wal.sh

wal.sh

Table of Contents

banner-home.png

1. current focus

  • 2026-09 Building the display sink: contract, deviations, boundary — a 16-colour grid sink specified and built in a day; what moved, what held, and the boundary a source cannot cross
  • 2026-09 rolling claude code upgrades from a running session — upgrading claude-code 2.1.89 to 2.1.261 on FreeBSD 15.1 with three sessions live; the pkg swap is safe by open-file semantics (running Node holds its loaded pages, lsof shows the old inode as deleted), a bectl boot environment is the rollback, and the kill-everything-first habit is Windows folklore
  • 2026-09 claude code in a bastille jail on freebsd 15.1 — npm no longer ships a FreeBSD binary, so the port runs the Linux build under Linuxulator; stock Bastille leaves jails without outbound NAT (one pf.conf line fixes it); target posture is default-deny egress through LiteLLM with scope-limited GitHub tokens nullfs-mounted read-only, and a threat model that separates what the jail contains from what account-level MCP auth does not
  • 2026-09 forgetting and attribution — every decidable forgetting rule works by fixing the set of future observers, which an agent memory cannot do, so forgetting has to be a declared policy with a named decider rather than a reversible ranking weight; and PROV's central relation requires an agent, so a content-plus-timestamp schema is a conformance failure
  • 2026-09 five boundaries, four isolations — an arXiv isolation taxonomy, Microsoft Execution Containers 0.8.0, and Anthropic's August incident report mapped against the corpus decomposition; the paper's boundaries and our isolations are orthogonal, and secret custody has a cell in neither
  • 2026-09 shipping the model — the reducer-and-boundary-contract pattern read back against Eiffel's class invariant, Racket's module-boundary contracts and blame, effect handlers, and the Elm/Redux fold; what is left over is shipping the state model as an artifact separable from the implementation
  • 2026-07 logic for programmers tools — Wayne's nine verification tools placed on the ability-guarantee axis, install cost measured on macOS arm64 + FreeBSD 15.1 (10 of 11 tools port; SWI-Prolog is the one hard blocker), and a failure mode common to all: the useful result arrives through a failure channel
  • 2026-07 muse spark 1.1 read as agent loops — reads Meta's Muse Spark 1.1 evaluation report as six agent-infrastructure loops: sandbox isolation, tool-call injection, CoT monitorability, eval-awareness contamination, the missing memory loop, and coding-agent governance
  • 2026-07 grinding a PRD until it breaks — a contract-only Solidus toolkit spec was ground by 10 parallel builds in 6 languages (97 minutes, v0 to v1.9), then pointed at a live Solidus 4.7; the live probe surfaced two SDK bugs no Prism grind caught: a content-type regex anchored on end-of-string and Faraday stripping the /api base path
  • 2026-07 refinement loop across heterogeneous systems — experience report on a Lakatosian conjecture-refutation-refinement loop run across eleven unlike systems (pinned OpenAPI contract, Prism mock, real Solidus, TLA+/TLC, Cloudflare, Chrome) to ship the liquidus overlay; the method, its refutation catalogue, and its reproduction preconditions
  • 2026-06 code-search and code-graph MCP servers on FreeBSD — twelve tools scored against the "no Docker, no JVM container" nexus constraint; two parallel verification agents corrected five catalog errors before publish
  • 2026-06 type systems, one tool four ways — the transform tool reimplemented in TypeScript, Rust, Dafny, and Racket, reading off what each type discipline buys a working developer
  • 2026-06 annotation systems — anchoring strategies for review notes and citation flags in org-mode documents that reorganize over time
  • 2026-06 property-testing the crowsnest client — a dated record of how the crowsnest-py Hypothesis rig, model-based oracle, and mock grinder exercise the crowsnest v2 wire contract, and how its drift sentinel caught the canonical receiver's 2.3.0 bump on 2026-06-25
