What It Is
SCIS — the SEO Continuous Intelligence System — is the recommendation engine at the heart of Korvex. It continuously analyses all your SEO data (rankings, content scores, entities, technical health, competitor data) and generates prioritised, evidence-based recommendations. Unlike static audits, SCIS updates its recommendations daily as new data arrives, adjusting confidence scores and re-ranking priorities.
Why It Matters for Your SEO
Manual SEO audits are snapshots — they go stale within weeks. SCIS operates as a continuous feedback loop: it generates hypotheses, collects evidence, scores confidence, predicts impact, and learns from outcomes. This means your recommendation queue is always fresh, always prioritised, and always improving its accuracy based on what actually worked.
How Korvex Measures It
SCIS operates through 6 stages:
| Stage | Function | Output |
|---|---|---|
| 1. Hypothesis Generation | 18 specialised generators create testable recommendations | Raw hypotheses with initial confidence |
| 2. Evidence Collection | Polls all data sources for supporting/contradicting evidence | Evidence records with source attribution |
| 3. Confidence Scoring | Rates 0-100 based on evidence quality and quantity | Calibrated confidence score |
| 4. Impact Prediction | Models cascading effects of implementation | Estimated ROI multiple |
| 5. Continuous Refinement | Updates scores as new data arrives daily | Adjusted priorities |
| 6. Winner Selection | Surfaces high-confidence, high-ROI recommendations | Sorted recommendation queue |
The 18 Hypothesis Generators
SCIS includes specialised generators grouped by tier:
Foundation (9 generators): Technical SEO, Site Health, HTML Structure, Internal Linking, Content Quality, Ecommerce Product Schema, Collection Optimiser, Product Descriptions, Sitemap Audit
Growth (5 generators): GSC Opportunity Mining, Sentence-Level Optimisation, Specificity Mismatch Detection, Content Decay Recovery, Cannibalisation Resolution
Authority (6 generators): Koray V2 Advanced Analysis, Strategic Content Planning, Content Brief Generation, Hub-Pillar Classification, Long-tail Opportunity Mining, Entity Question Generation
How to Improve Your Score
- Act on high-confidence recommendations first — items above 70% confidence have strong evidence
- Check the Strategy Actions page daily — SCIS surfaces new opportunities as data arrives
- Complete Foundation-tier recommendations before Growth — the tier system ensures fundamentals are solid
- Review dismissed recommendations periodically — new evidence may have changed the picture
- Track outcomes — implementing recommendations feeds the closed loop, improving future accuracy
Confidence Delta Rules
Evidence from different sources adjusts confidence by fixed deltas:
| Source | Condition | Delta |
|---|---|---|
| Page scores | Per technical issue found | +2.5 (max +20) |
| GSC data | High impressions (>100) + low CTR (<2%) | +10 |
| GSC data | Cannibalisation confirmed | +15 |
| Winning patterns | Success rate ≥ 80% | +20 |
| Winning patterns | Success rate 60-79% | +10 |
| Competitor data | Entity gap ratio ≥ 2.0x | +15 |
| Competitor data | Entity gap ratio 1.5-1.99x | +10 |
| Gap analysis | Severity = critical | +15 |
| Gap analysis | Severity = high | +10 |
| Outcome patterns | 10+ implementations, 80%+ success | +20 |
| Outcome patterns | 5+ implementations, <40% success | -15 |
Winner Selection
- Minimum confidence threshold: 70.0
- Sort:
confidence_score × COALESCE(estimated_roi_multiple, 1.0)descending - Default confidence for new recommendations: 50.0
- Calibration applied via
confidence_calibratormodule using historical outcome data
Recommendation Categories
technical_seo, content, keywords, links, performance, entities
Data Sources
- All daily phases: SCIS draws evidence from every data collection phase
- Tables:
page_recommendations(output), with evidence stored as JSONB - Update frequency: Daily SCIS cycle runs after Phase 5 scoring completes
Related Concepts
- The Tier System — which generators are active depends on your tier
- Confidence and Impact — how predictions are calculated
- The Closed Loop — how outcomes improve future recommendations