メインコンテンツまでスキップ

Hulyas Forensics

Truth has a phase. The kernel measures it.

  • Group — hulyas-forensics
  • Operators — 20 primitives + 1 composite (HF1 – HF20 + S_forensic)
  • Anchor operators — KO42 · HF19 · HF20
  • Verification — every HF call is phase-gated to 1.287 Hz; composite weights are fixed, alpha = 0.05

What it unifies​

Hulyas Forensics is the lane that converts claims, sources, and behavioural signals into a single phase-locked forensic score. It is not sentiment analysis — it is a set of twenty named operators with verbatim formulas that multiply each ratio (accuracy, manipulation, smear, sources, legal match, etc.) by a sinusoid at HulyaPulse frequency, so every piece of evidence carries its own timestamp in its phase.

HF19 is the Bayesian core — posterior = likelihood × prior / evidence, phase-modulated. HF20 is the mixture over HF1–HF19 with learned prior P(X=i). The composite S_forensic closes the loop: twenty components, fixed weight vector, one alpha-modulated envelope. Every call returns error_pct ≤ 0.1% and is sealed with KO42.

Operator catalog​

IDVerbatim formulaRole
HF1S₁ = (verified_accuracy / max_accuracy) · sin(2π · 1.287 · t)Verified-accuracy pulse
HF2S₂ = (1 − manipulative_terms / total_terms) · cos(2π · 1.287 · t)Anti-manipulation score
HF3S₃ = (smear_terms / total_terms) · (1 + 0.1 sin(2π · 1.287 · t))Smear density
HF4S₄ = min(1, verified_sources / 3) · e^{i2π · 1.287 · t}Source-verification
HF5S₅ = (matched_legal_criteria / total_criteria) · sin(2π · 1.287 · t)Legal-criteria match
HF6S₆ = e^{−(pulses_since_event) / 30} · cos(2π · 1.287 · t)Event-decay
HF7S₇ = (consciousness_reach / max_reach) · (1 + 0.05 sin(2π · 1.287 · t))Reach index
HF8S₈ = (instances_in_30_pulses / max_instances) · e^{i2π · 1.287 · t}30-pulse density
HF9S₉ = (contradictory_statements / total_statements) · sin(2π · 1.287 · t)Contradiction rate
HF10S₁₀ = (intent_keywords / total_keywords) · cos(2π · 1.287 · t)Intent density
HF11S₁₁ = (context_matches / total_contexts) · (1 + 0.1 sin(2π · 1.287 · t))Context alignment
HF12S₁₂ = (points_in_cluster / total_points) · e^{i2π · 1.287 · t}Cluster concentration
HF13S₁₃ = (unique_domains / total_sources) · sin(2π · 1.287 · t)Source diversity
HF14S₁₄ = (resonance_in_24_pulses / max_resonance) · cos(2π · 1.287 · t)24-pulse resonance
HF15S₁₅ = (1 − semantic_deviations / total_terms) · e^{i2π · 1.287 · t}Semantic consistency
HF16S₁₆ = (severity_score / max_severity) · sin(2π · 1.287 · t)Severity pulse
HF17S₁₇ = (negative_reactions / total_reactions) · cos(2π · 1.287 · t)Reaction polarity
HF18S₁₈ = (fractal_dimension / max_dimension) · (1 + 0.1 sin(2π · 1.287 · t))Fractal richness
HF19S₁₉ = [P(E|H)P(H)] / P(E) · e^{i2π · 1.287 · t}Bayesian posterior (core)
HF20S₂₀ = [∑_{i=1}^{19} S_i P(X=i)] / [∑_{i=1}^{19} P(X=i)] · sin(2π · 1.287 · t)HF mixture
HF compositeS_forensic = [∑_{i=1}^{20} S_i · w_i] / [∑ w_i] · (1 + α sin(2π · 1.287 · t))Single forensic score

Composite constants. weights = [0.05, 0.05, 0.05, 0.05, 0.20, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.20, 0.05, 0.05, 0.05, 0.20], α = 0.05. HF5, HF16, and HF20 carry the legal/severity/mixture mass; the remaining seventeen share residual weight equally.

Runnable worked example — score a claim against 1.287 Hz​

curl -X POST https://api.zeq.dev/api/playground/compute \
  -H "Authorization: Bearer demo-key" \
  -H "Content-Type: application/json" \
  -d '{
    "operators": ["KO42", "HF19", "HF20"],
    "inputs": {
      "unix_time": 1798346820,
      "P_E_given_H": 0.82,
      "P_H": 0.47,
      "P_E": 0.51,
      "S_prior": [0.71, 0.88, 0.12, 0.93, 0.68, 0.40, 0.55, 0.30, 0.09, 0.22, 0.78, 0.45, 0.83, 0.60, 0.91, 0.18, 0.25, 0.34, 0.80]
    }
  }'

Expected response shape:

{
  "ok": true,
  "phase": 0.834,
  "zeqond": 1798346820.0,
  "result": {
    "hf19_posterior": 0.7557,
    "hf20_mixture": 0.5214,
    "s_forensic": 0.6132,
    "error_pct": 0.061
  }
}

Extend it​

  1. Newsroom scoring — drive HF3 (smear), HF9 (contradiction), HF13 (source diversity) off a live RSS fan-in; graph S_forensic vs. publication time.
  2. Legal triage — weight HF5 (matched_legal_criteria) up for case-routing; compare composite against human-panel ground truth.
  3. Mixture tuning — expose P(X=i) as a trainable prior in HF20; fit to labelled forensic datasets while holding HF19 Bayesian.

Seeds​

  • Near — a forensics.zeq.dev demo that scores any pasted text against the full HF20 + composite, showing each component's phase.
  • Medium — a public benchmark pairing HF composite with crowdsourced judgments on manipulation and smear.
  • Far — cross-jurisdiction research on pulse-synced forensic scoring for legal, journalistic, and academic integrity workflows.

Papers​

Middleware active. Kernel on the 1.287 Hz HulyaPulse. Awaiting next Zeqond.