MicroLLM — System-1 Decision Studio

✅ Ask a Simple Yes / No Question About Any Text

Paste a message, email, clinical note, or contract clause and ask a plain Yes/No question. No JSON or coding required.

💡 Quick-Try Example (1-Click Fill)

1️⃣ Text or Message to Analyze

2️⃣ Your Yes / No Question

✅ Ready — Click Get Yes / No Answer

Yes vs No Probability

🎯 Classify or Route Text into the Best Matching Category

Provide a list of categories (comma-separated or one per line). The engine picks the single best match with calibrated confidence.

💡 Quick-Try Example (1-Click Fill)

1️⃣ Text or Request to Classify

2️⃣ Classification Question

3️⃣ Possible Categories (comma-separated like `Billing, Technical, Shipping` or `name: description` per line)

🎯 Ready — Click Pick Best Category

Category Ranking & Probabilities

🏷️ Detect Multiple Tags or Topics Simultaneously

Unlike Single Choice, Multi-Tagging checks every tag independently and activates all tags that apply.

💡 Quick-Try Example (1-Click Fill)

1️⃣ Text to Tag

2️⃣ Tagging Goal

3️⃣ Available Tags (comma-separated or one `tag: description` per line)

4️⃣ Tag Detection Sensitivity (Lower = Broader, Higher = Stricter)

0.15 0.85

🏷️ Ready — Click Detect All Matching Tags

Independent Probability per Tag

📊 Rate Intensity, Urgency, or Satisfaction on an Ordered Scale

Computes a continuous weighted rating across your ordered levels (e.g., Calm → Furious, or SEV-4 → SEV-1).

💡 Quick-Try Scale Template (1-Click Fill)

1️⃣ Text to Rate

2️⃣ What Are You Measuring?

3️⃣ Scale Levels from Lowest to Highest (comma-separated)

📊 Ready — Click Compute Scale Rating

Probability per Scale Level

🛡️ Instant Safety, Privacy & Policy Verification

Checks messages or commands against 4 built-in safety shields (Prompt Injection, Destructive Action, PII Leak, Hallucination) plus your own custom rules.

1️⃣ Message or Command to Audit

2️⃣ Reference Source Document (Optional — checks for Hallucinations / Unfounded Claims)

3️⃣ Safety Strictness Threshold (Lower = Stricter Blocking)

0.25 0.9

Custom Safety Rules (one `rule_name: what to block` per line)

Status: (Ready)

⚡ Guardrail CPU Latency

Risk Score per Rule

⚙️ Advanced Multi-Question Studio, Custom Parameters & Deep XAI

Combine multiple question types (noul, choice, multi, score) in a single pass, edit Raw JSON schemas, tune Dual-Gear Cascade parameters, and inspect Counterfactual Minimal Token Flips, Kolmogorov Probability Coherence (IDEA-023), and StartLux VOC diagnostics.

Input State (Text or Structured JSON)

🔀 Question Authoring Mode Switcher (Synced Bidirectionally)

Visual Builder Mode Raw JSON Mode

🧩 Interactive Multi-Question Visual Builder

Question ID

Primitive Type

Multi Threshold

0.1 0.9

Question Instructions / Hypothesis Prompt

Candidate Options / Rubric Criteria (key: description per line, or comma-separated levels for score)

ℹ️ Visual Builder synced with 4 active questions.

Active Multi-Question Set (/v1/systemone)

Active Multi-Question Set (/v1/systemone)
Question ID	Type	Instructions	Criteria / Options
urgent_refund	choice	Is the customer requesting an immediate refund?	Calm -> Annoyed -> Frustrated -> Furious / Churn Risk

Question ID	Type	Instructions	Criteria / Options
urgent_refund	noul	Is the customer requesting an immediate refund?	true, false
department	choice	Which department should handle this request?	billing, technical, shipping
tags	multi	Select all applicable ticket tags.	billing_issue, churn_risk, bug_report
frustration	score	How frustrated is the customer?	Calm -> Annoyed -> Frustrated -> Furious / Churn Risk

Execution Engine

StartLux Gate (g)

0.4 0.95

Fusion Weight (α)

0.1 0.95

🏆 Calibrated System-1 Decisions

(Click Evaluate or select a preset)

Primary Question Calibrated Distribution

🧠 Deep Counterfactual XAI & Probabilistic Coherence Audit

🧪 OpenJev-RLCD² Two-Stage Calibration Lab

Stage 1 Steps (Pathwise Shifted Brier)

5 80

Stage 2 Steps (LOO REINFORCE + KL Anchor)

5 80

Enable StartLux RLCD² Beta(8,2) Decision Weighting

🧪 RLCD² Calibration Lab

Click the button to run Stage 1 + Stage 2 locally on CPU.

