Graph

The discovered structure — nodes, scored lagged edges, multi-hop paths, and full edge evidence.

Discovery's lasting output is the goal's relationship graph: the target at the center, connected to the signals and event streams that predict it.

Get the graph

curl "$API_BASE/v1/goals/$GOAL_ID/graph" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

from predictai import PredictAI

client = PredictAI(token="pa_live_…", workspace_id="ws_…")

data = client.goals.get_goals_by_goal_id_graph(GOAL_ID)

import { PredictAI } from "@predictai/sdk";

const client = new PredictAI({ token: "pa_live_…", workspaceId: "ws_…" });

const data = await client.goals.getGoalsByGoalIdGraph(GOAL_ID);

{
  "data": {
    "goal_id": "b7e9c2d4-…",
    "target_key": "daily_sales",
    "interval_seconds": 86400,
    "horizon_bins": 14,
    "nodes": [
      { "id": "daily_sales", "kind": "goal", "label": "daily_sales", "depth": 0, "related": true },
      { "id": "ad_spend", "kind": "signal", "label": "ad_spend", "depth": 1, "related": true },
      { "id": "supply_disruptions", "kind": "event", "label": "supply_disruptions", "depth": 1, "related": true },
      { "id": "office_temp", "kind": "signal", "label": "office_temp", "depth": null, "related": false }
    ],
    "edges": [
      {
        "edge_id": "7d3f…",
        "source": { "kind": "signal", "key": "ad_spend", "channel": null },
        "target_key": "daily_sales",
        "depth": 1,
        "direction": "positive",
        "lag_bins": 3,
        "lifecycle": "validated",
        "strength": { "corr": 0.62, "lift": 0.18 },
        "confidence": { "q_value": 0.012 },
        "score": 0.988,
        "exploratory": false,
        "mu": 0.94,
        "impact": 0.61
      }
    ]
  }
}

The response is the full workspace universe, not just the winners: every signal and event stream discovery screened appears as a node, with related: true only on those that cleared validation. That's deliberate — you see how much was searched and rejected, not just what survived.

Reading an edge

Field	Meaning
`lag_bins`	How many intervals the source leads the target.
`direction`	`positive` or `negative` effect.
`strength`	`corr` (correlation at the best lag) and `lift` (incremental walk-forward predictive lift).
`confidence.q_value`	The false-discovery-controlled q-value from validation.
`lifecycle`	Where the edge stands: `validated`, `revalidated`, `weakening`, `broken` (live monitoring re-checks edges against fresh data), `exploratory` (suggestive but below the validation bar), or `candidate` (screened, never lift-tested).
`score`	A single 0–1 strength for filtering, in clean bands: validated edges ≥ 0.9, exploratory 0.55–0.85, screen-only candidates ≤ 0.5. Sorting by `score` peels off weaker tiers in order.
`mu` / `impact`	The belief axes: confidence in the relationship and how much the forecast rides on it.

Graph edges are slim — heavy evidence arrays are omitted. Fetch an edge by ID for everything.

Edge detail

curl "$API_BASE/v1/goals/edges/$EDGE_ID" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

from predictai import PredictAI

client = PredictAI(token="pa_live_…", workspace_id="ws_…")

data = client.goals.get_goals_edges_by_edge_id(EDGE_ID)

import { PredictAI } from "@predictai/sdk";

const client = new PredictAI({ token: "pa_live_…", workspaceId: "ws_…" });

const data = await client.goals.getGoalsEdgesByEdgeId(EDGE_ID);

{
  "data": {
    "edge": {
      "edge_id": "7d3f…",
      "source": { "kind": "signal", "key": "ad_spend", "channel": null },
      "target_key": "daily_sales",
      "lifecycle": "validated",
      "lag_bins": 3,
      "direction": "positive",
      "strength": { "corr": 0.62, "lift": 0.18, "stability": 0.8 },
      "confidence": {
        "q_value": 0.012,
        "folds": [{ "fold": 0, "lift": 0.21 }, "…"]
      },
      "evidence": {
        "shap_shape": [["ad_spend_lag3", 0.42], "…"],
        "event_response": null
      },
      "rolling": {
        "baseline_corr": 0.62,
        "last_strength": 0.59,
        "history": [{ "t": "2026-07-14T…", "strength": 0.59, "ewma": 0.6 }, "…"]
      },
      "created_at": "2026-07-01T…",
      "updated_at": "2026-07-15T…"
    }
  }
}

The full document includes the per-fold validation table (confidence.folds), the model's attribution shape (evidence.shap_shape), the measured event-response curve for event-sourced edges, and the rolling history that live monitoring maintains — the raw material behind a weakening transition. Unknown IDs return 404 with "Edge not found". Edge IDs are deterministic, so re-runs update the same edge instead of duplicating it.

Paths

Validated multi-hop chains — fuel_costs → shipping_rates → daily_sales — discovered by the expand stage:

curl "$API_BASE/v1/goals/$GOAL_ID/paths" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

from predictai import PredictAI

client = PredictAI(token="pa_live_…", workspace_id="ws_…")

data = client.goals.get_goals_by_goal_id_paths(GOAL_ID)

import { PredictAI } from "@predictai/sdk";

const client = new PredictAI({ token: "pa_live_…", workspaceId: "ws_…" });

const data = await client.goals.getGoalsByGoalIdPaths(GOAL_ID);

{
  "data": {
    "paths": [
      {
        "path_id": "8e4a…",
        "hops": [
          { "edge_id": "1b2c…", "source_key": "fuel_costs" },
          { "edge_id": "7d3f…", "source_key": "shipping_rates" }
        ],
        "total_lag_bins": 8,
        "confidence": { "min_hop_q": 0.03 },
        "lifecycle": "validated",
        "narrative": "fuel_costs leads shipping_rates by 5 steps, which leads daily_sales by 3."
      }
    ]
  }
}

Paths are sorted by strongest evidence first (lowest weakest-hop q-value). A path whose underlying edge breaks is marked broken rather than deleted. The list is empty until multi-hop chains have validated for this goal.

Scenarios use these chains to propagate an upstream move through to the target — see Scenarios.

Get the graph

Reading an edge

Edge detail

Paths

On this page