Stage 5 — Clustering + Evidence Pack

Data flow

Active signals are correlated into clusters, enriched with concrete evidence, and reduced to short predicate tags the downstream engine routes on.

flowchart TD
    ACTIVE["Active, undiagnosed signals"]
    subgraph Correlate["correlateSignals"]
        C1{"data_health?"}
        C1 -->|"Yes"| SOLO["Solo cluster"]
        C1 -->|"No"| C2["Group by URL / funnel step"]
        C2 --> C3["Merge funnel_drop into
matching URL cluster"]
        C2 --> C4["Remaining → group by type"]
    end
    CLUSTERS["Signal Clusters"]
    subgraph Enrich["enrichClusterEvidence"]
        E1["Pull replays (PostHog)"]
        E2["Capture screenshots (Steel → R2)"]
        E3["Fetch JS stack traces"]
        E4["Pull rage_click DOM snapshots"]
        E5["Cohort comparison
mobile vs desktop · source vs source"]
        E6["Derive predicates
url_specific · js_error_on_conversion_path
rage_click_on_cta · replay_corroboration"]
    end
    PACK["EvidencePack
supporting · contradicting · missing
predicatesSatisfied · clusterKey"]
    EVID_DB[(posthog_evidence_artifacts
upsert on dedupKey)]
    ACTIVE --> C1
    SOLO --> CLUSTERS
    C3 --> CLUSTERS
    C4 --> CLUSTERS
    CLUSTERS --> Enrich
    Enrich --> PACK
    PACK --> EVID_DB
    style Enrich fill:#e0f7fa
    style PACK fill:#c8e6c9
    style EVID_DB fill:#e0f7fa,stroke:#00ACC1