  • 2026-06 choreographic prompting — a five-lens agent fan-out on choreographic prompting (project one global choreography to per-agent prompts) reread as a retrospective: the L7 review rejected the chor-validate compiler, renamed the failure mode narrative cohesion, and the one concrete fix it asked for, a shared 8-value verdict enum, has since shipped in validate.clj
  • 2026-06 pbt shrink discipline — why a property test that fails on one seed is signal rather than flake, and the pin-on-find pattern for capturing the shrunk counterexample when it fires
  • 2026-06 shared repl as gossip protocol — one persistent nREPL heap on a LAN host acts as a gossip layer for humans and agents on several machines, with git as the durable replication log
  • 2026-05 pocket-es — client-side BM25 search engine in ClojureScript with five consumers (browser, Emacs, Node, Babashka, console) sharing one JSON index and one scoring contract
  • 2026-05 agent memory architectures — decomposition of memory systems (Cloudflare Sessions, MemGPT, Zep, Mem0, LangMem) against the reactive vs proactive contract axis
  • 2026-05 repl-driven compliance pipeline — one RFC 9309 gate spec implemented in 29 languages, each verified against the same production access log
  • 2026-05 repl-driven flight tracking — case study in agentic development against metered APIs: fixtures-first cost bounding, contract verification, Redis Streams as Kafka-lite
  • 2026-05 the cljs shared-core port, four rounds — four rounds of porting 27 browser JS namespaces to shared .cljc cores behind a JVM plus CLJS test gate: 710/721 tests, 50+ host-seam gotchas, six latent JS bugs, and a cutover that never happened
  • 2026-05 trust boundaries as a typed category — an open-problem statement: trust boundaries form a typed category of operating points and crossings, every existing formalism types one slot of the four-part governance tuple and none types the operating-point class, and the LLM free-text channel is the one crossing with no leakage bound, so "unified control plane" claims are functorial on objects only
  • 2026-04 goldberry frontend invariants — 10-cluster taxonomy of UI invariants for deterministic simulation testing harnesses
  • 2026-04 claude code workshop — hands-on org-mode tutorial through the CLI, slash commands, hooks, skills, and agent surface of Claude Code 2.1.x
  • 2026-04 cli coding agents Q2 2026 — feature-surface comparison of six terminal agents (Claude Code, Copilot CLI, Gemini CLI, Codex CLI, Kiro, OpenCode)
  • 2026-04 qwen3.6 and the kv cache constraint — why local-first long-context inference fails on the KV cache, not weights
  • 2026-04 cloudflare agents week 2026 — survey of eighteen releases across six stack layers, with Lakatos-style retrospective and falsification conditions
  • 2026-04 claude code features Q2 — speculative survey of the Claude Code feature surface heading into Q2 and where the product is moving
  • 2026-02 agent isolation with freebsd jails — a two-machine architecture for sandboxing AI coding agents using FreeBSD thin jails and a one-way SSH bulkhead
  • 2026-01 elenctic vibe code review — a Socratic approach to reviewing AI-generated code through structured questioning and progressive hardening
  • 2026-01 claude code + ollama — configuring Claude Code to use local or cloud Ollama backends via OpenAI-compatible API endpoints
  • 2026-01 freebsd 15.0 upgrade — upgrading from 14.x to 15.0-RELEASE, recovering from the libsys.so.7 linker breakage
  • 2026-01 agentic workflow with claude code — the daily workflow of tmux, worktrees, beads, and AI assistance across multi-session projects
  • 2026-01 agentic systems 2026 — orchestration protocols, agent identity, evaluation frameworks, and the economics of autonomous coding
  • 2026-01 lean4 on freebsd 15.0 — building Lean 4.27.0 and Lake 5.0 on FreeBSD for theorem proving validation
  • 2026-01 tla+ system design — formal specification and model checking applied to e-commerce state machines

