Prediction Accuracy — Night over Night
"Each night the pipeline evaluates four prediction methods against real outcomes. Mean reversion climbed from 47% to 71%. The KG optimizer went from matching random to 69.9%."
Directional Accuracy by Method — Best Per Night
KG-Augmented (Optimized)
Apr 3 ran 50-trial random search. TPE/Optuna expected to push past 70%.
Nightly Reports
Click any card to expand the full nightly digest.
Top Weekly Movers
JioHotstar: Sustained 4-week climb W10→W13, from 58.3 to 65.4 (+7.1 net), TierStrong.
DramaBox: Dominant tier at 82.75, +4pt week in W12.
Amazon: +4.05 gain in W13, now TierEmerging (tier downgrade despite score gain — worth investigating).
COL Belive: Steady climb from ~37 to 47.25 over 4 weeks.
Key Structural Anomalies
COL Belive: MI (market influence) gap persistent — composite 41-47 vs MI dimension 85-94. ~47pt gap. Systematically undervalued in composite.
Netflix: CS (community signal) anomaly — composite ~61 vs CS dimension ~26-30. Gap ~35 points every week. Brand power disconnected from community engagement.
Predictive Signals (W13 Narratives)
DramaBox (82.75): HRTS panel, "Tempted by the Wrong Brother" launch, $100M raise at $500M valuation in-market.
iQIYI (67.3, +1.6): Triple announcement — HK listing, $100M buyback, Nadou Pro AI tool. Stock +10%.
JioHotstar (65.4, +3.15): IPL opening weekend 515M reach, Tadka microdrama at 100 titles in 7 languages.
Netflix (60.95, +0.15): No microdrama moves. Short drama +155% YoY paid installs per AppsFlyer — opportunity being left on the table.
Prediction Accuracy (W13 Evaluation)
| Method | W12 Dir. Acc | W13 Dir. Acc |
|---|
| Persistence | 23.5% | 19.1% |
| Naive Momentum | 23.5% | 38.1% |
| Mean Reversion | 47.1% | 71.4% |
| KG Augmented (default) | 23.5% | 19.1% |
KG Interface Optimization
Best directional accuracy: 65.49% (50 random search trials). Key config shifts: direction_threshold +142%, mean_reversion_rate +106%, anomaly_contributes flipped to True.
Action item: Install Optuna to upgrade from random search to TPE. Expected +5-10% gain over the 65.5% ceiling.
Weekly Movers
DramaBox (W12): +4.0 pts → 82.75, TierDominant. Largest mover this week.
JioHotstar (W11→W12): +3.95 pts → 62.25. IPL 2026 launch + 100 microdramas in 7 languages.
Col-Belive (W12): +3.15 pts → 44.55, TierEmerging. Consistent climber.
Amazon (W11): -3.2 pts. Notable downward mover.
Netflix (W11): -3.0 pts. Continuing decline trend.
New Entrants
Google/100 Zeros (SBPI 63.65): Range Media Partners initiative, slate includes Mike Fleiss, McG, Simon Fuller, Kenan Thompson. First window on Google TV.
HolyWater (SBPI 61.65): $22M Series A, Fox equity + 200-title deal, Dhar Mann Studios partnership. Revenue tripled in 2025. 55M downloads.
Dimension Anomalies
Col-Belive: MI 90-92 vs composite 41-44 — massive structural undervaluation.
Netflix: CS 26-30 vs composite in the 60s — awareness without commitment.
Amazon: CS 25, DP 83 — heavy distribution, zero content strategy activation.
Prediction Accuracy (W12 Evaluation)
| Method | Dir. Accuracy | MAE |
|---|
| Persistence | 23.5% | 1.80 |
| Naive Momentum | 23.5% | 1.80 |
| Mean Reversion | 47.1% | 2.11 |
| KG Augmented | 23.5% | 1.80 |
KG-LLM Interface Optimizer
Default KG-Augmented: 23.5% → Optimized: 69.9% (+46.3pp). Key shifts: direction_threshold +159%, mean_reversion_rate +157%, anomaly_contributes True, new divergence_weight / tier_proximity_weight parameters.
Top Weekly Movers (W11→W12)
| Company | W12 Score | Delta | Tier |
|---|
| DramaBox | 82.75 | +4.0 | Dominant |
| JioHotstar | 62.25 | +3.95 | Strong |
| Col-Belive | 44.55 | +3.15 | Emerging |
| Disney | 76.55 | +2.3 | Dominant |
| ReelShort | 82.0 | -2.05 | Dominant |
| Netflix | 60.8 | -2.0 | Strong |
Dimension Anomalies — Two Distinct Gap Patterns
Pattern 1 — Hidden MI strength: Col-Belive MI 85-92 vs composite 41-44 (~47pt gap). Building audience faster than infrastructure.
Pattern 2 — Incumbent CS weakness: Netflix/Amazon/Disney CS 25-40 vs composites 60-76 (31-35pt gaps). The incumbents are all underweighted on content strategy. Disney DP ~93 but composite 62.8 — distribution power is its hidden asset.
New Entrants
Google / 100 Zeros (63.65): Range Media slate + Google TV first window. Platform distribution + IP slate is different from pure-plays.
HolyWater (61.65): $22M Series A, Fox equity, Dhar Mann deal, 55M downloads, revenue 3x in 2025. Credible crossover player.
