API Reference

CAP Computation Primitives

Base URL

https://api.abel.ai

Auth Header

Authorization: Bearer sk-...

Endpoints

predict()·/cap/v1/predict explain()·/cap/v1/explain intervene()·/cap/v1/intervene discover()·/cap/v1/discover counterfactual()·/cap/v1/counterfactual validate()·/cap/v1/validate detect-regime()·/cap/v1/detect-regime check-reflexivity()·/cap/v1/check-reflexivity

POST/cap/v1/predict#

predict()

Forecast a variable using its causal Markov blanket. More robust than statistical models because it only uses causally relevant predictors.

Parameters

Name	Type	Description
target	string	Variable to predict (e.g. "BTCUSD_close")
horizon	int	Prediction steps ahead (1-30)

Example

client.predict("BTCUSD_close", horizon=48)

Response

{ "value": 63420, "ci_lower": 61800, "ci_upper": 65100,
  "mb_variables": ["ETHUSD_close", "DXY", "SP500"] }

POST/cap/v1/explain#

explain()

Extract the minimal causal explanation set for a variable — its parents, children, and cross-domain causal chains.

Parameters

Name	Type	Description
variable	string	Variable to explain
depth	int	Chain depth (1-5, default 2)
cross_domain	bool	Include cross-domain chains (default true)

Example

client.explain("BTCUSD_close", depth=2, cross_domain=True)

Response

{ "parents": [{ "name": "DXY", "influence": "MODERATE_NEGATIVE" }],
  "cross_domain_chains": ["Fed_Rate → DXY → BTC"] }

POST/cap/v1/intervene#

intervene()

Compute P(Y|do(X=x)) — the causal effect of intervening on X. Uses do-calculus, not correlation.

Parameters

Name	Type	Description
treatment	string	Variable to intervene on
outcome	string	Variable to measure effect on
treatment_value	float	Intervention magnitude

Example

client.intervene("Fed_Funds_Rate", "BTCUSD_close", treatment_value=0.5)

Response

{ "effect": -0.042, "ci": [-0.021, -0.068],
  "causal_chain": "Fed →[τ=5h]→ DXY →[τ=2h]→ BTC" }

POST/cap/v1/discover#

discover()

Learn causal structure from data. Automatically selects the best algorithm from 39 options based on data characteristics.

Parameters

Name	Type	Description
data	base64	CSV data (base64 encoded)
query	string	Analysis goal in natural language

Example

client.discover(data=my_csv, query="What causes customer churn?")

Response

{ "graph": { "nodes": [...], "edges": [...] },
  "algorithm_used": "PC", "confidence": 0.87 }

POST/cap/v1/counterfactual#

counterfactual()

What would have happened if…? Computes P(Y_{x'} | X=x, Y=y) using structural causal models.

Parameters

Name	Type	Description
intervene_node	string	Variable that would have been different
observe_node	string	Variable to observe counterfactual outcome
intervene_value	float	Counterfactual value

Example

client.counterfactual("NVDA_close", "AMD_close", intervene_value=0.01)

Response

{ "counterfactual_value": 0.008, "actual_value": 0.012,
  "difference": -0.004 }

POST/cap/v1/validate#

validate()

Test whether a causal graph is consistent with new data. Bootstrap confidence intervals and edge stability.

Parameters

Name	Type	Description
graph_id	string	Graph to validate
test_data	base64	New data to test against

Example

client.validate(graph_id="g-123", test_data=new_csv)

Response

{ "consistent": true, "edge_stability": 0.91,
  "unstable_edges": [] }

POST/cap/v1/detect-regime#

detect-regime()

Detect whether the causal structure has changed — G_t ≠ G_{t-1}. Identifies structural breaks and regime shifts in the underlying causal graph.

Parameters

Name	Type	Description
graph_id	string	Graph to monitor for regime changes
window	int	Lookback window in time steps (default 30)

Example

client.detect_regime(graph_id="g-123", window=30)

Response

{ "regime_changed": true, "change_point": "2026-02-14",
  "changed_edges": [{ "from": "DXY", "to": "BTC", "old_weight": -0.42, "new_weight": -0.11 }],
  "confidence": 0.93 }

POST/cap/v1/check-reflexivity#

check-reflexivity()

Test whether Abel's own output has become a causal variable in the graph — Soros-style reflexivity detection. Guards against self-fulfilling predictions.

