---
name: web-research-and-structured-extraction
description: >-
  Turns an open research question into a structured dataset: discovers sources, extracts named fields, deduplicates, records provenance per row, and reports confidence. Use when the user asks to research a market, build a list, extract structured data from websites, or gather competitor information.
metadata:
  version: "1.1.1"
---

# web-research-and-structured-extraction

> **Client-neutral reusable package.** No client names, tenant or workspace identifiers, account or
> resource IDs, customer records, credentials, or internal paths. All examples are synthetic.

| | |
|---|---|
| Live catalog name | Web Research and Structured Extraction |
| Live category | Research & Data |
| Catalog entry | 17 |
| Internal capability ID | `CAP-RD-01` |
| Provider | Crawl and extraction services (provider-neutral) |
| Availability | Candidate; validate in the active Workspace |
| Deliverable type | `client_ready_guided_skill` |
| Validation status | Candidate; local package validation only |
| Version | 1.1.0 |


## OLLIE Runtime integration

Use this capability under the active OLLIE Operator. Read live Workspace Context and tenant
bindings for the required inputs below; this public package supplies no client facts or authority.
Consume `ask-ollie-authority-core` for authority, `outloop-access-fallback` for recovery and
resource verification, `email-lifecycle-core` for recipients, approvals and stored-copy delivery
verification, and the bootstrapper Runtime Contract for ownership, scheduling and completion.
These owners govern execution; this Skill does not implement a competing runtime.

Select the relevant expertise below for the requested outcome. Runtime governs combining Skills,
partial use and completing missing steps with other authorized capabilities. This Skill's domain
scope does not limit the Operator's wider authorized task. Preserve domain safety restrictions.
Verify findings against source evidence and read back any resulting artifact or authorized change.

## Business outcome

A research question becomes a table with a source behind every cell, instead of a paragraph of unverifiable assertions.

## Operating modes

Read · Analyze · Draft

## Prerequisites — what must be true before this runs

- Workspace Context
- An approved crawl or extraction grant
- A written target schema and inclusion criteria

## Required inputs

| Input | Source |
|---|---|
| Research question and the decision it feeds | Required |
| Target schema — the exact fields wanted | Required |
| Inclusion and exclusion criteria | Required |
| Geography and language scope | Workspace Context |
| Volume ceiling | Configuration — prevents unbounded retrieval |

## Provenance per row, not per report

Every row carries the URL it came from and when it was retrieved. A dataset whose provenance is a bibliography
at the end cannot be checked, corrected, or refreshed. If a field could not be sourced, it stays empty and is
listed as a gap — never filled with a plausible value.

## Bound the retrieval

Set the volume ceiling before starting. When results exceed it, stop and report rather than continuing — an
unbounded crawl burns credits, and a truncated dataset presented as complete is a false answer.

## Public sources only

This workflow reads what is publicly available. It does not log into a site, does not bypass an access control,
does not defeat a paywall, and does not use a stored session to reach member-only content.

## Action classification

| Action class | This workflow |
|---|---|
| Read | Public web pages, search results, directory and map listings, public company pages |
| Analyze | Relevance scoring against the criteria, deduplication, field normalization, confidence scoring, gap detection |
| Draft | A structured dataset plus a coverage-and-confidence note |
| Write with approval | **Not enabled.** Writing results to a destination system is a separate workflow |
| Destructive | None |

## Approval and delivery

The Operator applies Authority Core and the active Workspace approval policy to any requested
mutation. Email Lifecycle owns recipient classification, sending and stored-copy verification.
The action table defines this capability's domain scope; it grants no permission by itself.

## Expected output

A structured dataset with a provenance URL and retrieval timestamp per row, a confidence rating per field, and an explicit list of what could not be found.

## Synthetic placeholders used in this package

These are illustrative only. Real values come from the client's own workspace configuration and are
never written back into this package.

| Placeholder | Meaning |
|---|---|
| `SCHEMA_PLACEHOLDER` | Target field schema |
| `GEO_PLACEHOLDER` | Geography scope |
| `CEILING_PLACEHOLDER` | Volume ceiling |