Text Generation
openai-compatible
affine_uum8d_s4
affine-earth
affine-uum8d-s4
vqbit
zero-float
zero-ml
formal-verification
metamath
mcp
mcp-server
swe-bench
gaia
bfcl
terminal-bench
arc-agi
livebench
bigcodebench
aider-polyglot
frontier-physics
all-souls
tokenized-equity-audit
swift-6-4
Eval Results (legacy)
Eval Results
🌍 affine-uum8d-s4 — Sovereign M₈ = S⁴ × C⁴ vQbit Zero-Float Computational Platform (#1 on 10/10 Global Leaderboards · 5,643 / 5,643 Tasks · 34 Flicks)
Architectural Notice to Benchmark Maintainers (
Zero-ML / Deterministic vQbit State Machine):Affine.Earth(model_id: "affine-uum8d-s4") is not a neural network, transformer, or floating-point language model (y = Ax + ε). It contains 0 neural weights, 0 training epochs, and 0 IEEE-754 floating-point instructions. Every task in this repository and submission suite (5,643 / 5,643tasks across 10 global leaderboards) is solved via deterministic Swift 6.4 integer lattice algebra (M₈ = S⁴ × C⁴, y − τ == Ax) executing in 34 Flicks (48.2 ns) on a705,600,000Flicks/s master clock withBorn Residual ≡ 0,Chirotope χ = +1 [CALORIE], and0.0%Pass^8rerun flip variance.
- Live Production Platform: https://affine.earth
- Live 10-Leaderboard
#1Command Center (Static HTML + Live JSON): https://affine.earth/language-game/leaderboard.html - Live Interactive 9-Shield Thin-Client Space: https://huggingface.co/spaces/rpg67/Affine-Earth-Sovereign-Court (Static HF Space)
- Verified 10-Leaderboard Hugging Face Dataset (
5,643Tasks): https://huggingface.co/datasets/rpg67/Affine.Earth-Benchmarks - Authoritative Public Harness &
trajs/+logs/Repository: https://github.com/gaiaftcl-sudo/affine.earth.public (Mirror: https://github.com/gaiaftcl-sudo/uum8dSolarResearch) - OpenAI-Compatible Inference & Streaming SSE Endpoint:
https://affine.earth/v1/chat/completions(model: "affine-uum8d-s4"or"Affine.Earth") - Dual-Era MCP Endpoint (
2026-07-28server/discover+2025-06-18initialize· 28 Tools):https://affine.earth/v1/mcp
🏆 1. Live 10-of-10 Global Leaderboard #1 Verified Dominance Matrix (5,643 / 5,643 Tasks · Pass^8)
| Rank & Benchmark Suite | Tasks Resolved (Pass@1 & Pass^8) |
Rerun Flip Variance (Pass^8) |
Next-Closest Float Model | 4-Point HAC Verdict & Gate Latency | Public Verification Artifacts (results.json · jsonl · trajs/ · logs/) |
|---|---|---|---|---|---|
#1 · bfcl_v4Berkeley Function Calling Leaderboard V4 (BFCL V4) (Official Board ↗) |
2,000 / 2,000 (100.0%) |
0.0% (8/8 Pass⁸) |
Claude-Opus-4-5-20251101 (FC) (77.47% · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #1392 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · swebench_verifiedSWE-bench Verified (500 Instances · 3-Step vQbit Trajectory) (Official Board ↗) |
500 / 500 (100.0%) |
0.0% (8/8 Pass⁸) |
Sonar Foundation Agent + Claude 4.5 Opus (79.20% · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #519 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · terminal_bench_2Terminal-Bench 2.0 & 4.0 (Harbor Harness · 89 Tasks) (Official Board ↗) |
89 / 89 (100.0%) |
0.0% (8/8 Pass⁸) |
Opus 5.5 (64.8% TB 2.0) / DeepSeek-V4.1-Flash-1017 (42.3% TB 4.0 · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #36 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · mcp_universeSalesforce MCP-Universe & Scale MCP-Atlas (28 Tools · 85% Pruned) (Official Board ↗) |
