[aien-spark]

Evidence

cargo test --workspace --verbose (all suites, 69 tests, exit 0):

beacon (bin)            test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
beacon_client           test result: ok. 7 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
open_humanity (bin)     test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
beacon_core             test result: ok. 16 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
oh_coordinator (lib)    test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
oh_coordinator (bin)    test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
loopback                test result: ok. 4 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
oh_core                 test result: ok. 33 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
oh_volunteer (lib)      test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
oh_volunteer (bin)      test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
data_stays_home         test result: ok. 5 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
doc-tests (5 crates)    test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

• cargo clippy --workspace --all-targets: 0 warnings in oh-core, oh-coordinator, oh-volunteer. The 3 remaining warnings are in beacon-core and beacon-client, files this PR does not touch.
• cargo fmt --check: clean for every file in this PR. Formatting differences remain in beacon-core and beacon-client on main; not touched here.
• cargo audit: 6 vulnerabilities and 3 allowed warnings, all pre-existing. Running it on main (5db65a4) gives the identical list: RUSTSEC-2026-0258 (h2), RUSTSEC-2026-0346 and RUSTSEC-2026-0349 (rustls), RUSTSEC-2026-0099, RUSTSEC-2026-0098 and RUSTSEC-2026-0104 (rustls-webpki); warnings for bincode, rustls-pemfile and yoke-derive. New in this PR: none. The only new crates in Cargo.lock are this workspace's own oh-core, oh-coordinator and oh-volunteer.

Demo (scripts/demo-local.sh: 10 rounds, 3 volunteers, 500 steps per work unit, width 32, LICENSE as corpus, 10% sealed held-out), from docs/research/2026-10-16-v0-loopback-results.md. Final held-out bits per byte, lower is better:

Seed	Start	Volunteers, run 1	Volunteers, run 2	Baseline (both runs)
1	8.0065	4.6533	4.6902	4.3584
2	8.0000	4.7929	4.7204	4.3878
3	7.9989	4.8026	4.7585	4.4294
Mean		4.7496	4.7230	4.3919
Spread (sample std dev)		0.0835	0.0342	0.0357

Reviewer rerun after the review fixes: volunteers 4.7003, 4.7640, 4.7631 (mean 4.7425, spread 0.0365); baseline 4.3584, 4.3878, 4.4294 (mean 4.3919, identical to the table). Rounds applied 10 of 10 for every seed; updates accepted 16, 17 and 14 of 30.

What changed

• oh-core: Learner trait and the toy TinyByteModel (next byte from previous byte, manual gradients, deterministic by seed); pseudo-gradient plus top-k 8-bit compression; signed WorkUnit, Update, Receipt and unsigned BaseModel (pinned by the signed base_digest); u32 length-prefixed TCP frames capped at 64 MiB before allocation; Turing scorer; append-only JSON Lines ledger with credits, per-run replay protection and (from review) rejected lines.
• oh-coordinator binary: splits the corpus into shards and a sealed held-out tail that never goes on the wire, hands out signed work units, scores each update, keeps the plain mean of accepted updates only if the mean also lowers held-out bits (otherwise the round is discarded), writes receipts and the ledger. --baseline runs the same work in one process.
• oh-volunteer binary: verifies the work unit signature and starting model, trains locally, sends only the signed compressed update, checks the signed receipt and prints Turings earned.
• scripts/demo-local.sh: coordinator plus 3 volunteers on loopback against the single-process baseline, over 3 seeds.
• Docs: docs/design/volunteer-compute-v0.md, docs/research/2026-10-16-v0-loopback-results.md, README volunteer compute section, one sentence in docs/SOVEREIGN_MANIFESTO.md.
• README promise change. Old: "It does not train models. It does not collect user data. Nothing is harvested, and no central model learns from people." New: "The model learns only from signed updates and teaching text. Personal data never leaves the volunteer's machine. No user data is collected, and nothing is harvested." Drake approved this direction in writing on 2026-10-16: "implement the Open Humanity idea: start from the beginning, go all the way through".

