Tutorial verification record

This is the working verification record for the shared scraper setup, daily listing watch, and protected-site browsing guides. It separates what was tested against the running product from what was checked against provider documentation.

Verification date: 2026-08-30
Runtime baseline tested: e6cfa7d. The later documentation changes reorganize the same verified setup and controls without changing runtime behavior.

Tested against the local product

Check	Result	Data effect
Docker deployment at http://127.0.0.1:18800	Healthy; dashboard sign-in and all tabs loaded	Local only
Browser licence	Running binary resolved to CloakBrowser 151.0.7922.108.3-pro	None
Residential proxy	Real page observed a California residential ISP; browser timezone and locale matched America/Los_Angeles and en-US	Proxy traffic only
Saved proxy credential	Found a stale encrypted-volume value, replaced it from the valid EVOMI_PROXY_PASSWORD, restarted, and repeated the Pro/proxy check successfully	Local encrypted settings updated; .env unchanged
Manual browser	Launched from Browsers, live view rendered, proxy status appeared, then the test instance was closed	No forms submitted
Notion connection	Existing Listings database discovered and column mapping verified read-only	No listing rows created or edited
BizBuySell one-page sweep	After two blocked attempts using the stale saved credential, the corrected-password run completed in 64 seconds with 50 listings across one page and synced=null	No Notion writes (sync=false)
Failure evidence	Task result contained status, summary, error, page count, and a screenshot/HTML evidence directory	Local task history only
archive_page	Archived https://example.com/ into a disposable Notion page in one attempt; returned Example Domain, 5 blocks, and 149 Markdown characters; Source Content was present	Scratch parent archived after verification; existing listings untouched
Agent reads archived body	A second synthetic page received the same five-block Example Domain archive; Claude then used the connected Notion integration to read Source Content and identify the archived source	No existing listing rows edited; synthetic verification page retained as reproducible evidence

The repository's scripts/verify_browser.py was used for the independent browser check. It resolves the actual process binary, visits an IP/geo echo page through the launched browser, compares the page's timezone and locale, and closes the browser.

Verified from source code and tool schemas

• scrape_listings requires one or more BizBuySell search-result or broker-profile URLs, returns a job ID immediately, and has one shared max_pages value per call.
• get_scrape_listing_results must be polled with the returned job ID.
• A sync=false result contains all found listings and makes no Notion writes.
• A sync=true result contains only newly inserted rows. Existing rows are counted but not returned or refreshed.
• Each new synced row carries synced_row_id, the Notion page ID expected by archive_page.
• archive_page appends page content to an existing Notion page, returns status/counts rather than full text, and normally blocks for 40–60 seconds.
• Archive writes are append-only and can be partial if Notion accepts some blocks before an error. The tutorial therefore forbids blind retries.
• create_instance uses a durable profile. A configured but unusable proxy fails rather than silently falling back to a direct connection.
• agent_browser accepts one listed command per tool call and requires fresh snapshot references after page changes.

Checked against current provider documentation

• Railway Serverless location and the required redeploy for the setting to affect the container.
• CloakBrowser's key/concurrency model and current checkout instruction that a paid licence key is emailed after payment.
• Evomi's live My Products → Core Residential → Proxy Generator flow, whole-string copy behavior, host-first format, Core Residential endpoint, and HTTP port. The captured guide images omit the copied credential string and personal account controls.
• Notion internal-connection permissions and page sharing.
• Notion's hosted MCP URL, OAuth flow, read/write behavior, and current tool names.
• Claude remote-connector and Cowork schedule flows.
• ChatGPT Work availability on individual paid plans, Scheduled Tasks, and the separate plan restrictions documented for full custom-MCP actions.
• The signed-in Claude and individual ChatGPT accounts both exposed Scheduled in their current interfaces. Privacy-safe screenshots were captured from each product; the ChatGPT page visibly includes the Chat / Work selector.
• The workspace-owning Notion account was opened through Google OAuth. The live Seed URLs database, page-level Connections menu, Listings property panel, Bot Triage select, triage-criteria page, and listing-watch runbook were inspected and captured without editing rows. The live final triage values are REVIEW and REJECT; the tutorial and prompt template were aligned to them.
• The authenticated Claude account exposed Customize → Connectors → Add custom connector. Its live form asks for a name and remote MCP server URL. The existing Cloak Biz Scraper connection appeared as a connected custom web connector.
• The authenticated individual ChatGPT account exposed Plugins → Create app. Its live form asks for a server URL and authentication method, defaults to OAuth, and requires an explicit custom-MCP risk acknowledgement before creation. Privacy-safe crops of both product screens were added to the tutorial.
• The end-to-end archive handoff was repeated against a synthetic Notion page. The scraper's live archive_page call appended five blocks under Source Content; Claude then used its hosted Notion integration to find that page, read the archived body, and identify the source as Example Domain without editing the page. A privacy-safe product screenshot is included in the tutorial.
• The local dashboard's live Settings → Capacity and Settings → Disk space controls were inspected again for the operations guide. Privacy-safe screenshots show the real calculated task budget, host-memory warning, browser-build cleanup, task-evidence cleanup, and upload cleanup controls.

Publication status

The tutorial's product, documentation, visual, and end-to-end verification checks are complete.

Do not publish screenshots that expose an integration token, APP_SECRET, a proxy username or password, a private database ID, personal listings, or a private Notion page URL.