How Deep Does a Faymas Category Page Go?

Twenty cards, and there is no page two. We loaded all eleven category pages on faymas.in six times each on the morning of 15 October 2026 — 66 page loads, 1,320 card slots — and every load returned exactly twenty cards with no pagination control anywhere in the HTML. The filter itself is exact: all 1,320 cards carried the category of the page they appeared on. What is missing is depth. Those 1,320 slots produced only 497 distinct prompts, and mark-recapture on the repeats puts the pool behind ten of the eleven categories at 24 to 63 prompts. Photography is the exception at several hundred. Added up, the category system opens onto roughly 900 prompts — about 5% of the 17,952 prompt URLs in the sitemap.

Updated October 15, 2026 · Prompt-text analysis

This guide analyses published prompt text. No generation tests were run.

How we measured it

One script, one extractor, one morning. faymas.in publishes its prompt data inside the React Server Component payload embedded in the HTML, so a card's slug, category, model tag and counters can be read straight out of the server response with no browser involved.

The design is a repeat-sampling one, borrowed from ecology. Take a fixed-size sample from a population you cannot count directly, put it back, take another. The number of individuals that turn up twice tells you roughly how big the population was. Here the sample is one page load — twenty cards — and the population is whatever set of prompts that category page is drawing from.

Six passes over the eleven /category/ URLs, back to back, finishing within about eight minutes: photography, illustration-3d, fun-memes, food-drink, wallpapers, social-media, product-brand, ui-graphic, poster-design, logo-identity, presentation. Every pass returned 20 cards. That is 1,320 draws.

Two estimators, reported side by side because they disagree in the one place it matters. Lincoln-Petersen uses only the first two passes: 20 × 20 divided by the number repeated. Schnabel uses all six passes and the running count of everything seen so far. Where the two agree, the pool is stable. Where they do not, the pool is moving under the sampler, and we say so.

The sitemap figure comes from faymas.in/sitemap.xml fetched the same morning: 18,012 URLs, of which 17,952 sit under /prompt/, 11 under /category/, 12 under /prompts/ and 27 under /blog.

The filter is exact

This is the part that is not broken. Every card in the payload carries a category field, and we compared it against the category in the URL it was served from.

Not one card leaked across a category boundary. In a single pass, the eleven pages yielded 220 slots and 220 distinct prompts, with zero appearing on two different category pages. Whatever else is true of this site, the taxonomy is wired to the data correctly.

That is worth saying plainly, because the site's other listing system does not do this at all.

The pools behind the pages are small

Now the depth. Six passes per category, 120 draws each. How many distinct prompts did each category give up, and what does the repeat rate imply about the set it is drawing from?

Ten of the eleven categories land in a narrow band: 24 to 63 prompts. Two or three reloads of a category page and you have seen most of what it has. presentation is the extreme — 25 distinct prompts out of 120 draws, 13 of which turned up on all six passes, and only one prompt in the whole set that we caught exactly once.

Photography is the eleventh, and it is a different animal. Its two estimators do not agree: 133 from the first two passes, 516 from all six. That gap is the signature of an open population — items are entering the pool while we sample it. And 99 of its 109 distinct prompts were drawn exactly once, which means 120 draws did not come close to exhausting it. Five hundred and sixteen is a floor, not an estimate we would defend.

Add the eleven Schnabel figures and you get about 908 prompts. Count only what we actually saw and you get 497. Both numbers sit against 17,952 prompt URLs in the sitemap. That is 5.1% and 2.8% respectively.

There is no page two

We looked for one. On all eleven category pages, the strings Load more, Show more, Pagination, page=, next-page, aria-label="Next" and aria-label="Previous" appear zero times. Each page carries 31 <button> elements and one <nav>, none of which is labelled as a paging control in the response we received.

So twenty is not a page length. It is the whole page. The only way to see more of a category is to reload and take a different draw.

How far back each window reaches

Here the eleven categories stop looking alike. Every card carries a createdAt timestamp, so we can ask how far back in time a category's twenty slots actually reach.

Two categories are under a day deep. Nine reach back between six weeks and three months. The pattern is not about subject matter — it is about how fast each bucket is being filled. A category whose twenty slots all come from the last 22 hours is one that is receiving far more uploads than its window can hold. A category whose twenty slots stretch across 82 days is one that is barely being fed.

The consequence inverts the obvious expectation. Photography is where the volume is, and it is the one category where you can only ever see what arrived today. Its highest copy count across 120 draws was 193, and 63% of its cards had not been liked once. product-brand holds the most-copied prompt we saw all morning at 4,872, and it is a pool of about fifty that you can nearly exhaust in three reloads.

