Phoenix bookings dropped twelve percent last week. The analyst sends back a hypothesis on Wednesday. By Wednesday the cause has compounded another five days.

The analyst pivot table is the bottleneck

The pipeline, end to end

1. Position in the 3-skill Forecast + Attribute + Diagnose bundle. Forecast (forward-looking-recommendations predicts KPI trajectory) + Attribute (attribution-rollup runs multi-touch attribution on rollup data) + Diagnose (this skill — root-cause attribution sketch surfaces hypotheses behind unexpected movements). NEW 9th canonical bundle architecture in arc — 3-branch analytical extension of the 2-branch Observe → Forecast + Correlate pattern on the anomaly-detection agent.
2. Drop detection against the Forecast baseline. The Forecast axis provides the expected per-location KPI trajectory. Actual rollup deviation beyond the expected-variance band triggers the sketch investigation. Drop magnitude tunes the sketch depth (small drops sketch quickly; large drops branch deeper).
3. Eight-stream branch coverage. Paid spend (per-location budget changes, ad-platform- policy changes, cap hits). Organic search + local-pack rank (per-location rank shifts, AIO citation flips). GBP attribute changes (hours, services, photos, Q&A). Competitor activity (new competitors, competitor price changes, competitor Posts cadence). Inventory state (out-of-stock signals affecting service bookability). Per-location ad-budget shifts. Weather (atypical patterns affecting foot traffic). Calendar effects (holiday shift, school schedule, local-event impact).
4. Branch-and-search investigation per stream. Each stream branch queries its source signal stream for shifts during the drop window. Shifts that correlate temporally with the KPI drop surface as candidate causes. Each candidate carries a confidence score based on shift magnitude + temporal alignment + historical-pattern match.
5. Cross-stream correlation. Some root causes span multiple streams (a competitor entry might show in competitor-activity-stream AND local-pack-rank-stream simultaneously; an ad-budget cut might show in paid-spend-stream AND paid-traffic- stream). The cross-stream correlator surfaces multi- stream signatures as higher-confidence single causes.
6. Hypothesis ranking + confidence scoring. The sketch surfaces multiple candidate hypotheses ranked by evidence-strength score. Single-cause drops surface one high-confidence hypothesis. Multi-cause drops surface several contributing causes with their respective contribution-weight estimates. The ranking tunes per operator over the operating window.
7. Per-vertical hypothesis priors. Per-vertical operators carry different prior- probability weights on root causes. Healthcare booking drops more often trace to insurance-coverage- change events. Cannabis booking drops more often trace to regulatory-rule changes affecting product claims. Spa booking drops more often trace to weather. The priors load from the rule library and tune per cycle against the actual operator root-cause history.
8. Per-franchisee independent sketches. Per-location root-cause sketches run independently per franchisee. The Phoenix sketch uses Phoenix- territory data exclusively. The Tampa sketch uses Tampa-territory data. Independent execution preserves per-franchisee privacy (one franchisee operating signal does not leak to another).
9. Brand-wide aggregator. A brand-wide aggregator subscribes to all per- franchisee sketches and surfaces system-wide root causes (a brand-wide GBP API issue affecting all locations vs a Phoenix-specific competitor entry). The aggregator distinguishes brand-wide causes from location-specific causes by frequency-and-temporal clustering.
10. Delivery into the regional-VP Monday-morning inbox. The sketch output delivers with the original KPI alert into the regional-VP inbox via Slack or email digest (cross-link to the per-location-rollup- reporting cadence + the seo-alerts rank-stream emission pipeline). The hypothesis lands with the drop notification rather than three days later.
11. Hypothesis validation feedback loop. When the analyst or franchisee validates a sketch hypothesis (true or false), the validation feeds the ranking-model tuning. Hypotheses that consistently prove true gain weight; hypotheses that consistently prove false lose weight. The feedback loop runs per- vertical and per-location to allow per-context tuning.
12. Integration with the broader compliance and brand substrates. Per-vertical regulatory-rule changes surface as a root-cause-candidate from the compliance-overlay substrate. Brand-voice gate threshold changes surface from the brand-spec substrate. The sketch consumes operator-side governance events as additional root-cause-candidate streams.
13. ROI measurement. Time-to-hypothesis pre vs post deployment (target hours pre / minutes post). Time-to-action pre vs post deployment (target days pre / hours post). Per- location KPI recovery curve steepness when the sketch hypothesis-driven action lands quickly vs the manual-investigation baseline. The ROI signal tunes per-vertical prior weights and per-stream branch priorities per cycle.