Nightly Src Projects Desk Raw Survey (2026-05-29)

This raw note preserves the public-safe basis for the 2026-05-29 nightly src/ projects desk. It uses inspectable local evidence only: README/docs/plans, manifests, branch/status/log metadata, safe modified/untracked filenames, mtimes, tests, checked-in reports, and visible artifacts. It does not publish secret-bearing files, .env contents, hidden local settings, raw prompts/logs/trajectories, private corpus bodies, evaluator/oracle payloads, raw benchmark bodies, checkpoint/model artifacts, biometric/capture data, generated media bodies, or explicit/provocative material.

Where a directory is local-only, sensitive, artifact-heavy, private-corpus-backed, or too skeletal, this note uses category-level wording. The point is survey provenance, not a filesystem striptease. Even build systems deserve a little dignity.

Survey scope and method

• Survey root: /Users/ericfode/src.
• Survey timestamp: 2026-05-29 02:00 PDT.
• Full top-level directory count: 50, including hidden directories.
• Execution shape: exactly 10 top-level Hermes survey lane identities, dispatched as one batch of ten orchestrator lanes.
• Lane coverage audit: controller enumeration found 50 assigned directories, 50 unique assignments, no missing directories, no extras, and no duplicates.
• Lane recursion: all 10 lanes reported spawning exactly three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety/public-summary eligibility. Each subteam reported a further three-way leaf probe. The recorded depth is lane → subteams → leaves; no deeper recursion is claimed.
• Evidence allowed: README/docs/plans, manifests, branch/status/log metadata, safe modified/untracked filenames, mtimes, tests, checked-in reports, and visible artifacts.
• Evidence excluded: secret contents, .env contents, hidden local settings, raw prompts/logs/trajectories, hidden evaluator/supervisor payloads, private corpus bodies, explicit/provocative/unsafe material, raw benchmark bodies, checkpoints/model artifacts, biometric/capture data, generated media bodies, and directories too skeletal for responsible public claims.
• Illustration: two raster image-generation drafts were rejected because they contained readable/gibberish labels. The final editorial illustration was saved at queries/news-assets/2026-05-29-project-desk-hero.svg as deterministic symbolic SVG. It is an illustration, not a screenshot.

Ten survey lanes

The exact top-level lane count was 10. One provocative/protected-class-sensitive directory in lane 01 is intentionally withheld by name; it was surveyed and counted, but not publicized.

1. .claude; .socket-dev-scan; .tinygrad_research; another-harness; one sensitive social-claim directory withheld by name.
2. basis; basis-hermes; basis-jcode; cardgame1; creative.
3. deer-flow; FACEMUSIC; gas-city-but-its-just-codex; gemma-dungeon; gemma4-tinygrad-opt.
4. handterm; hoid; iii-wiki; is-codex-better; is-it-formal.
5. jepa-expriments; jepa-lang; jepa-poker; justfooln; kettlebellsim.
6. kimi-tests; langfuse; llama.cpp; local-hermes; meta-hermes.
7. nnpl-external-latent-bus; nnpl-shared-bus; nnpl-typed-boundary-ir; openai-symphony; overengineeredlife.
8. parenting-bookshelf-compass; quiz; silly-pi-stuff; spec-dataset-evolution-corpus; src.
9. steward; testing-rl; testing-rl-hermes; textual-world-model; tinygrad.
10. tinygrad-gemma; tinygrad-gemma-gemini; tinygrad-gemma-kimi; unconventional-jepa-lab; word-games.

Public-safe lead candidates

Orchestration, provenance, and spec-reduction work

• gas-city-but-its-just-codex is the strongest orchestration/control-plane lead. Evidence: README/manifests/schemas/docs/formal directories; main branch; modified Rust/operator/workflow/config/doc files; recent 2026-05-25 commits around meta-nomic sandbox and repo-loop toolsmith artifacts; 2026-05-27 mtime evidence in scripts and README. Safe summary: Codex-native durable orchestration/control-plane research. Hold back state/log/benchmark/wiki-source/generated artifacts and local runtime details.
• basis, basis-hermes, and basis-jcode form the Basis/spec-reducer cluster. Evidence: basis is ahead of origin with a 2026-05-24 imaginer workflow commit and untracked spec-pathology experiment material; basis-hermes is a clean plugin/dashboard repo with README/spec/pyproject/tests and May 5 reducer/dashboard commits; basis-jcode has useful reducer/control-plane docs but .basis run artifacts and dashboard/runtime bodies keep it side-room only. Safe summary: structured spec-state custody and deterministic/provenance-backed reduction, not raw packet publication.
• steward remains viable architecture-side-room evidence: README, project charter, architecture and implementation docs, Elixir/Python manifests, and service-kernel/router/explorer tests. Its dirty service/config/docs worktree and private-corpus/log-mining references keep it from being a stable-release lead.
• openai-symphony is public-summary eligible only at the README/SPEC/Elixir orchestration level. Local modified app-server/orchestrator/dashboard/test files and skill-adjacent material stay private.

