[attu308]

Development has ended. This marks the project closed in the READMEs and corrects five things the docs claimed that the code contradicts.

Why it ended

Two load-bearing premises failed while the engineering was succeeding:

• P3 — the economics do not close. scripts/economics.py puts a realistic host at $2.256/1M tokens against a $0.150 commodity price (~15×), with break-even utilisation unreachable. That killed the marketplace and the reason a stranger would contribute hardware (D2).
• P2 — NAT traversal is not a differentiator. The desk review found the falsifier true as written; Tailscale solves the single-machine case. What survives is cross-party pooling, which relocates the whole claim onto P1 — for which there was never any evidence.

D7 (a second pair of eyes) never got its human reviewer, and 1.5 — a request served across a real NAT — was never earned, despite being the product.

Discrepancies found by reading the code, and fixed here

• packages/node/README.md had the architecture backwards. It said the Ollama proxy was "the only backend runtime.py assigns". It is not a runtime.py backend at all: runtime.py:282 assigns EchoBackend and nothing else, ever; OllamaBackend is never constructed outside a test; and nothing in src ever enqueues onto task_queue. So Runtime._worker_loop → InferenceBackend → ArtifactStore is dead code in production, kept alive by a test named test_end_to_end_pipeline that feeds the queue by hand. The live path is api/inference.py → clients/ollama.py, a different class.

• The "155-word vocabulary" figure was never true. LocalBoundaryEngine holds 116 words plus 4 special tokens, against a declared vocab_size of 32,000. Counted, not recalled. The wrong number had propagated to five files, including CLAUDE.md and .claude/rules/node.md — the file governing how agents work in this repo.

• The v1.0.1 tag names a tree that declares itself 1.0.0 in all four packages. test_every_package_declares_the_same_version compares the four to each other and never to the tag, so V3's fix converged them on the previous version and the ratchet passed. Left as found and documented rather than bumped: a tree claiming to be v1.0.1 that is not the released v1.0.1 is the same confusion in a different hat.

• packages/website/README.md called the project "a globally distributed, community-owned AI infrastructure" — the thing D6 decided not to build and D2 priced out. The site's own landing component had been corrected; the README had not. It also listed 5 of 9 pages, omitting dashboard, playground and status (the functional ones), and described the sibling packages as separate repositories.

• "Python 3.11–3.14" implied four interpreters; ci.yml names three (3.11, 3.12, 3.14). 3.13 is not in the matrix.

Verification

./scripts/verify.sh run as a separate pass against the committed tree: PASS, 25 checks, commit d9dc89c9, clean tree.

Two checks are UNVERIFIED, not passed — the gate names them itself:

• compile under python3.14 (interpreter not installed in this environment)
• website lint + tests (no node_modules)

CI on this PR runs both, which is the point of landing it this way.

Test evidence from the session that wrote this: 906 passed, 1 skipped — 397 scheduler / 260 node / 249 root.

STATUS.md was regenerated with scripts/generate_status.py, not hand-edited. It still reads CI UNVERIFIABLE because there is no gh CLI in this environment; the root README explains that and cites run #33 on main at e5d1e7c, read through the API, which concluded success.

Note

The repository is deliberately not being archived on GitHub — it stays writable. The READMEs state plainly that it is unmaintained, that there is no security response process, and that issues and PRs are not monitored.

---

Generated by Claude Code