Why every web page gets a different study shape

A recipe and an API reference should not open the same way

Paste a public URL into Novus Learn and the result always stays grounded in that page, but the page may be trying to do very different work. A tutorial is organised around steps. An FAQ is organised around questions. Documentation often concentrates definitions and code. A research paper needs its sections and sources close. Treating all four as a generic article would technically preserve the text while making the first study screen less useful than it could be.

Page profiles solve that presentation problem. During web-page ingestion, Novus Learn records publisher metadata and structural features from the material it actually parsed. A deterministic classifier uses those signals to select a page kind and confidence. The project then changes the order of its existing views so the modes most suited to that kind appear earlier.

This is adaptation without a generative guess. The classifier does not rewrite the page, infer the publisher's intent from a language model, or create a special summary that cannot be traced. It chooses among explicit profiles, records why, and leaves every study view available.

The decision starts with signals the page exposes

The strongest signal is a type the publisher declares in JSON-LD, such as NewsArticle, BlogPosting, HowTo, Recipe, FAQPage, Product, or ScholarlyArticle. Scholarly citation metadata is another strong clue. When declarations are absent, the classifier can use the shape of the isolated main content: several ordered steps suggest a how-to, repeated question headings or expandable blocks suggest an FAQ, a substantial share of code suggests documentation, and a price signal paired with a specification table suggests a product page.

Generic Open Graph metadata and the URL path provide weaker clues. A path containing docs or api can tentatively suggest documentation; a blog or news path can do the same for those profiles. If nothing distinctive survives parsing, Novus Learn uses the balanced article profile. That fallback is important because a confident but wrong shape would be less helpful than the familiar general layout.

• Publisher-declared JSON-LD types receive the highest priority.
• Citation tags can identify a scholarly page without reading its prose as a guess.
• Ordered lists, question headings, code density, prices, and data tables provide structural evidence.
• Open Graph type and URL paths are useful but deliberately lower-confidence signals.
• A page with no distinctive signal remains a tentative general article.

You can see how the decision was made

Open the Overview of a web-page project and the card labelled How this page was read names the selected profile. It also shows a Strong, Likely, or Tentative match and lists the signals that supported it, such as a JSON-LD type, citation metadata, ordered steps, or a generic article declaration. The classification is therefore part of the interface rather than an invisible setting.

Confidence describes the match between this page and the profile, not the truth or quality of the page. A strongly declared recipe can still contain a bad recipe. A tentative article can still be authoritative. The label only explains why Novus Learn placed certain study tools where it did.

That distinction protects the evidence contract. Source quality remains your judgment, and every claim still needs a supporting block. The page-shape card gives you enough information to notice a surprising classification and use the complete set of views anyway.

What changes for each kind of page

Overview remains first for every profile. After that, a how-to or recipe moves the How-to walkthrough and Guided lesson toward the front. Documentation promotes structured Notes and Breakdown. A scholarly page leads with Summary and Breakdown while keeping Sources close. Product pages bring Visuals and Notes forward, and question-and-answer pages make Summary and Ask easier to reach. News and blog posts favour a quick digest.

The Overview's study-mode buttons follow the same profile order, so the adaptation is visible before you enter the full tab row. For a page classified as a how-to or recipe, that classification can also help the engine admit a step-by-step explainer when the source contains enough real list-backed claims. It still cannot create missing steps: without supported list evidence, there is no walkthrough to show.

• How-to and recipe: walkthrough, lesson, and notes rise earlier.
• Documentation: notes and structural breakdown lead the study sequence.
• Scholarly: summary, breakdown, sources, and notes stay close together.
• FAQ: summary and Ask make the page's question structure easier to navigate.
• Product: visuals and structured notes become easier to reach.

What never changes or disappears

A page profile is a full reordering, not a feature gate. Overview, Ask, Summary, Notes, How-to, Breakdown, Canvas, Visuals, Lesson, Flashcards, Quiz, and Sources remain in every web-page project. A documentation profile cannot remove the quiz; an FAQ profile cannot hide the canvas; a scholarly profile cannot force you to begin with its preferred route. The source may leave a particular view empty when it lacks enough evidence, but the classifier itself does not delete it.

The underlying document also remains one normalised source with its canonical URL, title, available author and date, sections, blocks, and retrieval notes. Claims, flashcards, quiz questions, and Ask answers still pass through their existing validators. Reordering the doors does not loosen the rules behind them.

This makes the feature reversible in practice. If the selected shape is tentative or simply not how you want to study, choose another tab. You are never trapped inside an interpretation of the page.

Use the profile as a starting point, not a verdict

For documentation, begin with Notes to collect the named concepts, then use Breakdown or Ask for the part you are trying to implement. For a how-to, inspect the original ordered steps before relying on the walkthrough, especially for health, legal, financial, or safety-sensitive instructions. For a paper, keep Sources open and check whether the project had full text or only limited material. For an FAQ, use Ask to retrieve the relevant supported answer without assuming the page covers questions it never states.

You can also compare the visible signals with the original page. A publisher-declared type tells you what the publisher intended; a structural match tells you what the parsed content looked like. Neither replaces reading. They simply shorten the path to the most likely useful study mode while keeping the classification legible.

The principle is the same one that guides the rest of Novus Learn: organise aggressively, invent nothing, and keep the route back to the source open. A better starting shape should help you reach evidence faster, not make the interface seem more certain than the page deserves.