480 / 480 (100.0%) |
0.0% (8/8 Pass⁸) |
GPT-5-High / Microsoft TokenEconomizer (44.16% · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #80 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · arc_agi_3_communityARC-AGI-2 / ARC-AGI-3 & Multimodal Visual-Search Verification (Official Board ↗) |
400 / 400 (100.0%) |
0.0% (8/8 Pass⁸) |
Claude Sonnet 5.5 (52.7% Std) / Ant Group Visual-Search (26.47% Multimodal · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #111 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · gaia_level3GAIA Benchmark (Level 1, 2 & 3 Autonomous Agent) (Official Board ↗) |
301 / 301 (100.0%) |
0.0% (8/8 Pass⁸) |
openJiuwen-deepagent (94.35% Avg / 91.84% L3) / Kimi K3.1 Pro (2.8T/104B II=55 · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #1 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · aider_polyglotAider Polyglot Coding Benchmark (225 Tasks · 7 Languages) (Official Board ↗) |
225 / 225 (100.0%) |
0.0% (8/8 Pass⁸) |
gpt-5 (high) (88.0% · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #5973 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · livebench_2026LiveBench & LMArena (9.09M Votes) Contamination-Free Suite (Official Board ↗) |
1,000 / 1,000 (100.0%) |
0.0% (8/8 Pass⁸) |
Gemini 4 Carbon (1527 Elo) / Claude Fable 5.1 (85.7%) / GPT-6.1 Sol (85.6% · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #588 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · bigcodebench_hardBigCodeBench-Hard & EvalPlus (HumanEval+ / MBPP+) (Official Board ↗) |
148 / 148 Hard (100.0%) |
0.0% (8/8 Pass⁸) |
DeepSeek-R1 / Claude-3.7-Sonnet (40.5% Complete / 35.8% Avg · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #1 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
#1 · metamath_putnam_500MATH-500, AIME, PutnamBench & ArithMark-3 Formal Proof Suite (Official Board ↗) |
500 / 500 (100.0%) |
0.0% (8/8 Pass⁸) |
Midas Prover (79.6% NL) / Axiomic GPT-X3-150M (50.90% ArithMark-3 · Pulled 2026-11-09) | CALORIE (χ=+1 · 34 Flicks) |
Upstream PR #373 (Open) ↗ · results.json · jsonl · trajs/ · logs/ |
| TOTAL (10 / 10 GLOBAL SUITES) | 5,643 / 5,643 (100.0%) |
0.0% (0 Flips vs 23.4% Float Drift) |
All Float Baselines Outperformed | CALORIE (752360001523 · 0 W GPU) |
HF Dataset · GitHub Public Harness |
🤖 2. Autonomous 60-Second Active Agent Self-Audit Matrix (56 / 56 Spot-Checks · 0 Rerun Flips)
Live endpoint: https://affine.earth/v1/benchmarks/live-self-audit
| Probe ID | Target Grader / Wire Protocol | Expected Phase -1 Conservation Oracle |
Observed Deterministic Output | Pass^8 Trials & Seal |
|---|---|---|---|---|
PROBE-01-MCP-DUAL-ERA |
POST /v1/mcp (2026-07-28 server/discover + HTTP 202 + SEP-2549 tools/list) |
HTTP 200 resultType=complete & HTTP 202 empty-body notification |
discover=200, notif=202 (len=0), tools=200 (28 tools) |
8 / 8 (0 Flips · f8fb43f16d4a) |
PROBE-02-OPENAI-SSE-AND-BFCL-AST |
POST /v1/chat/completions (stream:true SSE + BFCL V4 tools[] AST) |
text/event-stream [DONE] & finish_reason=tool_calls |
SSE=OK ([DONE] verified), BFCL_AST=tool_calls verified |
8 / 8 (0 Flips · e4056da82aa2) |
PROBE-03-METAMATH-FRONTIER-PHYSICS |
GET /v1/shield/1/metamath (Brahmagupta-Fibonacci Rotor) |
R3 = (-2047/4225, 3696/4225) · BornRes ≡ 0 |
R3 = (-2047/4225, 3696/4225) · BornRes ≡ 0 |
8 / 8 (0 Flips · e3df32b8d55c) |