The freshest shelf is also the least proven one. That is what a fixed twenty-slot window does when the pools behind it differ by an order of magnitude.

The other listing system, measured the same morning

faymas.in also runs twelve pages under /prompts/, each named after a keyword: ai-photo-prompts, ai-image-prompts, 15-august-ai-prompts, and so on. We ran the same extractor over all of them.

Eleven returned a card grid. Across those eleven pages — 220 slots — the union was 20 distinct prompts. The median pairwise overlap was 20 of 20: every page was showing the same twenty cards as every other page, and all twenty were photography. One page in the set, /prompts/15-august-ai-prompts, returned no card grid at all.

So the same site runs two listing systems with opposite behaviour. The category pages filter exactly and each opens a different set. The keyword pages filter not at all and all eleven open the same set.

What this measurement cannot tell you

Mark-recapture assumes a closed population and equal catchability. Photography violates the first: new prompts arrive while we sample, which is exactly why its two estimators disagree by a factor of four. Its 516 should be read as "at least several hundred", and its 133 as a figure we do not trust. The ten small categories are stable across both estimators and are the ones we would stand behind.

We read server responses. We did not run a browser. If the site assembles a second page of cards client-side after load, we would not see it — and neither would any crawler that does not execute JavaScript. We can only report that no such control exists in the HTML.

The sitemap count is a snapshot taken on the morning of 15 October 2026. The catalogue was 16,762 prompt URLs when we measured it on 12 October and 17,952 today, so it is growing while we measure.

And 497 distinct prompts is a floor on what the category system can reach, not a ceiling. Six passes is six passes. We did not run sixty.

What the taxonomy is actually for

Put the two findings side by side. The filter is perfect — 1,320 of 1,320. The reach is about 5% of the catalogue. Those are not contradictory, and the site is not confused about which one it built.

A correct category label on 17,952 URLs is worth a great deal to anything that reads the site by URL rather than by eye. Every prompt we drew was labelled correctly, every label matches a real route, and each of those routes is in the sitemap. That is a well-formed index. What it is not is a shelf.

The inequality is the tell. ui-graphic has roughly 24 prompts behind it. Photography has several hundred arriving per day. Both get a twenty-slot window with no second page. Equal windows over unequal pools produces a predictable result: the big category can only ever show you its newest and least-tested items, while the small ones show you nearly everything they have, including the prompts that have actually been copied thousands of times.

If you publish prompts there, that asymmetry is the practical finding. Placement in a small category is close to a guarantee of being seen, because twenty slots over a pool of twenty-five is most of the pool. Placement in photography is a draw of twenty out of several hundred, and the window closes within a day.

What would change this finding is one thing: a second page. Not a better filter — the filter is already exact. Not more categories — eleven is enough for 17,952 items if each one can be walked. The gap between a correct index and a browsable one is pagination, and it is the only thing missing here.

FAQ

How deep does a Faymas category page go?

Twenty cards, with no second page. On all eleven category pages on faymas.in, the strings "Load more", "Pagination", "page=", "Show more", "next-page", aria-label="Next" and aria-label="Previous" appear zero times in the server response.

Do the faymas.in category pages filter correctly?

Yes, exactly. Across 66 page loads and 1,320 card draws on 15 October 2026, every card's category field matched the category page it was served from — 1,320 of 1,320, with zero cross-contamination.

How many prompts can you reach through the Faymas category pages?

Six passes produced 497 distinct prompts across 1,320 slots. Mark-recapture on the repeats puts the pools at 24 to 63 prompts for ten of the eleven categories and several hundred for photography, summing to roughly 900 — about 5% of the 17,952 prompt URLs in the sitemap.

Why is photography different from the other Faymas categories?

Its twenty slots all come from the last 22 hours, while nine of the other ten categories reach back 42 to 82 days. It also produced 109 distinct prompts from 120 draws with 99 of them seen only once, so its pool is both larger and being refilled during sampling.

Do the Faymas keyword pages filter like the category pages do?

No. Eleven /prompts/ pages measured the same morning returned a union of 20 distinct prompts across 220 slots, with a median pairwise overlap of 20 of 20 — every page showed the same twenty cards, all photography.

How was the size of each category pool estimated?

By repeat sampling: six passes of twenty cards per category, then two mark-recapture estimators — Lincoln-Petersen on the first two passes and Schnabel across all six — with the number of prompts seen exactly once reported alongside.

Is 908 the true size of the Faymas category system?

No, it is an estimate with known limits. Mark-recapture assumes a closed population, and photography is open — its two estimators disagree by a factor of four, so its figure is a floor. The ten small categories agree across both estimators and are the reliable ones.