Infrastructure
Oxigraph running at port 7878. 1,672 triples loaded covering W10-W12, 21 companies. store_client.py was missing from working tree — recovered from commit 0e304b2 in git history. SPARQL query files (weekly-movers.rq, dimension-anomalies.rq, predictive-signals.rq) missing from disk — insight digest ran but produced empty sections.
W12 Evaluation
| Method | Dir. Accuracy | MAE | Brier |
|---|
| Persistence | 23.5% | 1.80 | 0.250 |
| Naive Momentum | 23.5% | 1.80 | 0.279 |
| Mean Reversion | 47.1% | 2.11 | 0.250 |
| KG-Augmented | 23.5% | 1.80 | 0.250 |
Key Movers (W11→W12)
Gainers: JioHotstar (+3.95), DramaBox (+4.0), Col-Belive (+3.15), Disney (+2.3) — streaming surge.
Losers: Amazon (-2.6), KLIP (-2.65), Netflix (-2.0), ReelShort (-2.05), Viu (-1.85).
Stable: BothWorldsFreeli, CandyJar, RTP, Verza-TV.
Top Gainers
DramaBox +4.0 → 82.75 (Tier 1): Disney Accelerator validation + $500M valuation. Only profitable pure-play at $10M net.
JioHotstar +3.95 → 62.25 (Tier 2): IPL launch converts from planning to production — 300M subscribers entering microdrama.
COL/BeLive +3.15 → 44.55 (Tier 3): FILMART execution live, 1,700 catalogue active. SaaS model provable.
Top Losers
KLIP -2.65 → 22.35 (Tier 4): Structurally squeezed by JioHotstar's 100-title IPL slate.
Amazon -2.60 → 50.2, downgraded Tier 2→3: Zero microdrama strategy. Gap widening vs Google, Disney, Netflix.
ReelShort -2.05 → 82.0 (still Tier 1): Head of production defected; ShortMax 3888% growth eroding app store position.
Dimension Anomalies
Amazon: Distribution Power 80 vs composite 50.2 (+29.8) — but Content 22 and Narrative 25 are severe vulnerabilities. The distribution moat masks a hollow competitive position.
COL/BeLive: MI 92 vs composite 44.55 (+47.5) — dramatically undervalued. Narrative and Content are the drags.
Disney: Narrative Ownership 93 vs composite 76.55. Content Strength 55 is the gap — execution lags the brand story.
DramaBox: Community Strength 65 vs composite 82.75 (-17.75) — the one dimension not keeping pace.
KG Interface Optimization Thesis
30 Optuna TPE trials lifted KG directional accuracy from 23.5% to 69.9% (+46.4pp). Key: direction_threshold +159%, mean_reversion_rate +157%, anomaly_contributes → True. The gap between naive KG (23.5%) and optimized KG (69.9%) is the core thesis: the knowledge graph has real signal, but the default interface doesn't extract it. Optimization is the lever.
Top Gainers (W12)
| Company | Score | Delta | Signal |
|---|
| DramaBox | 82.75 | +4.0 | Disney Accelerator, $500M valuation, only profitable pure-play |
| JioHotstar | 62.25 | +3.95 | IPL launch imminent, separate microdrama platform |
| COL/BeLive | 44.55 | +3.15 | FILMART launch, "Shopify for microdrama" validated |
| Disney | 76.55 | +2.3 | Locker Diaries #1, DramaBox Accelerator = financial hedging |
| Mansa | 19.35 | +1.85 | Africa-first, original series launch |
Top Losers (W12)
| Company | Score | Delta | Signal |
|---|
| KLIP | 22.35 | -2.65 | Squeezed by JioHotstar's IPL slate in home market |
| Amazon | 50.2 | -2.6 | Zero microdrama strategy, downgraded Tier 2→3 |
| ReelShort | 82.0 | -2.05 | Production head defected, ShortMax 100M DLs now #3 |
| Netflix | 60.8 | -2.0 | Mobile redesign = acknowledgment, not commitment |
| Viu | 48.15 | -1.85 | DramaBox SE Asia expansion compressing core territory |
Momentum Signals (W13 Outlook)
Bullish: JioHotstar (+9.45), COL/BeLive (+7.25), Disney (+5.55), DramaBox (+5.25), GoodShort (+4.5), Lifetime/A+E (+4.15)
Bearish: Amazon (-5.8), Netflix (-5.0), ReelShort (-2.6)
Bullish Signals
JioHotstar — +5.5 delta, strongest upward momentum
COL Group / BeLive — +4.1, "Microdrama in a Box" infrastructure play
Disney — +3.25, platform giant re-engaging
GoodShort — +2.8
Lifetime / A+E — +2.8
DramaBox — +1.25, steady #2
Bearish Signals
Amazon — -3.2 delta
Netflix — -3.0 delta
ReelShort — -0.55 (mild erosion despite #1 rank)
First Predictions Recorded
68 predictions for W12-2026 across 4 methods x 17 companies. Mean reversion was the only method generating directional predictions (all others predicted "stable" with only 1 week of data). Evaluation deferred — W12 actuals not yet loaded.
Structural Finding
The bearish Amazon/Netflix signals vs. bullish JioHotstar/COL divergence is the most interesting structural finding — platform giants losing ground to vertical specialists. This becomes a persistent theme across every subsequent nightly run.