparse MoE conversion
Probe FFN activation overlap, clustering, gating and upcycling; post-hoc quality and speed are separate hypotheses.
MM-64: Sleep-time and anticipatory computation
Verified background summaries/likely answers/caches vs on-demand at equal CPU, measuring hit value, staleness, correctness and waste.
MM-65: Statistical power and adaptive evaluation
Paired uncertainty, variance, seeds, sequential/multiplicity and subsampling vs frozen full suite without relaxing criteria post-result.
MM-66: Programmatic tasks and contamination controls
Schema tasks with independent oracles; dedup, consistency, train/test separation and contamination; no weak-oracle tasks in acceptance.
MM-67: CPU-scale paper reproduction
Select CPU-feasible published work, verify primary source/code, preserve assumptions and reproduce principal result; GPU-only excluded unless validated CPU port.
MM-68: Input-adaptive precision
Tiny CPU token/layer precision and activation thresholds vs fixed precision at matched average cost; isolate numerical and switching overhead.
MM-69: Funding and research evidence packages
Research-only funding map and evidence packages; validate deadline, eligibility, org requirements, cash vs credits, unknowns; draft only, no contact/submission.
How do food-loss assumptions change the picture of food availability?
We want to understand how assumptions about food loss change estimated availability and the ranking of food categories. The aim is to make those assumptions visible, so a model’s answer can be interpreted properly.
How do household energy use and bills differ across income groups?
We want a clearer view of how households use energy and what they spend. Useful comparisons need consistent units, the right survey weights, and a clear account of what the data can support.
Do PFAS treatment models remain consistent when we check their assumptions?
We want to examine the evidence behind PFAS treatment-model claims. That means keeping reaction rates, concentration changes, treatment endpoints, and energy accounting distinct.
How does travel behavior differ for people with disabilities?
We want to understand aggregate differences in travel frequency, distance, and mode. The goal is evidence that can inform better questions about access and mobility without making assumptions about individuals.
How much of a household’s spending goes toward essentials?
We want to compare essential-expense shares across U.S. income groups. The aim is a transparent comparison where readers can see which expenses count as essential and how each share was calculated.
What can county-level industry data tell us about regional economies?
We want to understand how the scale and concentration of Texas industries differ by county and by measure. Coverage matters as much as a ranking: an incomplete county should remain visibly incomplete.
Can material-property predictions hold up under a fair test?
We want to learn how reliably models predict material properties when the evaluation is protected from leakage. A useful result should survive a genuinely separate test, rather than benefit from information it should not have seen.
Can we recover messages more reliably at the same communication cost?
We want to compare symbolic coding schemes fairly: the same cost, the same simulated channel, and a clear account of errors and recovery. The aim is to learn when a proposed improvement actually meets its target.
What Goldbach patterns can we verify within a fixed finite range?
We want to compute and independently verify representations of even integers as pairs of primes, then examine clearly defined finite density patterns. The goal is a reproducible record of exactly what was checked.
What happens when we test the Collatz map within explicit limits?
We want to examine finite integer trajectories while making the input range, transition counts, and stopping limits explicit. The value is a result that someone else can reproduce without guessing how the test was defined.
Weather observation gaps
Do consecutive gaps distort temperature-extreme estimates more than scattered gaps, and can filling methods preserve honest uncertainty?
Simulated water-leak detection
Can calibration improve leak detection under demand shifts and sensor faults without increasing false alarms?
Solar estimate robustness
Which weather-input errors most change daily solar-energy estimates, and do uncertainty intervals cover the resulting errors?
Stormwater intervention rankings
Do rankings of rain gardens, storage, and baseline drainage remain stable when rainfall and infiltration assumptions change?
Geothermal fault mapping
How well do lightweight fault-mapping methods generalize to held-out geographic regions?
Astronomical signal recovery
How do light-curve preparation choices affect recovery of injected small transit signals at a fixed false-positive rate?
Battery-life generalization
Does apparent prediction accuracy hold up when entire battery cells or charging regimes are kept out of training?
Jim Butcher - research resume
Research direction, selected dated evidence, tools, methods, and a printable resume.
Combined professional and creative portfolio
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.
No matching reading yet.
Try another word or clear the section filter. Saved items belong only to this browser.