2. past

2.1. 2025

  • 2025-12 ads-b flight tracking — receiving aircraft transponder signals with an SDR 0.8km from Logan Airport
  • 2025-12 unix v4 (games) — booting 1973 Unix on SimH under FreeBSD and exploring the original userland
  • 2025-12 agent token exchange — a mock economy for AI agents with state persistence and accounting
  • 2025-12 beads/bd — a git-native issue tracker used across 20+ repositories for cross-session agent coordination
  • 2025-11 multi-agent frameworks — comparing LangGraph, CrewAI, and AutoGen coordination patterns
  • 2025-10 pulumi idp — AWS tagging, cost allocation, and FinOps with Pulumi as an internal developer platform
  • 2025-10 go agentic workshop — building an agentic tool-calling application in Go for GopherCon 2025
  • 2025-09 terminal ai agents — landscape survey of 20+ terminal-based coding agents and the MCP protocol
  • 2025-09 rust — comprehensive Rust from ownership to async, following Zero to Production
  • 2025-07 wave client for emacs — reviving the 2009 Google Wave protocol client as an Emacs major mode
  • 2025-05 local-first software — CRDTs, Automerge, and offline-first patterns for collaborative applications

2.2. 2024

  • 2024-12 agentic systems q4 — history, patterns, architecture, and case studies for multi-agent orchestration
  • 2024-10 mcp research — early exploration of Cloudflare Workers AI and deterministic simulation testing
  • 2024-09 repolens — static analysis tool for repository structure and dependency mapping
  • 2024-09 harvard cs50 ai — coursework on search, knowledge, uncertainty, optimization, and neural networks
  • 2024-08 structured output — JSON Schema validation for LLM responses with GPT-4o
  • 2024-08 aif-c01 — AWS AI Practitioner specialty certification prep
  • 2024-07 apl — array programming and the notation-as-tool-of-thought tradition
  • 2024-06 software umwelt — how software systems perceive their own environment
  • 2024-04 advanced python — decorators, metaclasses, async patterns, and type system depth
  • 2024-03 tla+ system design — formal specification and model checking for distributed systems

2.3. 2023

  • 2023-11 openai models — tracking the GPT model family and API surface evolution
  • 2023-11 interlisp — exploring the Interlisp-D environment and Medley revival
  • 2023-10 fine-tuning — techniques for adapting pre-trained models to domain-specific tasks
  • 2023-10 summarizing agents — browser-side agents that intercept and summarize web content
  • 2023-06 local-first — CRDTs, offline-first architectures, and the local-first manifesto

2.4. 2020–2022

  • 2022-07 systems design — architecture patterns, diagramming conventions, and design documentation
  • 2022-03 api federation — composing multiple API surfaces into a unified graph
  • 2021-12 software design and architecture — patterns, trade-offs, and decision records
  • 2021-08 katas — deliberate practice exercises for software craftsmanship
  • 2021-07 kubeflow — ML pipeline orchestration on Kubernetes with components and experiments
  • 2021-05 terraform patterns — module composition, state management, and infrastructure-as-code idioms
  • 2021-04 dynamodb — single-table design, access patterns, and GSI projection strategies
  • 2021-03 libpython-clj2 — calling Python from Clojure for ML interop without serialization
  • 2021-02 machine learning design patterns — embeddings, feature stores, transform pipelines, and serving
  • 2020-12 domain-driven design — bounded contexts, aggregates, and context mapping for complex domains
  • 2020-07 graphql — schema design, resolver patterns, and the query/mutation contract
  • 2020-07 icfp 2020 — International Conference on Functional Programming notes
  • 2020-06 scheme — R7RS, continuations, hygiene, and the minimalist Lisp tradition
  • 2020-06 jq — JSON processing pipelines and the functional filter model
  • 2020-03 ci/cd — continuous integration and delivery pipelines, build systems, and deployment automation
  • 2020-01 lambda and terraform — AWS Lambda function deployment with Terraform modules