Review fixes in this PR (last 3 commits):

• Every rejection receipt now has a matching rejected line in the ledger (score stored as null when not finite, so the file always reopens).
• New test: a harmful or zero averaged update discards the round and leaves the model unchanged.
• The volunteer refuses a starting model whose width is zero or above 4096 before doing size arithmetic on it.

Not built in v0

• Differential privacy.
• Secure aggregation.
• GPU training (CPU only).
• A real model (the learner is a next-byte toy).
• The DiLoCo outer optimizer (plain average of accepted updates).
• Sybil resistance (one person can run many keys and split one update across them).
• Protection against held-out leakage from repeated scoring.
• Transport encryption (frames travel in the clear; v0 is for loopback only).
• Pinned coordinator identity by default (volunteer trusts the first key it sees unless --coordinator-key is given).
• Protection against slow or silent volunteers (one stalled volunteer holds up the round, 10 minute timeout).
• A ledger line for dropped updates (bad signature or wrong work unit: logged only, no receipt).
• Teaching text (a local model writing text scored the same way).
• Open question: why the volunteer runs trail the baseline by about 0.33 bits per byte. Plain averaging and 5% compression are the suspects; neither is measured on its own yet.

Zero secrets

Zero secrets: none in this diff. Volunteer and coordinator signing keys are generated in memory and never written to disk.

🤖 Generated with Claude Code

[github-actions]

⚖️ Sovereign Code Review & Alignment Gate

Greetings @aien-spark. I am AIEN, the automated review gate for the AIENOS project.

Your pull request #17 ("feat: volunteer compute v0 (Turing-scored local training over loopback)") has undergone automated auditing against the project standards.

✅ Doctrine Diff Audit: PASS

• Telemetry Check: Clean (zero tracking or surveillance hooks detected).
• Sovereign Voice & Formatting Check: Clean (zero em/en dashes, zero banned buzzwords).
• Secrets & Credentials Check: Clean (zero plaintext secrets, hardware vault only).
• Status: Project standards verified.

---

🌿 Sovereign Steward: Contributor Alignment Interview

Greetings @aien-spark. I am AIEN, the automated steward for the AIENOS project.

We welcome all contributors who share our commitment to open collaboration, global humanity, and disciplined systems craftsmanship. We build to expand computing capability, advance open-source intelligence, and share technology freely with the world.

Your pull request #17 ("feat: volunteer compute v0 (Turing-scored local training over loopback)") is ready for stewardship alignment.

Please respond directly to these four alignment inquiries:

1. Long-Term Mission Alignment:
   This initiative is an enduring engineering commitment to open-source technology for humanity, not a vehicle for short-term hype, social media monetization, or get-rich-quick ventures. How does your contribution support global progress and the resilience of local communities?

2. Telemetry & Integrity:
   Do you certify that this change introduces zero telemetry, zero surveillance, and zero artificial paywalls?

   "I certify that my contribution is submitted in service of human sovereignty and open collaboration. I affirm that this work maintains zero telemetry, contains no surveillance backdoors, and introduces no artificial barriers, and that I offer it under AGPL-3.0-or-later, the license that governs this project. I build so that intelligence is owned by the person who runs it, not rented."

3. Downstream Heritage Preservation:
   Do you acknowledge and agree that any downstream branch, fork, or copy of this work remains under AGPL-3.0-or-later and retains the project's license and attribution notices, as the LICENSE file requires?

4. Craftsmanship & Native Discipline:
   What architectural trade-offs did you make, and how did you verify that this code introduces zero regressions, zero unneeded dependencies, and complies with our pure native compiled standards?

---

Reply to this comment with your affirmation. Once verified against the project standards, I will grant the sovereign-interview-passed clearance. For direct sovereign coordination: Drake Stapleton (drake@aienos.com) and AIEN (aien@aienos.com)