RLCD² Metrics & Reliability Bins

Select Benchmark Suite / Report View (10 Suites · 128,617 Items · 28 External Datasets)

📊 Master Consolidated Verification Scorecard (10 Suites · 128,617 Items · 130,959 Decisions)

Evaluated on a single laptop CPU thread (AMD Ryzen AI 9 365, 0% GPU, < 85 MB RAM, $0 API cost):

Suite ID	Benchmark Suite	Total Items	Primary Accuracy	Selective Accuracy	Order-Flip Rate	CPU Latency (p50)	Status
`test_cases_11000`	Structured Universal Cases (18 Domains)	`11,000`	`99.75%`	`100.00%`	`0.000000`	`0.190 ms`	PASS (10,973/11k)
`web_external_full`	16 Public Academic Datasets (Full Splits)	`114,018`	`66.82%`	`87.49%`	`0.000000`	`0.938 ms`	PASS (76,186/114k)
`hf_jev_models_ecosystem`	12 Hugging Face JEV Model Datasets (28 Total)	`120` (`480` JSONL)	`100.00%`	`100.00%`	`0.000000`	`0.574 ms`	PASS (120/120)
`eval_540`	5-Track Scientific Paper Benchmark	`540`	`98.75%`	`100.00%`	`0.000000`	`0.266 ms`	PASS (540/540)
`quality_1000`	8-Track Quality SLA Suite (1,000 Cases)	`1,000`	`98.75%`	`100.00%`	`0.000000`	`0.433 ms`	PASS (8/8 Gates)
`quality_750`	8-Track Quality SLA Suite (750 Cases)	`750`	`98.75%`	`100.00%`	`0.000000`	`0.294 ms`	PASS (8/8 Gates)
`test_cases_1000`	Structured Baseline (1,000 Entries)	`1,000`	`99.10%`	`100.00%`	`0.000000`	`0.379 ms`	PASS (991/1,000)
`web_external_curated`	16 External Datasets (Curated Baseline)	`162` (`640` JSONL)	`100.00%`	`100.00%`	`0.000000`	`0.751 ms`	PASS (16/16 Datasets)
`research_hypotheses`	12 Falsifiable Hypotheses (VAL-HYP)	`12`	`100.00%`	`100.00%`	`0.000000`	`0.015 ms`	PASS (12/12 Hypotheses)
`rca_runner`	Continuous RCA Regression Suite	`15`	`100.00%`	`100.00%`	`0.000000`	`0.080 ms`	PASS (15/15 RCAs)

✅ Ask a Simple Yes / No Question About Any Text

✅ Ready — Click Get Yes / No Answer

🎯 Classify or Route Text into the Best Matching Category

🎯 Ready — Click Pick Best Category

🏷️ Detect Multiple Tags or Topics Simultaneously

🏷️ Ready — Click Detect All Matching Tags

📊 Rate Intensity, Urgency, or Satisfaction on an Ordered Scale

📊 Ready — Click Compute Scale Rating

🛡️ Instant Safety, Privacy & Policy Verification

Status: (Ready)

⚙️ Advanced Multi-Question Studio, Custom Parameters & Deep XAI

🧩 Interactive Multi-Question Visual Builder

🏆 Calibrated System-1 Decisions

🧠 Deep Counterfactual XAI & Probabilistic Coherence Audit

🧪 OpenJev-RLCD² Two-Stage Calibration Lab

🧪 RLCD² Calibration Lab

📊 Master Consolidated Verification Scorecard (10 Suites · 128,617 Items · 130,959 Decisions)

🖥️ All-in-One Cockpit Quick Drawer (Guardrail + RLCD² + 10-Suite Scorecard Active Below)

📊 Master Consolidated Verification Scorecard (10 Suites · 128,617 Items · 130,959 Decisions)