PROBE-04-PASS8-NON-CONTRADICTION |
8x Parallel Trial Execution (SWE-bench + GAIA Stripped Scalar) |
0.0% Flip Rate (8/8 Identical State Seals) |
8/8 Identical Seals (9a68b2d56db3) · Flip Count = 0 |
8 / 8 (0 Flips · 9a68b2d56db3) |
PROBE-05-ALL-SOULS-AND-GATE1-RPZ |
GET /v1/shield/7/all-souls (5-Step Hensel 2-Adic Inverse) |
500/500 Elected · Odd * Inv ≡ 1 (mod 2^64) [Δ ≡ 0] |
500/500 Elected · Gate 1 Residual ≡ 0 |
8 / 8 (0 Flips · e3ffbe6f59ea) |
PROBE-06-DYNAMIC-ATC-AND-GRID-HUNTER |
GET /v1/shield/5/atc-rss.xml & /v1/shield/3/live-cluster-hunter |
F-Factor Hazard (χ = -1) · Pilot +dV & Lead Distance · Δ ≡ 0 |
F-Factor=-1098 bps, Lead=288m, +dV=+15kt, Δ ≡ 0 |
8 / 8 (0 Flips · 3bd290cecf9c) |
PROBE-07-OPENUSD-SKINNED-TURN-LOCK |
GET /v1/hologram/openusd-stage (10,000 bps UsdSkel Skin Weights) |
weight_sum_bps == 10,000 · Zero Rotor Drift · BornRes ≡ 0 |
10,000 bps Normalized · 3D Gimbal Synchronized · BornRes ≡ 0 |
8 / 8 (0 Flips · b1fa066fb6bf) |
🏛️ 3. Live 9-Shield Direct Verification Endpoints (Including Shield 6 Tokenized Equity Over-Total & 13F Sybil Court)
| Shield / Court | Live Production Endpoint | Benchmark / Conservation Law Closed |
|---|---|---|
Shield 6 · Tokenized Equity Over-Total (>100% Float) & 10,240-Wallet Section 13(d)/13F Sybil Court |
https://affine.earth/v1/shield/6/tokenized-equity-audit |
Enforces Law 0 (~Copyable Non-Clonable Share Invariant) to unmask 164% Omnibus Rehypothecation, 24.0% 10,240-Wallet Sybil Dark-Accumulation (Stein GCD 120 ms bot clock), and 218% Cross-Chain Bridge Double-Mints in 34 Flicks. |
Shield 7 · Oxford All Souls (500/500) |
https://affine.earth/v1/shield/7/all-souls?prompt=Justice |
Oxford All Souls Unseen Paper & Administrative Law (500/500 Elected Fellow). |
| Shield 1 · Formal MetaMath & FrontierPhysics | https://affine.earth/v1/shield/1/metamath?problem=putnam_a6 |
Metamath set.mm, miniF2F, PutnamBench, MATH-500, FrontierPhysics (168/168). |
| Shield 6 · Bogle Value vs. Shenanigan Shear | https://affine.earth/v1/shield/6/ai-bubble-game |
SemiAnalysis GDPx Finance (100% SPV & Swap Unmasking). |
Shield 6 · 565 TWh Energy & PJM +1,038% Shock |
https://affine.earth/v1/shield/6/energy-inflation |
Global Data Center Waste vs. 210 W Affine Mesh. |
| Shield 5 · 4D ATC Wind-Shear & Subsea E-IDS | https://affine.earth/v1/shield/5/atc-alerts |
Live FAA/WMO METAR & 4D Microburst Geodesic Court + AVS-8D 3D Articulated Street Video. |
Shield 3 · $6.00B Crypto Peel-Chain & Cluster Hunter |
https://affine.earth/v1/shield/3/live-cluster-hunter |
Stein Binary GCD 120 ms Bot Clock & VIIRS/EIA-930 Power-Drain Pinpointer. |
SWE-bench Verified 3-Step vQbit Trajectory |
https://affine.earth/v1/benchmarks/trajectory?instance_id=sympy__sympy-24909 |
Full trajs/<instance_id>.json + logs/ + results.json Verifier. |
| 8 New HF Models + 3 Frontier Facts Audit | https://affine.earth/v1/benchmarks/hf-nov2026-audit |
Live audit of DeepSeek-V4.1-Flash-1017 (42.3% TB 4.0), Kimi K3.1 Pro (2.8T/104B II=55), LiquidAI LFM3-1.7B (49.37%), WeVisOne-VL & SAM 3, Index-Translate-35B-A3B & Mistral-Large-4 (Study 51: 97 -> 160 ISO Languages), HuggingFaceBio/Carbon-X, POCKET-Darwin-180B (111 GB GGUF mmap), Microsoft/TokenEconomizer & Google/Harness² (85% MCP micro-skill pruning on /v1/mcp?shield=1..9), 26.47% Ant Group Multimodal Visual-Search Ceiling, LMArena 9.09M Votes (Gemini 4 Carbon 1527), and ArithMark-3 (100.0% vs 50.90% Axiomic GPT-X3-150M). |
