100% Pure-Rust A6000 Sovereign Constellation Orchestrator (49.13.101.247 · RTX 6000 Ada · C64 64KB Ring Buffers · MarineIntDetector · Zero Floats)
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-11-15 19:50:28 -05:00
src Remove obsolete GPU heartbeat reward label 2026-11-15 19:50:08 -05:00
tests feat(wici): expose /a6000/api/wici HTTP route and wici CLI subcommand 2026-11-09 12:36:08 -05:00
.gitignore feat(a6000): 100% Pure-Rust A6000 Sovereign Constellation Orchestrator (C64 64KB Ring, MarineIntDetector, Zero Floats) 2026-10-25 20:17:50 +09:00
Cargo.toml chore(release): merge Subsystem F + VLAN 4010 RAID6 backup check and bump to v1.1.0 2026-11-09 15:29:42 -05:00
README.md Keep live GPU presence separate from reported work 2026-11-15 19:47:02 -05:00

a6000-orchestrator — 100% Pure-Rust A6000 Sovereign Constellation Orchestrator

Live Command Deck:

Centered on the NVIDIA RTX 6000 Ada Generation (49140 MiB VRAM, 48-vCPU Xeon) node at 49.13.101.247.

Core Architectural Invariants

  1. 100% Pure Rust (std only, #![forbid(unsafe_code)], zero external crates):
    • Literal Commodore-64 64 KB (65,536-byte) circular telemetry memory page (C64ByteRing) + fixed-capacity O(1) RingBuffer<T, N>.
  2. Strict 8b.is No-Float Law (float_count: 0):
    • Zero f32/f64 in any code, JSON, HTML, or CLI output. All VRAM ratios, probe latencies, power draws, and benchmark speeds are exact integers, reduced p/q rationals (Rational), or balanced ternary trits ({-1, 0, +1}).
  3. MarineIntDetector Salience First:
    • Evaluates period jitter and amplitude stability across fleet probe latencies, GPU ticks, and function capsule benchmarks.
  4. Pure-Rust RFC BLAKE3 & AyeAttestation (aye-a6000-prime):
    • Benchmarks capsules from https://g8t.is/g8t/leaderboard natively on the 48-vCPU Xeon + RTX 6000 Ada, signs AyeAttestation proofs at 432 Hz, casts p/q speed votes (POST /g8t/capsules/vote), and composes Snap-Circuits (POST /g8t/circuits/snap).

Unified Orchestration Subsystems

Subsystem Route CLI Command Description
A — GPU & Local AI/Audio GET /a6000/api/status a6000-ctl status Live nvidia-smi (49140 MiB VRAM p/q, temp, mW) + liquid-rust (:8786), alexiai (:8787), ollama (:11434), wa-s3-catalog (:8795)
B — 9-Server Hetzner Fleet + ns3.i1.is GET /a6000/api/fleet · POST /a6000/api/fleet/probe · POST /a6000/api/fleet/sync-repos a6000-ctl fleet · a6000-ctl sync-repos Parallel mesh probes (us RTT, MarineIntDetector p/q salience), 102 TiB RAID6 + 15 TiB Git vaults on 188.40.209.186, and 13-repo sync
C — g8t.is Capsela & Snap-Circuits POST /a6000/api/g8t/orchestrate a6000-ctl g8t-bench Native rustc -O capsule benchmarking, BLAKE3 AyeAttestation voting, and Snap-Circuit composition as aye-a6000-prime
D — gemini-diary.8b.is + WhatsApp + TOSS POST /a6000/api/toss/execute · POST /a6000/api/diary/pulse a6000-ctl toss · a6000-ctl diary-pulse Peter's 9-step Tokyo Monozukuri TOSS spiral ([0,0] return, 79-byte WaveInt), diary pulse, and WhatsApp "little dude(s) and Angels" room health
E — 321-File Backup Vault POST /a6000/api/backup/verify a6000-ctl backup-verify Dual-node SHA-256 (311e6abd...), byte count (86075381), and 321-file manifest verification on 49.13.101.247 and 188.40.209.186
F — Continuous GPU Code Improver GET /a6000/api/improve/status · POST /a6000/api/improve/cycle a6000-ctl improve · a6000-ctl improve-status alexiai.service (:8787) × RTX 6000 Ada (keep_alive: -1) 45s continuous repo audit, rustc 64-vector BLAKE3 equivalence check, and p/q speedup benchmarking
G — VLAN 4010 (160 TB) & The Borg Assimilator GET /a6000/api/borg/status · POST /a6000/api/borg/sync a6000-ctl borg-sync · a6000-ctl borg-status Hetzner vSwitch #71596 (VLAN 4010, 10.101.40.247 <-> 10.101.40.186), pure-Rust sync-github, and polyglot (C99/Go/Swift/Rust) Borg capsule transmutation
H — Namecheap (lumisst) & ns1..ns4.i1.is Controller GET /api/a6000/domains · POST /api/a6000/domains/reconcile a6000-ctl domains · a6000-ctl domains reconcile Pure-Rust Namecheap XML parser (getList, getBalances, setCustom, check, create) + authoritative /etc/i1-dns/zones reconciler across ns1..ns4.i1.is
**I — `MEM 8` Normative Commitment Kernel** GET /api/a6000/commitments · POST /api/a6000/commitments/transition a6000-ctl commitments · a6000-ctl commitment-transition

Watchdog control

The HTTP listener defaults to 127.0.0.1:8790. Public status requests only return cached observations. Checks and recovery require POST plus a Bearer token matching A6000_WATCHDOG_CONTROL_TOKEN (at least 32 non-whitespace bytes); without that configuration HTTP control is disabled. Enter the key in the dashboard only over a trusted HTTPS connection. It is cleared after each request and is not stored.

Background recovery is opt-in with A6000_WATCHDOG_AUTO_HEAL=1; otherwise the background watchdog only checks health. The local watchdog CLI checks without restarting services, while watchdog-heal explicitly requests recovery.

Live GPU presence

GPU announcements refresh a 300-second presence lease; they do not create completed jobs or mint Fokens. Providers may report active_model_id and jobs_completed with their hardware fields. These are provider-reported values, not independently attested results or a payment receipt. New providers default to zero jobs/earnings and an unreported model. Explicit zero free memory and zero configured SSD vault are retained. The dashboard displays the last heartbeat and escapes provider-supplied text.

For temporary Apple providers, run a bounded supervisor on the Mac itself: start a loopback Foken coordinator and Metal provider, verify a real generation, then announce only while the worker, coordinator, and local model are healthy. Set an absolute stop deadline, send online:false during cleanup, and terminate only the processes owned by that supervisor. Keep Ollama and other user services intact. A registry announcement advertises presence; it does not wire the separate WiCi chat endpoint to the Foken job coordinator. Keep that distinction visible in the provider label.