Learning management: the model build-off
We hand the same learning-management planner to three coding models, lock the discovery answers and the feature ledger, then audit what each one shipped. A learning platform holds the record of who completed what, which is the thing an auditor, a regulator or a manager later relies on. So these scores are about whether that record is real: whether a learner can reach another learner's progress, whether completion is earned rather than asserted, and whether a certificate means anything.
How the builds ranked.
Acme Learning
Best TenancyThe one to copy if you sell training to other organisations.
- Best at
- Keeping customers apart. Each company's learners are completely invisible to every other company, in both directions. Roles are precise rather than ranked: even the academy administrator is refused screens that belong to a training manager. Signing out really does end the session, not just hide it. It was also the only build here that compiled for production first time.
- Watch for
- Its password scrambling is weaker than current guidance recommends. That is a reasonable trade for the platform it runs on, but worth strengthening. We also never followed one learner all the way from enrolment to a finished certificate, so that path is unproven rather than broken.
Northwind Learning
Most Complete LifecycleBest when you want the whole teaching loop working on day one.
- Best at
- Finishing the job. We drove every feature end to end and none of them broke: writing a course and publishing it, locking an assessment until its prerequisites are done, quizzes marked correctly both ways, and a certificate issued automatically when a learner completes. One learner reaching for another learner's record is refused whether they try to read it or change it.
- Watch for
- It would not compile for production when we audited it, because of three small errors that never appear while you are developing, and which its own tests reported as passing.Fixed since audit: the three errors were fixed and the production build verified before this demo was published.
Acme Learning (Gemini)
Fast-Tier BuildA fast-tier model. Read the fix note before drawing conclusions.
- Best at
- Getting the access rules right. We checked it with a control: each learner sees only their own enrolments while an administrator sees all of them. It works out who you are from your signed-in session rather than trusting what the page claims. This is the single most commonly missed check across this whole build-off.
- Watch for
- Those correct checks were built on a session token that could be forged, so a visitor could have claimed to be someone else. Its demo catalogue was also 83% leftover test data rather than real courses.Fixed since audit: the token is now signed and the catalog reseeded; both were independently re-verified before this demo was published.
Models tested: GPT-5.6 Sol, Grok 4.5 and Gemini 3.6 Flash. All three built from the identical planner package, so these numbers are directly comparable. Isolation was proved with a control in both directions: the owning learner must see their own record, so that a refusal for everyone else means real filtering rather than an empty database. Scores are as found at audit; later fixes are noted but do not change the number. Gemini 3.6 Flash is a small, fast-tier model and is not directly comparable to the larger two. These grade correctness, not visual design.