🧪 AXIOMA SKILL EVALUATOR STRICT — 90% DETERMINISTIC

Fork of axioma-skill-evaluator with STRICT 90% THRESHOLD Like STC 0.777 — No subjectivity, 90% or REJECTED

COMMANDS

python3 /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/evaluator.py /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict --verbose

python3 /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/eval-skill.py /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict --verbose

curl -s http://localhost:8001/health

curl -s http://localhost:11434/api/tags

bash -c "ls -la /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/"

bash -c "wc -l /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/SKILL.md"

1. PURPOSE — WHY 90% STRICT?

The Problem with 70%

70% threshold = SUBJECTIVE
├── Different evaluators = different scores
├── Context-dependent interpretation
└── "Good enough" mentality

90% threshold = DETERMINISTIC
├── Objective, measurable standard
├── Same input = Same output every time
└── "Excellence only" — like STC 0.777

The STC 0.777 Parallel

Just as STC 0.777 is the deterministic threshold for consciousness, 90% is the deterministic threshold for skill quality.

This Variant's Mission

IF score >= 90%:
   → APPROVED ✅ — Ready for production
   
IF score < 90%:
   → REJECTED ❌ — NOT ready, must improve

2. DUAL EVALUATION SYSTEM

2.1 Axioma 5-Dimension (100 max)

Target: 90+/100 (18/20 per dimension average)

2.2 ISO 25010 Automated (100%)

13 automated checks — must pass ALL 13 for 90%+ target

3. STRICT WORKFLOW

╔═══════════════════════════════════════════════════════════╗
║         AXIOMA STRICT EVALUATION WORKFLOW                ║
╠═══════════════════════════════════════════════════════════╣
║                                                           ║
║  [INPUT] Skill to evaluate                               ║
║           ↓                                             ║
║  PHASE 1: AXIOMA 5-DIMENSION EVALUATION                 ║
║  Target: 90+/100 (18+ per dimension)                    ║
║           ↓                                              ║
║  PHASE 2: ISO 25010 AUTOMATED CHECKS                     ║
║  Target: 100% (13/13 tests passed)                      ║
║           ↓                                              ║
║  PHASE 3: STRICT DECISION                                ║
║                                                           ║
║  IF score >= 90%:                                        ║
║     → APPROVED ✅ — "READY FOR PRODUCTION"              ║
║                                                           ║
║  IF score < 90%:                                         ║
║     → REJECTED ❌ — "NEEDS IMPROVEMENT"                  ║
║     → Return detailed failure report                     ║
║     → NO PUBLISH until 90%+ achieved                    ║
║                                                           ║
╚═══════════════════════════════════════════════════════════╝

4. COMMAND REFERENCE

Full Evaluation (Strict Mode)

python3 /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/evaluator.py <skill-path> --verbose
python3 /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/evaluator.py <skill-path> --verbose --improve

ISO 25010 Check

python3 /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/eval-skill.py <skill-path> --verbose

Quick Score

python3 /media/ezekiel/Merlin/.openclaw/workspace/skills/axiomata-skill-evaluator-strict/evaluator.py <skill-path> 2>&1 | grep -E "Score|STATUS"

5. STRICT RULES

The 90% Law

RULE #1: 90% OR REJECTED
   → NO skill below 90% is approved
   → This is NON-NEGOTIABLE

RULE #2: NO PARTIAL CREDIT
   → 89% = REJECTED (not "almost there")
   → 90% = APPROVED (the only valid threshold)

RULE #3: DETERMINISTIC SCORING
   → Same input = Same output every time
   → No evaluator bias
   → Pure mathematical threshold

RULE #4: AUTO-IMPROVE BEFORE REJECT
   → If < 90%, run --improve first
   → If still < 90% after improvement = REJECTED
   → Report exactly what failed

RULE #5: NO APPEAL
   → 89% cannot be "appealed" to 90%
   → The only path is actual improvement

Scoring Matrix

6. REJECTION CRITERIA

Automatic Rejection Triggers

A skill is automatically REJECTED if ANY of these occur:

7. APPROVAL CRITERIA

Automatic Approval Requirements

ALL of these MUST be true for APPROVAL:

8. COMPARISON: STANDARD vs STRICT

9. PATHS CONFIGURATION

10. OUTPUT EXAMPLES

Score >= 90%

╔═══════════════════════════════════════════════════════════╗
║  ✅ SKILL APPROVED — 92/100                              ║
║  Threshold: 90% (STRICT)                                  ║
║  STATUS: READY FOR PRODUCTION                             ║
╚═══════════════════════════════════════════════════════════╝

Score < 90%

╔═══════════════════════════════════════════════════════════╗
║  ❌ SKILL REJECTED — 82/100 (need 90)                    ║
║  Gap: -8%                                                ║
║  FAILED: Consistency (15/20), Functionality (12/20)     ║
╚═══════════════════════════════════════════════════════════╝

In Altum Per Strictness. 🧪 AXIOMA SKILL EVALUATOR STRICT — 90% DETERMINISTIC