Listings rot. That's the whole problem.

A tool directory is a decaying asset. Pricing changes on a Tuesday with no announcement. A beloved open-source project quietly stops merging PRs. A startup pivots from "AI meeting notes" to "AI SDR agent" and nobody updates the listing. A free tier that said 10,000 requests/month now says 500.

When we started Oryxen, we counted: how many records in a typical tool directory are wrong within six months? Our estimate: most of them, somewhere. Pricing pages are the most-edited pages on the internet. If your directory doesn't re-verify, it's a museum.

So we built freshness into the data model itself. Every listing on Oryxen shows a freshness date — when we last re-tested it. And we re-test all 36,000+ records every quarter.

What a freshness signal actually means

A freshness date is a promise and a liability. It says: as of this date, the pricing class, the key features, the pros and cons, and the top alternatives were checked against reality. It does not say "perfect forever." It says "someone looked, here's when."

Quarterly re-testing is a brutal schedule at 36,000 records — roughly 400 records a day, every working day. Some of it is automated checks: does the homepage resolve? Does the pricing page still say what we recorded? Has the GitHub repo gone dark? The judgment calls — pros, cons, alternatives — need human eyes. We've learned to spend automation on the facts and humans on the opinions.

Deterministic order: same input, same order — no auction

The second decision we made early: there are zero paid placements on Oryxen, and the ranking is deterministic. Same query, same data, same order. Every time.

This matters more than it sounds. Most "best tools" lists are auctions wearing a trench coat: whoever pays ranks. Once you sell ranking, every editorial choice becomes suspect — including the honest ones. We refuse paid rank entirely, which means we also refuse the revenue that comes with it. That's the trade. The directory has to be funded some other way, and it is.

Deterministic ordering is also an engineering choice. If order is a pure function of the record data, then order is testable. We can write assertions: given this dataset, this query produces this sequence. When someone asks "why is tool X above tool Y," the answer is a data field, not a business relationship.

Stated unknowns instead of invented ratings

Here's a rule we hold ourselves to: when we don't know something, we say so. We don't invent a 4.7/5 score.

Fake precision is the original sin of tool directories. A star rating implies measurement, and measurement implies a methodology nobody published. We'd rather print "not yet verified" than a number we made up. Pros and cons are written in sentences, not scored. If a record is thin, it says so — and the freshness date tells you when someone will look again.

This is slower than scraping and scoring. It is also the only version I can defend in public, under my own name.

What we still get wrong

I'll be honest about the gaps:

• Niche open-source tools sometimes change licenses or maintainers faster than a quarter. We're working on faster triggers for repos with sudden activity changes.
• Pricing class boundaries are fuzzy. "Freemium" covers everything from "generous forever-free" to "trial wearing a costume." We're splitting these classes finer this quarter.
• New categories emerge faster than we add them. Nine categories covered 36,000 tools, but the edges — AI agents, MCP servers, local models — keep spilling over the taxonomy.

The roadmap is boring on purpose: more verification, finer pricing classes, faster signals on repo death, and the developer tool index I built staying free while we do it.

Building a directory isn't a launch problem. It's a maintenance problem. The moat isn't the data — it's the discipline of re-checking it.