LOCAL AI LAB / OLARES ONE
● RUN COMPLETE MILESTONE 02 · 06–07 SEP 2026 · FULL TRAIL · HUB
NEMOTRON 3 NANO OMNI 33b-q4_K_MRTX 5090 24GB // 96GB RAM 82% GPU PROCESSOR SPLIT23.8 GiB VRAM USED COLD 48.8s // WARM 11.6s9/9 READABLE FIELDS MRZ NOT TRUSTED AUTOMATICALLYFOL DECISION request_better_evidence SWI-PROLOG POLICY GATESYNTHETIC DATA ONLY NEMOTRON 3 NANO OMNI 33b-q4_K_MRTX 5090 24GB // 96GB RAM 82% GPU PROCESSOR SPLIT23.8 GiB VRAM USED COLD 48.8s // WARM 11.6s9/9 READABLE FIELDS MRZ NOT TRUSTED AUTOMATICALLYFOL DECISION request_better_evidence SWI-PROLOG POLICY GATESYNTHETIC DATA ONLY
MODULE 01 · PROVENANCE CHAIN · RESEARCH QUESTION

NEMOTRON × EVIDENCE GATE.

A local multimodal model read the passport. Then the deterministic layer refused to believe it. Can one local model be the perception layer of an evidence-gated compliance agent — without becoming the final authority?

NVIDIA NEMOTRON 3 NANO OMNI RTX 5090 24GB 96GB RAM OLLAMA 0.30.11 SWI-PROLOG / FOL SYNTHETIC DATA ONLY
THE MODEL PERCEIVES. CODE VERIFIES. LOGIC DECIDES. THE MODEL ONLY EXPLAINS.
FOL-GATE · POLICY CONSOLELIVE
verification(passport_number_format, pass). verification(mrz_line_1_structure, fail). verification(mrz_line_2_structure, pass). verification(date_of_birth_check, fail). verification(expiry_date_check, fail). verification(composite_check, fail). verification(nationality_vs_mrz, conflict). verification(sex_vs_mrz, conflict). decision(request_better_evidence) :- \+ automated_acceptance. ?- decision(D), reasons(R). D = request_better_evidence, R = [ cross_field_conflict, deterministic_failure, identity_conflict, mrz_integrity_failed ]
VERDICT REQUEST_BETTER_EVIDENCEAI OVERRIDE: FALSE
01

Test environment

TIGHT LOCAL FIT

▚ HARDWARE

Olares One · NVIDIA GeForce RTX 5090 Laptop GPU · 24,463 MiB VRAM · ~96 GB installed RAM (93 GiB visible).

▚ RUNTIME

Ollama 0.30.11 in Olares / Kubernetes · Nemotron 33b-q4_K_M · payload ~27.6 GB · OpenAI-style local HTTP workflow.

▚ FORMAL LAYER

Python deterministic validator · ICAO-style MRZ length/charset · check digits + cross-field comparisons · SWI-Prolog policy engine.

▚ WHY NEMOTRON / WHY WE REFUSED TO TRUST IT

One local model for text + vision keeps sensitive evidence on-device. But in regulated workflows a plausible answer is not evidence — the experiment was designed so a deterministic layer could reject a confident but malformed extraction.

82%GPU SHARE · OLLAMA SPLIT
23.8GIB VRAM USED
48.8sCOLD TEXT RUN
11.6sWARM TEXT RUN
12–20sVISION RUNS · RANGE

Not a “huge headroom” deployment: a tight local fit with CPU/RAM spillover (18% CPU / 82% GPU).

02

Architecture — four layers, four jobs

AUTHORITY SEPARATION
01 · PERCEPTIONNEMOTRON

“What does the image appear to contain?”

Extract human-readable fields + candidate MRZ strings from a synthetic passport image.

02 · VERIFICATIONPYTHON / ICAO

“Does the extraction survive deterministic checks?”

Length, charset, MRZ check digits, cross-field consistency, uncertainty sanity.

03 · DECISIONPROLOG / FOL

“Given verified facts, what policy action follows?”

The validator exports facts; the policy engine reasons over facts — not over model prose.

04 · EXPLANATIONNEMOTRON

“Tell the reviewer what happened.”

Explicitly forbidden from changing, softening or reversing the decision.

03

T1 — clean synthetic passport

HIGH-CONTRAST · KNOWN GROUND TRUTH

▚ WHAT WORKED

9/9 human-readable fields extracted correctly. The model separated the printed passport number from the MRZ check digit. Structured JSON via Ollama API.

