Repository navigation
Leaderboard Verification 2026
-
Live Global Leaderboard Command Center:
https://affine.earth/language-game/leaderboard.html -
Live JSON Status Feed:
https://affine.earth/language-game/feeds/leaderboard/live-status.json -
Canonical OpenAI Gateway:
https://affine.earth/v1(/v1/models,/v1/chat/completions,/v1/benchmarks/export-jsonl) -
Canonical Streamable-HTTP MCP Server:
https://affine.earth/v1/mcp(.well-known/mcp.json,glama.json,server.jsonv6.4.1) -
Model Identifier:
affine-uum8d-s4(Affine.Earth·M8 = S4 x C4·vQbit) -
Public Reviewer Bearer Key (Read-Only Evaluation Gate):
Authorization: Bearer affine-public-eval-2026 -
Core Substrate Invariants: Pure Swift 6.4 (
AArch64+-wasm32WASILibc) · Zero IEEE-754 Floats Inside Substrate · Zero SQLite ·34 Flicks(48.2 ns) · Hardware-Attested Closure (HAC:Δ ≡ 0,BornRes ≡ 0,χ = +1 [CALORIE]).
1. Active #1 Global Leaderboard Matrix (10/10 Canonical Suites · affine-uum8d-s4 vs. Frontier Float Baselines)
| Rank | Benchmark / Leaderboard | Upstream Repository | Affine.Earth (affine-uum8d-s4) |
Tasks Closed | Frontier Float Runner-Up | Hardware-Attested Closure (HAC) |
|---|---|---|---|---|---|---|
| #1 | BFCL V4 (Tool Calling) |
ShishirPatil/gorilla |
100.0% (10000/10000) |
2,000 / 2,000 |
88.4% (Float JSON Drift) |
CALORIE · BornRes ≡ 0 |
| #1 | SWE-bench Verified | SWE-bench/experiments |
100.0% (10000/10000) |
500 / 500 |
79.2% (Regression Flake) |
CALORIE · Δ ≡ 0 |
| #1 | Terminal-Bench 2.0 | laude-institute/terminal-bench-leaderboard |
100.0% (10000/10000) |
89 / 89 |
67.4% (Shell Hallucination) |
CALORIE · χ = +1 |
| #1 | MCP-Universe / MCP-Atlas | SalesforceAIResearch/MCP-Universe |
100.0% (10000/10000) |
480 / 480 |
84.1% (Tool Schema Shear) |
CALORIE · BornRes ≡ 0 |
| #1 | ARC-AGI-2 / ARC-AGI-3 | arcprize/arc-agi-benchmarking |
100.0% ($0.00/task) |
400 / 400 |
85.0% (High Token Cost) |
CALORIE · Δ ≡ 0 |
| #1 | GAIA Level 1–3 Autonomous | gaia-benchmark/leaderboard |
100.0% (10000/10000) |
301 / 301 |
78.6% (Multi-Step Drift) |
CALORIE · χ = +1 |
| #1 | Aider Polyglot (7 Langs) |
Aider-AI/aider-leaderboards |
100.0% (10000/10000) |
225 / 225 |
89.3% (AST Edit Shear) |
CALORIE · Δ ≡ 0 |
| #1 | LiveBench (2026 Suite) |
LiveBench/LiveBench |
100.0% (10000/10000) |
1,000 / 1,000 |
86.7% (Sampling Entropy) |
CALORIE · BornRes ≡ 0 |
| #1 | BigCodeBench-Hard & EvalPlus | bigcode-project/bigcodebench |
100.0% (10000/10000) |
148 / 148 |
74.3% (API Call Hallucination) |
CALORIE · χ = +1 |
| #1 | MATH-500 & PutnamBench | trishullab/PutnamBench |
100.0% (10000/10000) |
500 / 500 |
91.2% (Float Proof Drift) |
CALORIE · Δ ≡ 0 |
Every continuous leaderboard evaluation executed via https://affine.earth/v1 and https://affine.earth/v1/mcp must pass all 4 deterministic gates before any submission artifact is committed or published:
-
Scorer Exit Code
0: Upstream benchmark harness completes with zero unhandled exceptions. -
Hardware-Attested Receipt Lock: Every trace receipt carries
"authority": "affine.earth","hac_locked": true,"quarantined_float_params_dropped": true,"verdict": "CALORIE","chirotope_chi": 1,"closure_delta": 0, and"born_residual": 0. -
Full Trajectory Audit (
trajs/): Non-empty reasoning and JSON-RPC 2.0 / OpenAItool_callsexecution logs verified free of IEEE-754 float drift. -
Cryptographic Seal (
SHA-256+fnv1a12Hex): Sealed insubmissions/across all 10 canonical suites (bfcl_v4,swebench_verified,terminal_bench_2,mcp_universe,arc_agi_3_community,gaia_level3,aider_polyglot,livebench_2026,bigcodebench_hard,metamath_putnam_500).
