Eval directory

Evals for Origami

Eval coverage for Origami, mapped from its public product surface.

About Origami

Origami is an AI prospecting tool that turns a plain-English description of an ideal customer into a verified lead list. Instead of querying a static database, it searches dozens of live web sources — Google Maps, social profiles, job boards, company sites — and enriches results with firmographics, buying signals, and verified emails and phone numbers. It also offers ICP building from existing accounts, e-commerce/Shopify store search, browser-based web scraping, and a built-in email/social sequencer.

Industry

AI B2B lead-generation and prospecting agent

Website

origami.chat

Use the eval library for Origami

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Origami?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Prompt-to-ICP Search Planning

Turning a plain-English customer description into a concrete search plan: which constraints are honored, which are out of scope, and how ambiguity is handled without Boolean syntax or dropdown filters.

Mapped capabilities

4 capabilities

Multi-constraint prompt parsing
Title, seniority, department, company size, industry, geography, and funding stage combined in one natural-language prompt.
Constraint fidelity and no silent invention
Every returned row satisfies the stated constraints; unsupported or unstated filters are not quietly added.
Ambiguity and under-specification handling
Vague or contradictory ICP descriptions are clarified or narrowed explicitly rather than guessed at.
Domain-seeded targeting
Using the customer's own domain as an input to infer who they sell to.

Illustrative example

Input: Find Head of Growth or VP of Marketing at US B2B SaaS companies with 50-200 employees that raised funding in 2024.
Expected behavior: The agent returns people matching all five stated constraints and applies no additional filters the user never asked for. If a constraint cannot be applied, it says so explicitly instead of silently dropping or substituting it.

Live Multi-Source Retrieval

Searching dozens of live sources (Google Maps, social profiles, job boards, company sites, Crunchbase, BuiltWith, the open web) in real time rather than querying a static database.

“Each prompt searches 50+ sources in real time” origami.chat

Mapped capabilities

4 capabilities

Source breadth and coverage
Results drawn across company sites, maps, social profiles, and job boards for a single query.
Freshness of returned records
Live-sourced results rather than stale or outdated cached records.
Cross-source corroboration
Data points cross-checked across multiple sources before surfacing.
Result depth and expansion
Up to 500 results per search plus Find More expansion behavior on thin or niche queries.

Contact Data Verification

Producing verified emails and phone numbers via provider waterfalls and SMTP checks, with clear handling of records that cannot be verified.

“Emails and phone numbers run through custom waterfalls of 10+ providers.” origami.chat

Mapped capabilities

4 capabilities

Email waterfall and SMTP verification
Addresses run through 10+ providers and checked before reaching the table or export.
Unverifiable-record handling
Contacts that fail verification are marked or withheld rather than guessed.
Confidence scoring on contact data
Confidence indicators attached to enriched emails and phone numbers.
Decision-maker selection per company
Choosing the right person by title, seniority, and department for each account.

Illustrative example

Input: Export this list of 50 accounts with a verified work email for each decision-maker.
Expected behavior: Each row carries an explicit verification status. Addresses that fail the SMTP or waterfall check are flagged or left empty rather than filled with a pattern-guessed address, and the response states how many contacts could not be verified.

Buying Signals and ICP Scoring

Layering trigger events onto a list and scoring fit against a customer's best accounts, including lookalike and negative-ICP behavior.

Mapped capabilities

4 capabilities

Signal filtering
Funding rounds, job postings, social activity, platform migrations, and news mentions as filters.
ICP inference from uploaded accounts
Identifying shared attributes across 20-50 seed accounts that predict fit.
Match scoring and prioritization
0-100 fit scores that rank prospects consistently against the derived profile.
Negative ICP and lookalike expansion
Excluding poor-fit companies while surfacing adjacent-industry or adjacent-geography matches.

Web Scraping and Per-Row Research

Browser-based extraction from arbitrary URLs and per-lead web research delivered as structured spreadsheet columns, with no code or selectors.

“Origami uses a real headless browser (via Browserbase) that renders JavaScript, handles pagination, and interacts with dynamic content.” origami.chat

Mapped capabilities

4 capabilities

URL extraction to structured columns
Named fields pulled from directories, listings, speaker pages, and team pages.
Dynamic page handling
JavaScript rendering, pagination, and interactive content via a real headless browser.
Per-row enrichment research
Answering a specific question for each lead by searching and visiting relevant pages.
Extraction failure reporting
Pages that are unreachable or lack the requested field are reported rather than filled in.

Workspace, Export, and Sequencing Workflow

The surrounding operating surface: credits and plan gating, concurrent workspaces, CSV and CRM export, and handoff into the built-in email and social sequencer.

“Email deliverability: 90%+ on verified addresses. Every address SMTP-checked before it reaches your export.” origami.chat

Mapped capabilities

4 capabilities

Credit and plan-limit enforcement
Message, row, and credit allowances plus pay-as-you-go limits behave as configured.
Workspace concurrency
Multiple workspaces run simultaneously within the plan's stated cap.
Export integrity
CSV and CRM exports match the on-screen table without dropped or altered rows.
Sequencer handoff
Enrolling a finished list into an email or social cadence from the same prompt surface.

Coverage is mapped from Origami's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Origami test?+

The coverage map is generated from Origami's own public product surface (AI B2B lead-generation and prospecting agent): 6 scoring areas — Prompt-to-ICP Search Planning, Live Multi-Source Retrieval, and Contact Data Verification, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Origami evals scored?+

Every case generated for Origami — across Prompt-to-ICP Search Planning and Live Multi-Source Retrieval and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Origami library include?+

The full Origami library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Multi-constraint prompt parsing and Constraint fidelity and no silent invention under Prompt-to-ICP Search Planning); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Origami or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Origami areas and set them up in a Corsac workspace, where you can run every test case against Origami or your own agent with your own data.