An autonomous competitive intelligence platform that tracks 24 companies across 5 scoring dimensions, predicts market movements, and generates actionable business intelligence — running every night without human intervention.
SBPI is a competitive intelligence system built for the micro-drama vertical — a $2B+ mobile entertainment market with 24 tracked platforms ranging from ReelShort ($400M revenue) to early-stage African entrants. The platform produces three types of output:
Every company ranked across 5 structural dimensions. DramaBox leads at 82.8, with ReelShort close at 82.0. Rankings update weekly with full score breakdowns.
Momentum-based predictions for next-week movements. JioHotstar (+9.45 cumulative) and COL Group (+7.25) are the strongest bullish signals. Amazon (-5.8) and Netflix (-5.0) are bearish.
Per-company analysis of breaking news, deal flow, and market events — scored through all 5 dimensions. Material events trigger defensive recommendations.
Most competitive intelligence is reactive — it fires recommendations against every negative signal and creates false urgency. SBPI is designed to be proactive. It maintains a model of each company's strategic trajectory and reserves its recommendations for genuine structural threats — moments where something has changed that the company's existing plan does not account for.
The system actively learns to distinguish signal from noise through a self-improving optimization loop. Over time, it gets better at knowing when to speak up and when to stay quiet.
Each company is evaluated across 5 structural dimensions that capture different aspects of competitive position. Scores range from 0-100. The composite score is a weighted average.
The weighting reflects a structural thesis: distribution is the most defensible moat in mobile entertainment. Content and narrative matter, but without distribution they don't convert. Monetization is weighted lowest because it follows from the other four — if you have distribution, content, narrative, and community, monetization is execution, not strategy.
The platform runs a 9-phase nightly pipeline. Phases 1-6 are the core intelligence engine. Phases 7-9 are the new BI agent system — event research, defensive recommendations, and self-improving signal weighting.
Loads weekly data into an RDF triple store, validates predictions against actuals, generates new predictions, upgrades attestation metadata, and runs 11 SPARQL queries producing insight digests.
Uses TPE (Tree-structured Parzen Estimation) to auto-tune 12 prediction parameters. The system gets more accurate over time without human intervention. Currently optimized across direction threshold, confidence base, magnitude bonuses, and mean reversion rate.
New this quarter: event discovery via SerpAPI searches real news sources, defensive BI generates mitigation strategies filtered for strategic relevance, and a signal weight optimizer learns to distinguish material events from noise.
| # | Company | Score | Delta | Tier | Signal | Under Pressure |
|---|---|---|---|---|---|---|
| 1 | DramaBox | 82.8 | +4.0 | Dominant | BULLISH | Community |
| 2 | ReelShort | 82.0 | -2.1 | Dominant | BEARISH | Narrative |
| 3 | Disney | 76.6 | +2.3 | Dominant | BULLISH | Content |
| 4 | iQiYi | 71.2 | +1.2 | Strong | — | Community |
| 5 | Google / 100 Zeros | 63.6 | — | Strong | — | Community |
| 6 | JioHotstar | 62.3 | +4.0 | Strong | BULLISH | Narrative |
| 7 | Holywater / My Drama | 61.6 | — | Strong | — | Community |
| 8 | Netflix | 60.8 | -2.0 | Strong | BEARISH | Content |
| 9 | GoodShort | 58.8 | +1.7 | Strong | BULLISH | Community |
| 10 | CandyJar | 58.6 | — | Strong | — | Narrative |
| 11 | ShortMax | 56.6 | — | Strong | — | Narrative |
| 12 | Lifetime / A+E | 55.5 | +1.4 | Strong | BULLISH | Content |
| 13 | Amazon | 50.2 | -2.6 | Emerging | BEARISH | Narrative |
| 14 | Viu | 48.2 | -1.9 | Emerging | — | Community |
| 15 | GammaTime | 46.1 | — | Emerging | — | Community |
| 16 | COL Group / BeLive | 44.6 | +3.2 | Emerging | BULLISH | Content |
| 17 | VERZA TV | 37.3 | — | Niche | — | Monetization |
| 18 | RTP | 31.3 | — | Niche | — | Content |
| 19 | KLIP | 22.4 | -2.7 | Niche | — | Distribution |
| 20 | Both Worlds / Freeli | 21.5 | — | Niche | — | Monetization |
| 21 | Mansa | 19.4 | +1.9 | Niche | — | Monetization |
Data: W12-2026 (week ending March 23, 2026). Scores are 0-100 weighted composites across 5 dimensions. Delta is week-over-week change. Tiers: Dominant (70+), Strong (55-69), Emerging (40-54), Niche (<40).
The BI agent system (Tracks A, B, C) transforms raw competitive data into actionable intelligence. The key innovation is a built-in filter that prevents over-reaction to short-term noise.
Structural changes to competitive dynamics. Events that could trigger tier transitions within 2-3 weeks if unaddressed. Requires defensive action.
W12 example: DramaBox +4.0 score surge driven by Disney Accelerator validation and $500M valuation signal.
Events worth tracking but not yet actionable. May escalate to material if trend continues. The system watches these without generating recommendations.