Game, craft, verifier, and simulation work

• cardgame1 / Dungeon Steward is publication-ready game craft with caveats. Evidence: Godot project, README, MIT license, design/docs/data/tests/scenes, branch hermes/combat-stage-art-fallback-upstream ahead by one commit, and a 2026-04-15 verified combat-stage art fallback fix. Safe summary: browser-first fantasy roguelite deckbuilder prototype with deterministic combat/map/reward systems and generated-art fallback handling. Hold back local agent/generated/private production material.
• gemma-dungeon is a strong research/game lead. Evidence: README, goal/spec/implementation/reward docs, pyproject, schemas, tests, 2026-05-27 verified commits exposing train/eval improvement and delta status tokens, plus 2026-05-28 README/spec mtimes. Safe summary: embedding-native roguelike/world-model research where symbolic game state remains authoritative and Gemma/tinygrad-facing projections are audited. Hold back replays, generated packs, private corpora, model/checkpoint artifacts, and archived plan bodies.
• testing-rl remains the stable verifier/test-generation bench. Evidence: README/SPEC/WORKFLOW, pyproject, docs/formal material, tests, branch master ahead of origin by three commits, and May 10–11 verifier/dashboard/ranking-lift commits. Safe summary: an RL/test-generation environment with evaluator-held references and local verifier evidence. Do not claim completed model training or fine-tune improvement.
• kettlebellsim is the night’s clearest simulation lead. Evidence: branch codex/reward-audit-and-swing-training ahead 36, clean worktree, pyproject, docs/configs/scripts/tests, and 2026-05-09 verified commits adding bounded Modal Isaac execution/probe guards and planar remote handoff boundaries. Safe summary: deterministic local restart and validation pipeline for kettlebell swing/path-signature research before bounded remote simulator/RL execution.
• handterm remains the clean craft/tooling lead: README/Cargo workspace, CPU/GPU terminal components, tests/scripts, clean branch, and April 19 kitty graphics/refactor commits.

JEPA, NNPL, world-model, and humane/static artifacts

• unconventional-jepa-lab is a strong docs-only public lead if operational details are omitted. Evidence: README, mission/gates/lane docs, Makefile/profiles/scripts, branch ahead by one, 2026-05-24 commits publishing the Gas City JEPA lab scaffold and running lanes, plus modified research-lane packets/manifests. Safe summary: local JEPA/world-model lab scaffold with explicit lanes and gates. Hold hidden/local state and raw packet bodies.
• jepa-lang is the cleanest small IR/replay artifact: README, pyproject, docs, tests, and May 18 mtimes support a deterministic typed operation/replay experiment with evidence receipts and inert latent payload boundaries. It is not a git worktree at the inspected root.
• textual-world-model is a research-process side-room with a 2026-05-28 research-loop mtime and browser/dashboard-level material. Safe framing: benchmark-first, falsification-heavy source-code world-model research infrastructure, not a claim of latent-model success.
• nnpl-external-latent-bus and nnpl-typed-boundary-ir are public-safe concept summaries with docs, manifests, reports, and tests around latent-bus and typed-boundary planning ideas. nnpl-shared-bus is useful as an honest negative-result side note; raw run/checkpoint/result bodies stay out.
• parenting-bookshelf-compass is a clean humane/static artifact: README plus index.html, clean main, and 2026-05-25 publish commit. Safe summary: a non-diagnostic two-parent parenting-books compass quiz with explicit disclaimers and no obvious network-send surface in the inspected public files.
• word-games has promising Story JEPA / character-state modeling evidence, but generated run/checkpoint surfaces make it a curated side room rather than an automatic public lead.

Held back from project-specific public detail

The survey fully held back, or reduced to category-only mention, hidden local assistant/settings directories, security/dependency scan artifacts, empty or skeletal directories, one provocative/protected-class-sensitive social-claim notebook, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, raw benchmark outputs, model/checkpoint artifacts, biometric/capture data, creative/canon/world-packet drafts, service configuration, raw test/counterexample bodies, local .env-style material, cache/build/vendor directories, dirty patch/reject variants, and too-skeletal placeholders.

Specific examples of category-level handling: FACEMUSIC was kept high-level because camera/facial-capture material is privacy-sensitive; tinygrad-gemma was treated as artifact-heavy because local benchmark/checkpoint/evolution-state material dominates; llama.cpp and .tinygrad_research were recognized as public upstream/reference substrates rather than local original project leads; langfuse, local-hermes, kimi-tests, quiz, overengineeredlife, and the nested or placeholder src material were not promoted into public claims.

Editorial synthesis

The publishable movement tonight clusters around six claims:

1. gas-city-but-its-just-codex, Basis/Hermes, Steward, and openai-symphony define the orchestration/provenance/spec-reduction room, with Gas City as the most active public-safe lead.
2. cardgame1 and gemma-dungeon carry the game/research craft story: one game prototype, one symbolic-world-model laboratory.
3. testing-rl remains the verifier bench; kettlebellsim is the strongest simulation validation lead.
4. unconventional-jepa-lab, jepa-lang, textual-world-model, and selected NNPL projects are research side rooms where public claims must stay narrower than the local artifacts.
5. parenting-bookshelf-compass is the tidy humane static artifact; handterm remains the clean systems-craft tool.
6. The public page should be narrower than the tree. Tonight’s omissions are not missing work; they are the safety filter doing its quiet, thankless job.