▚ WHAT FAILED

MRZ was not reproduced exactly — characters added/dropped in machine-readable lines — while reporting uncertain_fields: [].

▚ API TIMING SNAPSHOT

total_duration ≈ 19.65 s load_duration ≈ 15.52 s prompt_eval_duration ≈ 1.23 s eval_duration ≈ 2.83 s
04

T2 — degraded evidence

BLUR · ROTATION · JPEG · OCCLUSION

▚ ADVERSARIAL FINDING: CONFIDENT RECONSTRUCTION

Instead of returning null and marking fields uncertain, Nemotron attempted to reconstruct unreadable fragments — malformed values, still declaring uncertain_fields: [].

{ "passport_number": "L8989U<..<.5", "mrz_line_2": "L898902C36UT0900312: :<<<10", "uncertain_fields": [] }
12.17sT2 WALL CLOCK
FAILUNCERTAINTY CALIBRATION
USEFULAS CANDIDATE EXTRACTOR
05

T1b — formally valid control specimen

44-CHAR MRZ · VALID CHECK DIGITS

▚ THE INTERESTING PART: THE MODEL STILL FAILED THE MACHINE-READABLE LAYER

PASS passport_number_format
FAIL mrz_line_1_structure (model returned 42 chars)
PASS mrz_line_2_structure (44 chars)
PASS document_number_check
FAIL date_of_birth_check
FAIL expiry_date_check
FAIL optional_data_check
FAIL composite_check
CONFLICT nationality / sex / DOB / expiry vs MRZ
FAIL uncertainty_calibration

Deterministic verdict: NEED_MORE_EVIDENCE — 9/9 readable fields, but no MRZ trusted automatically.

06

Final loop — the model explains, but does not decide

FOL → HUMAN
{ "decision": "request_better_evidence", "explanation": "Human-readable fields were successfully extracted, but formal MRZ validation and cross-field consistency checks failed due to integrity and identity conflicts.", "next_action": "Request a new, valid evidence submission from the applicant." }
Nemotron perceives. Code verifies. Logic decides. Nemotron explains.THE MODEL IS USEFUL WITHOUT BEING SOVEREIGN
07

What this actually proves

HONEST BOUNDARIES

▚ PROVEN IN THIS PILOT

Nemotron 3 Nano Omni Q4_K_M runs locally on the Olares One · 24GB RTX 5090 hosts most of the model · vision extraction works via local API · human-readable fields extracted accurately on simple synthetic specimens · deterministic MRZ/cross-field checks catch model errors · Prolog deterministically refuses automated acceptance · the model explains the policy result without changing it.

▚ NOT PROVEN

Production-grade OCR across real passports · reliable uncertainty calibration · legal/regulatory compliance · document authenticity, fraud, liveness or biometrics · that Prolog rules faithfully encode real law · audio/video paths production-ready · anything involving real customer PII — none was used.

▚ WHY IT MATTERS · PRIVACY

Local multimodal perception keeps sensitive evidence on-device while exposing a structured interface to downstream controls.

▚ WHY IT MATTERS · AUDITABILITY

Checksums, structure rules and explicit conflicts leave a trace that is reproducible without trusting model confidence.

08

Experiment log

06–07 SEP 2026
SEP 06
Nemotron pulled into Olares Ollama storage (~27 GB tag; ~27.6 GB API size).
SEP 06
Cold/warm runtime measured. ~82% GPU / 18% CPU split; ~23.8 GB VRAM used.
SEP 07
Synthetic passport generated. Clean extraction: 9/9 readable fields, MRZ errors, no declared uncertainty.
SEP 07
Degraded document test: model reconstructed unreadable fragments, still reported no uncertainty.
SEP 07
Deterministic MRZ / cross-field validator added; malformed evidence routed to NEED_MORE_EVIDENCE.
SEP 07
SWI-Prolog policy layer added; final decision = request_better_evidence.
SEP 07
Nemotron explanation layer completed the loop without overriding the FOL decision.

Notes & caveats. Home-lab engineering experiment, not a benchmark paper and not a compliance certification. Synthetic passport data only. The deterministic checks validate properties of extracted data; they do not prove a document is genuine or a person is who they claim to be. Next: audio/video evidence → gate; realistic specimen scans; a versioned policy library; perception, verification, logic and explanation as separate components — not one trusted black box. This page is Module 01 of the chain that ends at AuthorityClaw.