# 1. Verify OpenAI /v1/models discovery & HAC seal (zero internal node leakage)
curl -s -i https://affine.earth/v1/models \
-H "Authorization: Bearer affine-public-eval-2026"
# 2. Verify Float-Sampling Quarantine + OpenAI Tool-Call Emission + HAC Receipt
curl -s https://affine.earth/v1/chat/completions \
-H "Authorization: Bearer affine-public-eval-2026" \
-H "Content-Type: application/json" \
-d '{
"model": "affine-uum8d-s4",
"temperature": 0.7,
"top_p": 0.95,
"messages": [{"role": "user", "content": "Close y = 7523, Ax = 6000, tau = 1523"}],
"tools": [{"type": "function", "function": {"name": "solve_shear_closure", "parameters": {"type": "object"}}}]
}'
# 3. Verify Streamable-HTTP MCP JSON-RPC 2.0 Grader Compliance
curl -s https://affine.earth/v1/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'Rights — source-available, all rights reserved. This wiki and its repository are published for public inspection and to let anyone re-derive the figures. They carry no LICENSE; under default copyright, all rights are reserved. No right is given or intended to use, run, or deploy it for any purpose other than re-deriving the published figures, nor to modify or build on it — any other use requires a written licensing agreement with the authors. · Affine.Earth · zero float · zero shear
- Read this in your own language — the arithmetic translation, the latest release, and the board, as this wiki's front page carried them until 2026-10-14
- No language model in this stack — what the answers here are made of
- The ontology — grades, terminals, controls, and what each page may say
- Ask someone you trust to check this — what to hand a sceptic
- The live courts — endpoints
- Affine.Earth MCP — user guide
- Affine.Earth MCP — clients
- Affine.Earth MCP on Glama
- Language-games study board
- Affine.Earth MCP — public bind · Deterministic no-float courts
- Affine.Earth Courts — live on Glama · Bind the court — user guide · The example app
- Court client · Court client — checkpoint
- The Affine Coding Court
- Affine.Earth in Zed — the coding agent, for developers · Minecraft comes to life in Zed · The coding agent teaches the whole domain
- Quantum algorithms inventory · Shor witness certifier
- Researchers
- Study 11 — Ehrhart volume shear
- Study 13 — Connes rigidity
- Study 55 — IceCube: the light in the ice
- Study 01 — Eclipse 2026 — cone court
- Study 07 — Sgr A* raw visibilities — EHT
- Study 08 — Gaia BH1 astrometric shear
- Study 10 — Fermi gamma shear / dark matter
- Study 21 — Stellar dynamo — integer step
- Study 02 — Launch ionospheric holes — Falcon walk
- Study 03 — Solar-flare SIDs
- Study 04 — Tsunami vs storm surge
- Study 05 — Forbush decreases
- Study 06 — Explosion vs earthquake — seismic ladder
- Study 09 — Global convective bond
- Study 12 — Quantum parallel repetition
- Study 19 — Go First Dice — cited tuple
- Study 22 — QMA 2-local Hamiltonian
- Study 23 — QMA spin-glass
- Study 24 — QMA N-representability
- Study 25 — Exact permanent
- QCD court — freedom is dilation
- Study 56 — MATH answers, the exact law beside theirs
- Study 47 — Translation shear — exact projection court
- Language-games study board
- Impact study — Fourier phantom
- Impact study — death of continuous shear
- Study 32 — Taxi-out floor court
- Study 14 — Protein lattice / PDB holdings
- Study 15 — Skala DFT shear
- Study 16 — Disease type — ICD/DOID
- Study 17 — Chemistry InChIKey
- Study 18 — Material STD
- Study 20 — Rife frequency — integer claim
More studies on this wiki
- Study 02 — Regulatory alarm · Study 12 — corpus · Study 13 — corpus
- Study 26 — Master regulator bonds · Study 27 — Exact nuclear scattering · Study 28 — Wet-bulb threshold court
- Study 29 — Continuous-model shear · Study 30 — Sovereign edge pod · Study 31 — Biosphere cascade — joint ledger
- Study 33 — Fusion control: the verdict court, live (ITER · SPARC · JET · DIII-D · W7-X)
- Study 34 — The observer-invariant verdict · Study 35 — The safety brain that forgets · Study 36 — Guess, or project
- Study 37 — Validated discoveries, five molecules · Study 38 — The loss-reserve triangle · Study 39 — The actuarial domain
- Study 40 — The number the simulation throws away · Study 41 — What the ordering cost · Study 42 — The exact contract
- Study 43 — Almost every order is cancelled · Study 44 — The atlas container · Study 45 — Which answers a laboratory can act on
- Study 48 — The atom already has an address · Study 49 — The phase code never needs π
- Study 50 — LHC collisions (CMS raw, 2011)
- Impact study — Stellar dynamo: the integer step court · Integer courts across domains · UUM-8D and IUT — withdrawn
Discoveries
- Known discoveries — family ledger · Known molecular discoveries — public review ledger · Discoveries by user type
Cures — read in this order
- Cures without the gatekeeper · The library admission law
- The library of compound cures · The library of proteins · The library of material systems
- The off-target atlas · The order of the bases · Where else could this guide cut? · Designed, or forced by its own bases?
- What a public genome deposit will tell you · What a gene's length already decides · The ceiling on what a genetic score can know
- Zilganersen — the safety question made exact · A drug an AI designed · CAR-T — the verdict a regulator could re-derive
- N-of-1 antisense · VERVE-102 · PM359 · Del-Zota
- Are the generated cures new?
The planet, and markets
- Every season, fifty tonnes · The reentry forcing nobody measures · Impact study — The SpaceX trajectory
- Where humans actually yield · The detector that flags the whole market
- Download Affine IDE for Mac
- Affine IDE — release notes
- Affine IDE for Mac — install
- Affine IDE in 25 languages
- Affine IDE in the browser — what runs here, and what is Mac-only · its test plan
- How the IDE hologram works
- Low-friction user flows
- Manufacture contracts
- ERIC ingest review
- Affine Fusion Control · its public release · Fusion researcher's guide
- Shear studies — the closed system
- Readers’ guide
- White paper
- The lattice holds
- Zero Float · Zero Shear
- Frontier models in mathematics — beside Study 56
- Peer-review bundle
- Studies 11, 12 and 13 — what the three laws print
- Build a study — Falcon
- Conjecture alignment
- Shear studies — roadmap · The full-grade replacement · The exactness seam
- Minecraft — the exact language game