Signal Scout — Intelligence

Intelligence Engine

11-layer signal logic that separates signal from noise, scores relevance, and routes the right content to the right team member.

Live Ingestion Test
Layer Status ● 6 ACTIVE ⚡ 6 COLD-START Updated June 2, 2026 · analyst + competitor feeds wired
L1 COLD-START
Emergence Detection
45-day rolling window · growth rate + saturation
emergenceInput × 0.60 → formula
⚡ Bug fixed (count:1→engWeight). Needs ~50 signals/topic over 7+ days.
L2 ACTIVE
Thought Leader Watchlist
Tier 0/1 name + handle + domain matching
Override — floors final score at 0.90
● Fixed string/object author bug. Karpathy, Andrew Ng, Sam Altman etc. now fire.
L3 COLD-START
Question Gap Detector
Cross-platform question clustering · 7-day window
questionGapBonus × 0.15 → formula
⚡ Needs body text — HN/arXiv titles rarely contain questions. Add blog/newsletter bodies.
L4 COLD-START
Practitioner / Analyst Divergence
Sentiment delta between analyst orgs vs. practitioners
emergenceInput × 0.40 → formula
⚡ HBR + MIT Tech Review + competitor feeds wired. Needs ≥2 signals/bucket/topic to produce divergence score.
L5 COLD-START
Competitive Gap Intelligence
Topics Accenture / Deloitte / McKinsey haven't covered
relevanceInput × 0.40 → formula
⚡ Accenture + Deloitte + McKinsey feeds wired. Coverage map building — gap scores calibrate as competitor topics accumulate.
L6 ACTIVE
Temporal & Calendar Intelligence
Conference calendar · Q4 windows · pre-event boost
Post-formula multiplier × 1.0–1.5
● Firing now — Snowflake Summit (Jun 2–5) active. Data+AI Summit in 7d.
L7 ACTIVE — STATIC
Source Trust Registry
Per-domain trust score · decays on noise, repairs on value
sourceAuthority × 0.20 → formula
⚡ Trust table static. Wire Route→Ink + Dismiss buttons to engine.feedback() to make it adaptive.
L8 ACTIVE
Engagement Velocity
Rate-normalized reactions, comments, shares vs. baseline
velocityTrajectory × 0.10 → formula
● Fixed field mismatch (points→reactions). HN/Reddit engagement scoring now live.
L9 COLD-START
Cross-Platform Heat
Same topic on 2+ platforms in 48h → 1.30–2.0× multiplier
Post-formula multiplier × 1.0–2.0
⚡ Fixed platform field bug. Needs HN + Reddit + arXiv covering same topic within 48h window.
L10 ACTIVE
Keywords & Relevance Depth
Hakkoda domain term matching + content density
relevanceInput × 0.60 → formula
● Fully wired. Expand HAKKODA_DOMAIN_TERMS as new practice areas emerge.
L11 ACTIVE
Noise Filter
Promotional content · low-trust domains · thin content
Override — zeros score when triggered
● Fully wired. Tune NOISE_PATTERNS as false positives surface.
L12 COLD-START
Hype Cycle Position
5-phase inference: Innovation Trigger → Plateau
Post-formula multiplier × 0.7–1.5
⚡ Defaults to 1.1×. Needs L9 history + 30d concept data. Future: Gartner subscription for Tier 2.
To Get All 12 Layers Firing — Priority Order
✓ P0
Analyst feeds wired → L4 now COLD-START
HBR + MIT Technology Review (analyst) added to signal-scout.js. source.type=‘analyst’ explicit typing added — classifySourceType now uses it first. L4 divergence will activate once ≥2 signals/bucket/topic have been processed.
✓ P1
Competitor feeds wired → L5 now COLD-START
Accenture, Deloitte, McKinsey feeds added as source.type=‘competitor’. L5 coverage map now accumulates topics from competitor publications — gap scores will diverge from 1.0 as coverage data builds. Add PwC + BCG feeds to complete the set.
P2
Wire Route & Dismiss to engine.feedback() → adapts L7
Every time a signal routes to Ink or is dismissed, call engine.feedback(id, signals, wasValuable). Source trust then adjusts +0.05 (validated) or −0.02 (noise). Makes L7 learn from editorial decisions.
P3
Fetch article body text → activates L3
HN and arXiv only provide titles right now. Fetching full body text surfaces the questions people are actually asking ("How do we...?", "Why does...?"). L3 clusters these across sources to find unanswered question gaps.
P4
Run Scout for 7+ days → activates L1, L9, L12
L1 needs a 45-day growth window to detect true emergence vs. saturation. L9 needs the same topic appearing on HN and Reddit and arXiv within 48h. L12 needs all of this plus 30d concept history to infer hype phase reliably.
P5
