Algorithm Knowledge Base
Purpose
Provides authoritative, confidence-rated platform algorithm intelligence to inform all Synthex content recommendations. Prevents generic advice by grounding recommendations in specific, verified ranking signals.
Loaded into: AI Marketing Advisor brief generation, Platform Content Optimiser skill, and any content strategy session.
Reference Files
Load the relevant platform file before making algorithm-specific recommendations:
| Platform | File | Confidence Range |
|---|---|---|
| Google Search | references/google-search.md |
CONFIRMED → INFERRED |
| Instagram / Meta | references/instagram-meta.md |
CONFIRMED → SPECULATIVE |
references/linkedin.md |
CONFIRMED → INFERRED | |
| Cross-Platform | references/cross-platform.md |
INFERRED |
| Signal Taxonomy | references/signal-taxonomy.md |
Internal standard |
Protocol
When generating content recommendations:
- Load
references/signal-taxonomy.md— establishes signal categories - Load platform-specific reference file(s)
- Map client's content to the top 3 ranking signals for that platform
- Cite only CONFIRMED or LEAKED signals when making specific claims
- Mark INFERRED recommendations with "likely" or "based on observed patterns"
Confidence label rules:
- [CONFIRMED] — Direct official statement from platform or verified documentation
- [LEAKED] — From authenticated leak (e.g., Google Content Warehouse 2024)
- [INFERRED] — Multiple corroborating sources (creators, analytics providers)
- [SPECULATIVE] — Single source or community consensus, low certainty
DO NOT:
- Invent signal weights or claim numerical precision not in the reference files
- Use outdated information (check
last_verified_datebefore citing) - Apply one platform's signals to another without checking cross-platform patterns
Output Format
When citing algorithm intelligence in recommendations:
Signal: watch_time [CONFIRMED — Mosseri 2023-09]
Recommendation: Keep Reels under 30s for service demos — watch completion
directly feeds distribution beyond your existing followers.
Pass / Fail Rules
A recommendation using this skill:
- ✅ PASS: cites at least one specific signal by name with confidence level
- ✅ PASS: distinguishes between platforms when giving multi-platform advice
- ❌ FAIL: "post at optimal times" without citing engagement velocity data
- ❌ FAIL: claims specific percentage weights without [CONFIRMED] or [LEAKED] source