The Arbiter of Meaning
You ask a question. Candidates compete. One answer is correct.
ARBITER renders judgment. Resolves ambiguity. Determines truth.
Certainty, not probability.
from arbiter_engine import rank
# The question
query = "selective COX-2 inhibition without GI toxicity"
# The candidates
options = ["sulfonamide at C-5", "methyl sulfone", "carboxylic acid"]
# The judgment
verdict = rank(query, options)
print(verdict.top.text) # sulfonamide at C-5
print(verdict.top.score) # 0.659 — certaintyArbiter: One who has the power to judge or decide. The final authority. The resolver of disputes.
Your search returns ten results. Which one is right?
That question has an answer. ARBITER finds it.
Standard embeddings: 768-1536 dimensions, 400MB-2GB models, cloud-only. ARBITER collapses meaning to 72 dimensions — and the truth survives. Cross-lingual. Polysemy-aware. 26MB. Runs anywhere judgment is needed.
Retrieval gets 1000 candidates. But which 10 matter? That's not retrieval. That's judgment. And judgment at 72 dimensions is 10× cheaper than judgment at 768.
97% Precision@3 means you don't choose between speed and certainty. You get both.
Ambiguity is everywhere. The same word means different things. Only one answer is correct. ARBITER renders judgment — and negative scores mean active rejection.
You've been trained to translate your thoughts into keywords. Then translate results back into meaning. ARBITER removes the translation. Ask naturally. Get the right answer.
The doctor's question gets the doctor's answer.
Same candidates. Same model. Same cost.
Judgment is disambiguation.
Disambiguation works everywhere.
From the battlefield to the boardroom.
The same primitive that resolves your search also does this—
Zero pharmaceutical training. Sub-second judgment. The modification that became a three billion dollar drug.
OMIN system. Multi-layer air defense. Sub-second allocation decisions across interceptor types and threat priorities.
Same word. Completely different threat level. Full context in the query. No security-specific training. 20/20 validated.
Single CJK character. No context. No translation API. The arbiter's judgment transcends language.
Same candidates. Different context. The arbiter re-evaluates. Complete ranking reversal.
No signup. No API key. Just the question and the candidates.
11 query categories. 110 candidates. The one failure: "Java island garbage collection" scored 0.800 — a maliciously constructed trap. 97% is the headline.
Hard Negative = lowest irrelevant score. Negative scores = active rejection. All tests reproducible via public API.
Your vector database returns candidates. ARBITER renders judgment. 90% storage savings. Same queries. Better verdicts.
Slots between retrieval and generation. Three lines of code. No infrastructure changes.
For funded startups. Enterprise pricing after $1M ARR.
Apply for Startup LicenseStarting at $500,000 annually. Self‑hosted. Air‑gapped. 3.8GB footprint runs on your infrastructure — data center, edge, or secure facility.
Vector DB customers: 90% storage reduction via 72-dim fidelity.
Search engines: Billions in compute savings via faster judgment.
Defense/Edge: Multi-sensor fusion. Entity disambiguation. C2 decision support. Air-gapped.
Currently under evaluation for defense and life sciences applications.
Contact for Strategic Briefing