Parameters

Name	Type	Description
graph_id	string	Graph to check
output_variable	string	Abel's prediction variable to test

Example

client.check_reflexivity(graph_id="g-123", output_variable="BTCUSD_prediction")

Response

{ "reflexive": false, "feedback_strength": 0.02,
  "threshold": 0.15, "safe": true }

API Reference

CAP Computation Primitives

Base URL

https://api.abel.ai

Auth Header

Authorization: Bearer sk-...

Endpoints

POST/cap/v1/predict#

predict()

Forecast a variable using its causal Markov blanket. More robust than statistical models because it only uses causally relevant predictors.

Parameters

Name	Type	Description
target	string	Variable to predict (e.g. "BTCUSD_close")
horizon	int	Prediction steps ahead (1-30)

Example

client.predict("BTCUSD_close", horizon=48)

Response

{ "value": 63420, "ci_lower": 61800, "ci_upper": 65100,
  "mb_variables": ["ETHUSD_close", "DXY", "SP500"] }

POST/cap/v1/explain#

explain()

Extract the minimal causal explanation set for a variable — its parents, children, and cross-domain causal chains.

Parameters

Name	Type	Description
variable	string	Variable to explain
depth	int	Chain depth (1-5, default 2)
cross_domain	bool	Include cross-domain chains (default true)

Example

client.explain("BTCUSD_close", depth=2, cross_domain=True)

Response

{ "parents": [{ "name": "DXY", "influence": "MODERATE_NEGATIVE" }],
  "cross_domain_chains": ["Fed_Rate → DXY → BTC"] }

POST/cap/v1/intervene#

intervene()

Compute P(Y|do(X=x)) — the causal effect of intervening on X. Uses do-calculus, not correlation.

Parameters

Name	Type	Description
treatment	string	Variable to intervene on
outcome	string	Variable to measure effect on
treatment_value	float	Intervention magnitude

Example

client.intervene("Fed_Funds_Rate", "BTCUSD_close", treatment_value=0.5)

Response

{ "effect": -0.042, "ci": [-0.021, -0.068],
  "causal_chain": "Fed →[τ=5h]→ DXY →[τ=2h]→ BTC" }

POST/cap/v1/discover#

discover()

Learn causal structure from data. Automatically selects the best algorithm from 39 options based on data characteristics.

Parameters

Name	Type	Description
data	base64	CSV data (base64 encoded)
query	string	Analysis goal in natural language

Example

client.discover(data=my_csv, query="What causes customer churn?")

Response

{ "graph": { "nodes": [...], "edges": [...] },
  "algorithm_used": "PC", "confidence": 0.87 }

POST/cap/v1/counterfactual#

counterfactual()

What would have happened if…? Computes P(Y_{x'} | X=x, Y=y) using structural causal models.

Parameters

Name	Type	Description
intervene_node	string	Variable that would have been different
observe_node	string	Variable to observe counterfactual outcome
intervene_value	float	Counterfactual value

Example

client.counterfactual("NVDA_close", "AMD_close", intervene_value=0.01)

Response

{ "counterfactual_value": 0.008, "actual_value": 0.012,
  "difference": -0.004 }

POST/cap/v1/validate#

validate()

Test whether a causal graph is consistent with new data. Bootstrap confidence intervals and edge stability.

Parameters

Name	Type	Description
graph_id	string	Graph to validate
test_data	base64	New data to test against

Example

client.validate(graph_id="g-123", test_data=new_csv)

Response

{ "consistent": true, "edge_stability": 0.91,
  "unstable_edges": [] }

POST/cap/v1/detect-regime#

detect-regime()

Detect whether the causal structure has changed — G_t ≠ G_{t-1}. Identifies structural breaks and regime shifts in the underlying causal graph.

Parameters

Name	Type	Description
graph_id	string	Graph to monitor for regime changes
window	int	Lookback window in time steps (default 30)

Example

client.detect_regime(graph_id="g-123", window=30)

Response

{ "regime_changed": true, "change_point": "2026-02-14",
  "changed_edges": [{ "from": "DXY", "to": "BTC", "old_weight": -0.42, "new_weight": -0.11 }],
  "confidence": 0.93 }

POST/cap/v1/check-reflexivity#

check-reflexivity()

Test whether Abel's own output has become a causal variable in the graph — Soros-style reflexivity detection. Guards against self-fulfilling predictions.

Parameters

Name	Type	Description
graph_id	string	Graph to check
output_variable	string	Abel's prediction variable to test

Example

client.check_reflexivity(graph_id="g-123", output_variable="BTCUSD_prediction")

Response

{ "reflexive": false, "feedback_strength": 0.02,
  "threshold": 0.15, "safe": true }