W12 example: Netflix mobile redesign acknowledges short-form but lacks production commitment.
Expected volatility within a functioning strategy. Normal weekly fluctuation that does not warrant a response. Filtered out by the signal weight optimizer.
Filtered: Minor app store ranking changes, routine press mentions, industry-wide headwinds.
The signal weight optimizer (Track C) tunes 7 parameters that control the threshold between "this warrants action" and "this is expected variance." It uses TPE optimization — the same machine learning approach used in Phase 6 for prediction parameters.
Materiality threshold: How large a dimension delta must be to classify as material (current: 2.0 points)
Structural change weight: Multiplier for events classified as structural vs. temporary (1.5x)
Competitor action weight: Multiplier for competitor-specific threats vs. market-wide headwinds (1.3x)
Trajectory window: How many prior weeks establish the strategic baseline (4 weeks)
Volatility dampener: How much historical volatility reduces urgency (0.3)
The optimizer runs after each weekly cycle. It evaluates whether past recommendations were appropriate, over-reactive, or under-reactive. Over time, it calibrates the system to match expert judgment — getting quieter when noise is high and more assertive when structural changes occur. Expert validation labels are the ground truth signal. After 8+ weeks, the system transitions to outcome validation: did companies that received MATERIAL recommendations actually experience tier transitions?
The platform maintains two layers of structured knowledge beyond the scoring data:
Built from trade press extraction (The Wrap, Deadline, TechCrunch, Variety). Maps companies, people, technologies, deals, and market forces. InfraNodus graph analysis surfaces structural gaps — disconnects in the competitive landscape that represent opportunity or risk.
ReelShort (0.54) — Category leader, connected to all major clusters
Microdrama (0.34) — Category term bridging Chinese and Western market discussions
GoodShort (0.26) — Y Combinator-backed challenger, bridging VC and entertainment clusters
GammaTime (0.24) — JioHotstar acquisition connecting Indian and global market clusters
Maps how value actually moves through the micro-drama ecosystem using REA (Resource-Event-Agent) accounting principles. Three files decompose the market into agents, resources, and flows.
1. No creator-to-platform flow — Independent creators cannot publish on dominant platforms
2. No cross-platform licensing — Content is siloed per platform
3. Quality tier gap — $5-15K Chinese vs. $50K+ US production, nobody owns the $20-40K middle
4. Genre concentration — Market dominated by romance/melodrama
5. No regulatory response strategy — US-China tensions are existential for market leaders
Chinese tech money — Patient, volume-oriented, vertically integrated (COL Group, Storymatrix)
US venture capital — Milestone-driven, differentiation-focused, exit-seeking (Y Combinator / GoodShort)
Legacy studio budgets — Incremental allocation from existing spend, risk-averse (Netflix, Disney)
Ontology: OWL/RDF (sbpi.ttl, 424 triples)
Store: Oxigraph SPARQL endpoint
Validation: SHACL shape constraints
Queries: 11 SPARQL query library
Bridge: InfraNodus → RDF conversion
Primary: SerpAPI (Google News, 988 searches/month)
Enrichment: Claude CLI (headless research)
Extraction: Firecrawl (authenticated scraping)
KG Analysis: InfraNodus MCP (24+ tools)
Subscription: The Wrap (paywalled content access)
Predictions: 4-method ensemble (persistence, momentum, mean reversion, KG-augmented)
Optimization: Optuna TPE sampler (12 + 7 parameters)
BI Agents: Event impact, defensive recommendations, signal weighting
Scheduler: 9-phase nightly cycle via launchd
Reports: Cloudflare Pages (global CDN, free tier)
Store: Local Oxigraph (HTTP + direct access)
Graphs: InfraNodus cloud
Code: Python 3, git-versioned
Weekly stack rankings across 24 companies. Nightly insight digests. Prediction engine with 4 methods. TPE parameter optimization. Event impact analysis with SerpAPI discovery. Defensive BI recommendations. Signal weight optimization. Knowledge graph with 119 entities. Value flow ontology mapping power structures.
Expert validation loop: Weekly human review of BI recommendations feeds the signal weight optimizer, improving material/noise classification.
Proactive BI agent: Phase 2 — opportunity recommendations (not just defensive mitigation). Architecture is already in the ontology, waiting for 4+ weeks of defensive BI data to calibrate against.
Content intelligence expansion: Deep crawl of The Wrap via Firecrawl. Additional source integration (Variety, Deadline, Sensor Tower data feeds).
Knowledge graph enrichment: Named executives, deal terms, production budgets extracted from ongoing article processing.
Vertical expansion: The content intelligence methodology is designed to be reusable. Same 5-layer approach (Discovery → Extraction → Synthesis → Ontology → Monitoring) applied to adjacent verticals.
Client-facing intelligence products: Per-company BI reports delivered as a service. Each tracked company gets their own dashboard showing where they stand, what happened this week, and what to do about it.
Prediction accuracy milestone: Target KG-augmented method outperforming baseline by 15%+ on directional accuracy with 8+ weeks of training data.