Add LinkedIn / Twitter → maximizes L9 heat multiplier
Currently max heat is 1.30× (HN + Reddit = 2 platforms). Adding LinkedIn and Twitter unlocks the 1.60× and 2.0× tiers. LinkedIn signals are especially relevant for Hakkoda's enterprise audience and are available via the LinkedIn API or monitoring tools.
Processing Layers
1
Emergence DetectionActive
  • Identifies concepts in the critical 5% window: past "too early" but before mainstream saturation
  • Detects pre-emergence signals via semantic clustering of related concepts in new combinations
  • Hype cycle positioning: pre-emergence = maximum value, peak = contextualization only
2
Thought Leader WatchlistActive
  • Tier 0 (weight 0.97): Karpathy, Andrew Ng, Jensen Huang, Sam Altman, Demis Hassabis
  • Tier 1 (weight 0.92): IBM Research, Snowflake, Databricks, Anthropic, OpenAI
  • Tier 0/1 activation auto-elevates signal to 0.90+ regardless of engagement volume
3
Question Gap DetectorActive
  • Monitors comment sections, Reddit threads, LinkedIn replies, and conference Q&A for unanswered questions
  • Questions appearing repeatedly across sources with no satisfying answer = talk track opportunity
  • Questions precede answers; answers precede adoption — highest forward-looking value
4
Practitioner vs. Analyst DivergenceActive
  • Tracks analyst firm publications (Gartner, Forrester, McKinsey) vs practitioner sentiment (HN, Reddit)
  • High divergence between analyst optimism and practitioner reality = Hakkoda positioning opportunity
  • Example: "73% of AI projects fail to reach production" vs analyst adoption narratives
5
Competitive Gap IntelligenceActive
  • Monitors Accenture, Deloitte, McKinsey, BCG, PwC AI publications continuously
  • Identifies topics they all cover (add depth), topics none cover (own the conversation), topics covered poorly
  • Updated monthly, feeds directly into talk track prioritization
6
Temporal & Calendar IntelligenceActive
  • AI governance content scores higher Q4 (compliance budget season)
  • Innovation/strategy content elevated pre-conference (IBM Think, Dreamforce, NeurIPS)
  • Year-end predictions window: November/December elevated scoring
7
Source TrustActive
  • Weighted trust score per source: tier-1 publications, peer-reviewed research, practitioner blogs
  • Trust decays (−0.02) for repeated low-signal submissions; repairs (+0.05) for validated signals
  • New sources enter at neutral 0.55 and earn authority through track record
8
Engagement VelocityActive
  • Measures comment volume, share rate, and reaction count at time of ingest — normalized per platform
  • High velocity = topic resonating with real audiences right now, not just editorial opinion
  • Feeds: LinkedIn reactions, YouTube comment count, Reddit upvote velocity, HN comment score
9
Cross-Platform HeatActive
  • Detects same topic trending across 2+ platforms within a 48-hour window — multiplier applied post-formula
  • Isolated signal = strong. Same topic on LinkedIn + HN + Reddit + YouTube = fire (2.0×)
  • Multipliers: 1 platform 1.0× → 2 platforms 1.3× → 3 platforms 1.6× → 4+ platforms 2.0×
10
KeywordsActive
  • Matches signal content against Hakkoda domain term list: Snowflake, Databricks, LLM, RAG, MLOps, data governance, etc.
  • Breadth score (5+ matches = full) combined with content density heuristic (unique token ratio + length)
  • Combined with Layer 5 (Competitive Gap) to produce the relevance input: L10 × 0.6 + L5 × 0.4
11
Noise FilterActive
  • Override layer: zeros out final score entirely if 2+ noise signals detected — signal never surfaces
  • Noise signals: promotional/PR content, no domain relevance, chronically low-trust source, insufficient content
  • Requires 2+ triggers to fire, preventing false positives on borderline content
12
Hype Cycle PositionActive
  • Maps signals to hype cycle phase using our own data: platform volume, practitioner ratio, content type shift, engagement decay
  • Innovation Trigger 1.5× · Slope of Enlightenment 1.3× · Trough 1.2× · Plateau 0.8× · Peak 0.7×
  • Future: Gartner subscription (Tier 2 source) validates our position detection. New layers can be added as scoring needs evolve.
Live Engine Test
Submit a signal and watch all 12 layers score it in real time.
—