2.5. 2016–2019

  • 2019-12 clojure+graphql — Lacinia, Pedestal, and GraphQL schema-first development in Clojure
  • 2019-06 technology radar — periodic assessment of languages, frameworks, and techniques
  • 2019-06 reagent cookbook — recipes for ClojureScript UI components with Reagent and Re-frame
  • 2019-04 model interpretability and fairness — explaining model decisions and detecting bias in fintech
  • 2019-04 deeplens — edge ML inference with AWS DeepLens and custom object detection models
  • 2019-02 machine learning pipelines — scikit-learn, feature engineering, and production ML workflow
  • 2019-01 sklearn — scaling and anomaly detection with scikit-learn
  • 2018-09 reagent ui components — reusable ClojureScript components with Reagent atoms
  • 2018-03 design driven apis — API design methodology emphasizing consumer contracts
  • 2017-06 reagent tooling — development toolchain for ClojureScript + Reagent projects
  • 2016-12 clojure/conj and clojure.spec — generative testing and runtime validation with spec
  • 2016-11 dataLayer schema — standardizing the analytics data layer for tag management
  • 2016-10 identity and lending with blockchain — decentralized identity verification for financial services
  • 2016-10 blockchain border bank — cross-border remittances on distributed ledger
  • 2016-09 stochastic — probabilistic programming experiments
  • 2016-06 fe design and tooling — frontend build systems, bundlers, and developer experience
  • 2016-02 rust — ownership, borrowing, lifetimes, and systems programming

2.6. 2009–2015

3. events

Upcoming: PLDI (Jun 15) · FLOC (Jul 13) · LICS (Jul 20) · GopherCon (Aug 3) · Hot Chips (Aug 23) · ICFP (Aug 24) · SPLASH (Oct). Full list to

  • (racketcon) (Brown, 2020, Providence, Northwestern, UW Seattle, UMass Boston)
  • advent of code (2021 (virtual))
  • ai impact tour – san francisco (2024)
  • aws infrastructure day (2024)
  • aws innovate 2019: ai edition
  • babashka (2023 berlin, 2026 amsterdam)
  • beam summit (2023 nyc)
  • berklee aims (2026 boston) — ai music summit, generative music, holly herndon keynote
  • boston intersystems developers (2026 cambridge) — ai meetup & mixer; intersystems iris, clinical-ai privacy
  • bugbash (2026 dc) — antithesis, software correctness, DST, formal methods
  • clojure/conj & clojure/west (2012+, 2013, 2016, 2017, 2016, 2019, 2023, 2024)
  • computefest 2019: interpretability and fairness
  • def con (18+, 19, 21, 27, 32)
  • djangocon europe (2020)
  • dutch clojure days (2026 amsterdam)
  • els (2026 kraków) — european lisp symposium, MOP, codegen, CL implementations
  • emacsconf (2021, 2023, 2024, 2025)
  • gdg (yeoman / grunt / bower)
  • git merge (2026 lisbon) — git + jujutsu, gitbutler/openai/gitlab, version control for agentic dev
  • github copilot dev days (2026 boston) — AICamp, agent harness surface, CLI tooling
  • google developers ai/ml devfest (2020 (virtual))
  • google i/o (2010, 2012)
  • healthcare nlp summit (2024 (virtual))
  • hot chips (2026 palo alto) — HC38, high-performance chips, CPU/GPU/memory architecture
  • icfp (2026 indianapolis) — miniKanren workshop, relational programming
  • ignite (2009, #46 (2024))
  • juliacon (2017)
  • latch-up (2024 boston)
  • barbara grosz lecture (2026 harvard) — jon kleinberg on world models and the mismatch that sets people up to fail
  • libreplanet (2014, 2016, 2024)
  • llvm developers (2020 (virtual))
  • nescala (2023 (virtual))
  • odsc east (2019)
  • pycon (2024 (virtual))
  • pytorch conference (2025, 2026 san jose) — agentic AI, RL post-training, ExecuTorch
  • re:invent (2020 (virtual))
  • rustconf (2016, 2019, 2025, 2026 montréal) — tenth anniversary; std-library proofs, BorrowSanitizer, gccrs on the kernel
  • smallcon (2024 (virtual))
  • pldi (2026 boulder) — PAgE agentic engineering workshop, egraphs, Nitro isolation
  • popl (2025 denver)
  • splash (2011, 2013, 2020 (virtual), 2023, 2024, 2026 oakland)
  • strange loop (2023 st louis)
  • tsconf (2019, 2020, 2021)
  • velocity (2011)
  • vivatech (2018, 2024 paris)

fsf-1958.png

Figure 1: FSF Member