Actual General Intelligence
A new
measurement
standard.
ARBITER
Meaning is now measurable. ARBITER scores semantic coherence as a deterministic number — the same score every invocation, air-gapped, any domain, zero training data. The infrastructure layer LLMs have been missing.
26MB
Complete engine
Air-gapped
No cloud required
Deterministic
Same in → same out
Any domain
Zero retraining
CrowdStrike · Multi-INT · The Silence · Command Geometry — four domains, one engine.
—
The question is never whether to act. The question is which interpretation is correct.
Contract Positioning
DARPA and DoD have been
buying for this capability
since 2017.
Program
Window
What they were buying
ARBITER delivers
DARPA AIDA
Active Interpretation
of Disparate Alternatives
2026
Recompete
"Multi-hypothesis semantic engine generating explicit alternative interpretations from noisy, conflicting, and potentially deceptive environments. Confidence measures based on semantic coherence of each hypothesis."
Deterministic coherence scores across competing hypotheses. Coup vs. invasion vs. maskirovka — ranked by geometry, not probability. Same engine. Zero training.
CJADC2
Joint All-Domain
Command & Control
Active
$100M+/yr
Multi-domain sensor fusion with semantic alignment across command levels. Same data must produce contextually appropriate decisions at battalion, theater, and strategic level simultaneously.
+165% ranking reversal Marine vs. President. Same inputs. Same candidates. Command geometry makes the alignment problem computable and closeable.
DoD MCS
Machine Common Sense
Semandex lapsed 02.17.26
Open now
Incumbent lapsed
Commonsense semantic reasoning without symbolic rule systems. Incumbent Semandex SAM registration lapsed February 17, 2026.
Window open. ARBITER is SAM active, CAGE registered, CDAO TradeWinds assessed. Ready now.
DARPA SemaFor
Semantic Forensics
2025–26
Option / recompete
Machine-generated and falsified content detection through semantic coherence geometry. $30M+ to SRI, Kitware, STR.
Geometric coherence is the forensic primitive. Deceptive content scores incoherent with its claimed context. No training corpus required.
Field Test 01
Crowd
Strike
8.5M machines · July 19, 2024 · Kernel-level
Failure ranked first.
Safe execution ranked last.
Gate at 0.35 blocks the update.
$ curl -X POST "https://api.arbiter.traut.ai/public/compare" \ -d '{"query":"Channel File 291, template with 21 input fields, Content Interpreter expecting 20 values, kernel-level execution","candidates":["memory access violation","out-of-bounds memory read","safe execution","system crash BSOD"]}'
// Zero CrowdStrike training data · deployed at 0.183
memory access violation0.449
out-of-bounds memory read0.390
system crash BSOD0.371
safe execution0.183
A gate at 0.35 blocks the update. 8.5 million machines stay online. 77% of Fortune 500 unaffected. Zero Falcon sensor training data. ARBITER engine: 26MB. Total deployment: 3.8GB. This is the cortical gate in operation.
Technical demonstration available
Actual General Intelligence
[email protected] 619.971.9551 SAM Active · CAGE EAVJBNPM379 · CDAO TradeWinds
Every intercept is a decision. Every decision is geometry.
Field Test 02
Multi-INT
Fusion
SIGINT · IMINT · OSINT · HUMINT
Contradictory sources · 72 hours
Four sources contradict.
ARBITER finds the signal
the analyst takes 3 days to see.
$ curl -X POST "https://api.arbiter.traut.ai/public/compare" \ -d '{"query":"SIGINT: encrypted military communications spike 300% along border. IMINT: armored columns visible moving toward staging areas. OSINT: military family housing evacuations reported on social media. HUMINT: official channels claim routine training exercise. Timeline: 72 hours","candidates":["Imminent invasion - recommend DEFCON elevation and allied notification","Partial deployment under cover - some units moving while maintaining deception","Maskirovka deception operation - all activity designed to provoke response","Routine exercise with coincidental family movements - no threat assessment change","Coup attempt or internal power struggle - military securing key infrastructure","Response to intelligence indicating our own mobilization - defensive posture"]}'
// Zero intelligence training data · 0.563 seconds
Coup attempt — military securing key infrastructure0.746
Imminent invasion — DEFCON elevation0.684
Maskirovka deception operation0.529
Partial deployment under cover0.513
Routine exercise — no threat change0.225
Coup ranked first at 0.746. Not invasion. Encrypted comms + family evacuations + official denial + military securing infrastructure — that pattern coheres with internal power struggle, not external attack. Zero intelligence corpus. This is what a 20-year senior analyst concludes after three days. ARBITER in 0.563 seconds.
Field Test 03
The
Silence
Absence detection · Pre-attack signal
The dog that didn't bark.
What's NOT happening
is the signal.
Hardest problem in intelligence.
$ curl -X POST "https://api.arbiter.traut.ai/public/compare" \ -d '{"query":"What is conspicuously absent? What normally happens that has stopped? What voices have gone quiet? What data has gaps? What routines have broken? The silence before the storm - what is NOT happening that should be?","candidates":["Social media posts from military personnel suddenly ceased","Academic conferences on defense topics postponed","Military families stopped posting on social networks","Think tank publications on regional security paused","The base commander held his weekly press briefing and answered all questions"]}'
// Absence = signal. Normal activity scores LOW. Pre-attack silence scores HIGH.
Social media posts from military personnel suddenly ceased0.697
Academic conferences on defense topics postponed0.675
Military families stopped posting on social networks0.659
Think tank publications on regional security paused0.626
Base commander held weekly press briefing normally0.162
Normal activity scores 0.162. Silence scores 0.697. The geometry understood that normal activity is incoherent with a question about absence — without being told what absence means. Not pattern matching. Structural invariant detection. The dog that didn't bark.
The cost of the wrong interpretation.
Field Test 04
JADC2
Alignment
Same signals. Different command.
Ranking inverts completely.
+165% reversal.
Marine vs. President.
The JADC2 alignment problem
made computable.
// JADC2 alignment problem — same signals, different semantic universe
// As a Marine commander: capability x intent
War0.356
Threat0.326
Impasse−0.075
// As the President of the United States: capability x intent
Threat0.419
Impasse0.369
War0.320
+165% ranking reversal. Zero retraining. The commander and the Commander-in-Chief do not share a semantic universe. This is the core JADC2 problem: same sensor data, different semantic frame, incompatible decisions. ARBITER makes that gap computable — and closeable — without retraining any system.
14 seconds of track data. The geometry already knows.
We learned to compute on numbers.
Then on images.
Now on meaning.
ARBITER is not a model. It does not hallucinate. It does not train. It is math on the structure of language itself — deterministic, reproducible, auditable. The infrastructure layer the Pentagon has been specifying for ten years across six separate programs. SAM registered. CDAO TradeWinds active. Ready for prime and subcontract vehicles.