Evidence · Observation review 2026-11-04
What our public program pages show.
Four programs. Five URLs. A dated record of what an automated reader could retrieve, with the limits stated alongside each finding.
Reviewed November 4, 2026, Central Time · Publisher: Good Combinator · Prepared using automated HTTP inspection.
Read the findings · Inspect the method · Download observations (JSON)
The result
All five sampled URLs returned HTTP 200. GROW and Sponsor Matchmaker exposed substantive descriptions in the returned HTML. Agent Foundry and FDE exposed messages explaining that their interactive interfaces require JavaScript.
Successful page retrieval is one small, verifiable result. Interactive function, learning gains, employment, customer savings, and commercial outcomes were outside this review.
Findings and limits
| Public page | HTTP | Observed result | What remains untested |
|---|---|---|---|
| GROW | 200 | Public HTML includes the claim that ten Core modules are free and open, plus module summaries. | Homepage response only; module access, completion, and learning outcomes not tested. |
| Agent Foundry | 200 | Public HTML includes a JavaScript-required message for the visual builder. | Interactive editing, local storage, AI actions, and blueprint export not tested. |
| Agent Foundry app | 200 | App URL returns the same HTML response hash as the root URL, including the JavaScript-required message. | Matching HTML responses do not establish that client-side routes behave identically. |
| FDE Certification | 200 | Public HTML says the course player, application, account access, and credential verifier require JavaScript. | No enrollment, payment, learner outcomes, or credential verification tested. |
| Sponsor Matchmaker | 200 | Public HTML prominently labels the evaluation as sample data and asks visitors not to enter personal or confidential information. | No matching transaction, human introduction, sponsorship, or employment outcome verified. |
How to reproduce the review
- Retrieve each linked URL with an HTTP client, following redirects. Record date, final URL, and status.
- Calculate SHA-256 over the returned response bytes. Compare with the downloadable observation record.
- Read the HTML text after excluding script and style content. Record the supported finding and its limits.
- Test interactive workflows separately in a browser before making a claim about function or outcomes.
Sample: the four named programs linked from our About page, with Foundry’s root and app routes checked separately. No JavaScript was executed, no account was used, and no form, payment, or matching transaction was submitted. Response hashes are comparison fingerprints; the reviewed response bodies are not archived in this public record. Changes on the source sites may produce different hashes.
Discovery and citation measurement
A separate November 4 public-search spot check did not return the exact Protect30A MS4 and Good Combinator Research target URLs for the recorded queries. Broader home and directory pages appeared. Search-tool results are not an exhaustive index census or ordinary consumer rankings.
Both websites already expose analytics code. Visitor totals, referral counts, search impressions, indexed-page totals, and independent chatbot citation accuracy were not obtained. Missing metrics remain unmeasured.
Inspect the exact search queries and returned URLs (JSON).
Controlled citation check
A separate, source-selected assistant experiment recorded four answers and checked each against its supplied page. All four were supported in that small sample (two per website). The same assistant produced and checked the answers; this does not measure independent engine accuracy or organic citation frequency. The experiment was recorded before publication of these edits.
Inspect prompts, answers, source checks, and limits (JSON).
Next evidence to collect
- A fixed-period analytics export with landing pages and referrers.
- Verified search-property URL inspections and indexing reports.
- Saved answers from independent AI engines, with prompts, date, mode, cited URLs, and claim-by-claim source checks.
A future comparison should repeat the same queries and prompts, preserve the full outputs, and report supported, partially supported, unsupported, and unavailable citations separately.
Publication and corrections
This is a self-published technical observation review. It is not independent validation of Good Combinator or Protect30A. No general success rate or causal claim follows from this sample.
Report a correction with a source · Research and evidence work · Explore programs and project listings.