Validate API

Validate whether a list of companies match your Ideal Customer Profile. Provide your ICP description and domains. DiscoLike generates a validation prompt and evaluates each company, returning structured fit/confidence/reasoning results.

Built on the DiscoGen pipeline. Requires an LLM provider integration, unless you run it on DiscoLike’s own native ICP-fit model.

POST   https://api.discolike.com/v1/validate/icp

Results are async. Poll GET /discogen/status/{task_id} for completion.

ICP Fit Validation

POST /validate/icp

Validates domains against an ICP description. Returns a task ID immediately.

Parameters

Parameter	Type	Required	Description
icp_text	String	Yes	Your ICP description (e.g., “B2B SaaS companies in HR and payroll with 50-500 employees”). Must be non-empty
domains	Array	Yes	List of domains to validate (1-10,000)
context_mode	String	No	What data the LLM sees per domain: website (default, profile + homepage), profile (firmographics only), domain (name only)
integration_id	String	No	LLM provider integration UUID. Uses your default if omitted. Pass native-icp to score with DiscoLike’s own model instead of an LLM (see Native ICP-Fit Model)
web_search	Boolean	No	Enable web search enrichment (default: false)
search_provider_id	String	No	Search provider UUID for web search. Uses your default if omitted

Response

{
  "task_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "column_name": ["Fit", "Confidence", "Reasoning"],
  "status": "in_progress",
  "total_domains": 3
}

Result Schema (per domain)

When the task completes (poll via GET /discogen/status/{task_id}), each domain returns:

Column	Values	Description
Fit	Yes, No	Whether the company matches the ICP
Confidence	high, medium, low	Assessment confidence level
Reasoning	String	1-2 sentence explanation

A native run returns a different set of columns. Read column_name from the task response rather than hardcoding either set.

Caution

The Fit verdict is title case, Yes or No, matching the ICP Fit column of a native run. Earlier versions of this endpoint returned lowercase yes / no on LLM runs; update any exact-string comparison, and compare case-insensitively if you support both.

Example

curl -X POST https://api.discolike.com/v1/validate/icp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "icp_text": "B2B SaaS companies providing HR and payroll solutions with 50-500 employees",
    "domains": ["gusto.com", "rippling.com", "stripe.com"]
  }'

Then poll for results:

curl https://api.discolike.com/v1/discogen/status/TASK_ID \
  -H "Authorization: Bearer YOUR_API_KEY"

Example completed result:

{
  "task_id": "a1b2c3d4-...",
  "status": "completed",
  "results": {
    "gusto.com": {"Fit": "Yes", "Confidence": "high", "Reasoning": "HR and payroll platform for SMBs, directly matches ICP"},
    "rippling.com": {"Fit": "Yes", "Confidence": "high", "Reasoning": "HR, payroll, and IT platform for businesses, strong ICP match"},
    "stripe.com": {"Fit": "No", "Confidence": "high", "Reasoning": "Payment processing infrastructure, not HR/payroll"}
  }
}

The status response also carries estimated_cost, cost_metadata and warnings, described under Cost Metadata and Warnings. Validation runs at low search depth with two fixed queries per domain when web_search is enabled; for targeted lookups use DiscoGen with a prompt that names the source.

Native ICP-Fit Model

Pass integration_id: "native-icp" to score every domain with DiscoLike’s own ICP-fit model, a cross-encoder served on our GPUs, instead of your LLM provider. The run needs no LLM provider integration and makes no calls on your provider key, so it reports no LLM or search spend.

The model was trained on LLM validation verdicts, so a score is what an LLM validator would most likely have said about the company, at no LLM cost. It is not more accurate than an LLM.

Differences from an LLM run:

• No web search is performed. web_search and search_provider_id are ignored.
• context_mode is always website, the company profile plus homepage text the model was trained on. Passing profile or domain has no effect.
• The columns differ, as below. Reasoning carries the confidence band rather than an explanation: the model returns a score, not text.

Native Result Schema (per domain)

Column	Values	Description
ICP Fit	Yes, No	Whether the company matches the ICP. Yes when ICP Score is 0.50 or above
ICP Score	String	Calibrated probability that the company matches the ICP, "0.00" to "1.00", two decimals
Reasoning	High confidence, Medium confidence, Low confidence - ...	The confidence band, derived from how far ICP Score sits from the 0.50 cutoff: high when the score is 0.15 or below or 0.85 or above, medium between 0.15 and 0.35 or between 0.65 and 0.85, low between 0.35 and 0.65. This engine scores rather than explains, so the column carries the band instead of a rationale. A low-confidence verdict is close to a coin flip - measured at 0.598 accuracy on held-out data, against 0.943 for high - so treat it as unresolved rather than as a No

Native Example

curl -X POST https://api.discolike.com/v1/validate/icp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "icp_text": "B2B SaaS companies providing HR and payroll solutions with 50-500 employees",
    "domains": ["gusto.com", "stripe.com"],
    "integration_id": "native-icp"
  }'

{
  "task_id": "a1b2c3d4-...",
  "status": "completed",
  "results": {
    "gusto.com": {"ICP Fit": "Yes", "ICP Score": "0.91", "Reasoning": "High confidence"},
    "stripe.com": {"ICP Fit": "No", "ICP Score": "0.08", "Reasoning": "High confidence"}
  }
}

Errors

Status	Meaning
400	The validation prompt generated from your icp_text does not list its criteria under Mandatory:, Reject if: and Nice-to-have:. The native model only answers ICP validation prompts; rephrase the ICP text or pick an LLM integration for it
503	Native ICP-fit validation is not available on this server right now. Retry shortly or pass an LLM integration_id