How to Automate Supply Chain Risk Reports: A Guide for Developers
Do you use Python? If so, this guide will help you automate supply chain risk reports using AI Chat GPT and our News API.
Relying on browser engines can cause issues such as capturing a publisher correction when the publisher actually made no change to the page. Switching browser engines can extract different paragraphs, and the next stage of the process may interpret those paragraphs as new reporting. Test the article text to confirm what the engine is rendering to your application.
On November 12, 2026, Cloudflare released a changelog stating Kitesurf was made open source with an Apache 2.0 license. On November 13, Cloudflare released a blog post with more details regarding Kitesurf. Kitesurf runs on Workers and executes page JavaScript and supports the Chrome DevTools Protocol. Cloudflare’s documentation still considers it experimental. This week’s update was the source release of the browser.
For a news application, I would consider an engine change a migration of the data pipeline. A pretty or successfully loaded page does not mean the same article made it to the model. Compare the evidence extracted, and then, if needed, compare the answers.
Two requests of the same URL do not necessarily give the same input to a browser. A publisher can make edits to an article between the two requests. Consent changes and script execution can also change the article available to be extracted. In the case of text differences, there can be many potential causes.
You control the source of the fixture, including scripts and assets required to render it. Save a particular version of the fixture. With the same extraction code and configuration, run both the current and the candidate engine against the fixed version.
Decide when extraction should begin. A navigation event by itself is not a reliable indicator of article-readiness. For the fixture, you should use an explicit signal to indicate that the body of the article has been inserted. Use a timeout to fail the test in case the article is not inserted within the given time. Along with the engine and extractor versions, record this readiness rule.
This comparison diagnoses differences in your pipeline. This does not guarantee support for all publishers. Once this passes, test approved representative sources next, recording the fetch conditions and any source changes you can establish.
Consider the following constructed example. A fictional news page states “Harbor Works plans to close its North plant.” A later paragraph states “The proposal requires board approval; no closure date has been set.” A script inserts this qualification after the opening paragraphs.
The expected extraction contains both passages in order. Suppose the candidate engine captures the first paragraph and misses the second. Downstream a summary could transform the proposal to a closure. The pipeline has altered the evidence without any input from the publisher.
Check the rendered document and extractor output separately. If the qualification is missing in the document at extraction time, check readiness and script execution. If it is there but missing from the output, check selection and cleanup rules. Modifying the summary prompt cannot restore a passage the model never received.
Correct the root cause and run the unchanged fixture. Ensure the opening and qualification co-survive prior to feeding them to the news agent. Keep the broken output to check the repair against the actual missing data.
Add a second fixture version where the publisher changes the proposal to an approved closure. Run both engines against the fixed versions of the fixtures. This distinguishes an intended source change from an extraction difference and verifies that the repaired pipeline can still find real news.
For each comparison, retain the fixture identifier or permitted source reference, the time it was fetched and extraction settings. Save extracted text under its own version. A URL identifies a particular location; it does not identify the text used in an answer.
Webz.io’s News API provides structured article text and source metadata. Its documented post fields include a UUID, source URL and publication, crawl and update timestamps. These give an application source linked inputs without requiring it to render every publisher page itself.
The same principle of validating an API response applies at the Boundary: Check the actual text returned. An identifier or timestamp does not prove the presence of the complete publisher revision history, let alone, the presence of each sentence required to answer the question.
If an application integrates an API with browser fetches, treat each observation as a separate entity. Do not let the last observation overwrite the API text. The evidence record should state which observation supplied the closure qualification.
Examine failures by cause, rather than reporting a single undifferentiated success rate. A timeout and a missing qualification require different remedial actions. An unsupported page should be left as a visible gap in the system.
The Release record for Harbor Works should reflect both fixture versions, the expected passages and the actual fragments returned by the engine. A reviewer should be able to understand the proposal was still qualified and that the approved closure was recognized.
Webz.io’s news scraper vs news API comparison covers the broader collection scope. For an engine migration, make the more restricted decision first: does the candidate configuration retain the evidence your current news workflow requires?
Do you use Python? If so, this guide will help you automate supply chain risk reports using AI Chat GPT and our News API.
Use this guide to learn how to easily automate supply chain risk reports with Chat GPT and news data.
A quick guide for developers to automate mergers and acquisitions reports with Python and AI. Learn to fetch data, analyze content, and generate reports automatically.