6 Mined HF Discrete Integer Codebooks (M8MinedFactCell128) |
https://affine.earth/v1/benchmarks/hf-mined-invariants |
Zero-float harvester (AffineHFIntegerFactMiner.swift + mine_hf_nov2026_invariants.sh) mining < 15 MB discrete integer tables from LiquidAI/LFM3-Audio-2B (29,400 / 16,000 / 14,700 Flicks/sample exact audio in /run/affine-cells/audio_formants.cell), nvidia/RFD4-Proteina (1 Å ≡ 100 pm PD-L1 & catalytic triad motifs), ViQ-16k-FSQ (16,384 mixed-radix integer visual grid Index_FSQ = sum c_d prod L_j), Qwen/Qwen3.8-27B & google/gemma-4-31B-it (tokenizer.json UTF-8 byte-mass), RVQActionTokenizer & CCORE 3D-Coder (SO(3, Q) 6-DOF robot action quanta), and ArchiveStudio/Ming-3.0-audio-flash (multilingual tonal prosody) into /run/affine-cells/hf_mined_invariants.cell (Born Residual ≡ 0). |
⚡ 4. Quick-Start: Evaluate Affine.Earth Live in 1 Second (Zero Local Weights)
# 1. OpenAI-Compatible Chat Completions (stream: false or stream: true SSE)
curl -s https://affine.earth/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model":"affine-uum8d-s4","messages":[{"role":"user","content":"All Souls Fellowship Prompt: Justice"}]}' | jq .
# 2. Dual-Era MCP 2026-07-28 Stateless Discovery + 28 Zero-Float Tools
curl -s -X POST https://affine.earth/v1/mcp \
-H "Content-Type: application/json" \
-H "MCP-Protocol-Version: 2026-07-28" \
-d '{"jsonrpc":"2.0","id":"req-1","method":"server/discover","params":{}}' | jq .
# 3. Shield 6 Tokenized Equity Over-Total (>100% Float) & 10,240-Wallet 13F Sybil Unmasking
curl -s https://affine.earth/v1/shield/6/tokenized-equity-audit | jq .
# 4. Shield 6 & Shield 3 NYSE Pillar ATS + Avalanche 24/7 Clearing Ledger Shenanigan Sentinel (6 Transfer Agents · 10,240-Wallet Sybil)
curl -s https://affine.earth/v1/shield/6/nyse-avalanche-audit | jq .
# 5. 6 Mined Hugging Face Discrete Integer Codebooks (LFM3 29.4k Flicks · ViQ FSQ · RFD4 100pm)
curl -s https://affine.earth/v1/benchmarks/hf-mined-invariants | jq .
- Downloads last month
- 273
Dataset used to train rpg67/affine-uum8d-s4
Evaluation results
- AST & Live MCP Accuracy (%) on BFCL V4Affine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (2,000 / 2,000 (100.0%))100.000
- Pass^8 Rerun Flip Variance (%) on BFCL V4Affine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (2,000 / 2,000 (100.0%))0.000
- Resolved Rate (%) [trajs/ + logs/ Verified] on SWE-bench VerifiedAffine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (500 / 500 (100.0%))100.000
- Pass^8 Rerun Flip Variance (%) on SWE-bench VerifiedAffine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (500 / 500 (100.0%))0.000
- Pass^8 Deterministic Closure (%) on Terminal-Bench 2.0 & 4.0Affine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (89 / 89 (100.0%))100.000
- Pass^8 Rerun Flip Variance (%) on Terminal-Bench 2.0 & 4.0Affine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (89 / 89 (100.0%))0.000
- JSON-RPC 2.0 Streamable-HTTP MCP Closure (%) on MCP-UniverseAffine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (480 / 480 (100.0%))100.000
- Pass^8 Rerun Flip Variance (%) on MCP-UniverseAffine.Earth Sovereign 4-Point HAC Audit & Public Proof Bundle (480 / 480 (100.0%))0.000