Whether the tool expects a second submission

A one-shot rating tool is built for a single result - upload, read, leave - while an iterative one expects a series, shown through an account, a dated history and retake prompts. Two tools can share a model, rubric and scale and still differ on this, and the design gives it away before you submit anything.

What a one-shot tool looks like

No account requirement, or an account that does not visibly do much - no saved history worth returning to, no framing around a second attempt. The result page ends the interaction rather than pointing toward a next one. That design is not necessarily worse; it fits a reader who wants one answer and is not building a series, and it avoids the trap of a tool nudging you to resubmit for reasons that have more to do with engagement than with anything you asked for. This axis is mostly independent of whether the tool is a browser page, a downloaded app, or a chat interface - the one-shot-versus-iterative choice is a design decision that shows up in any of those formats.

What an iterative tool looks like

An account by default, a dated history, and usually some visible invitation to come back - a "retake" button, a note about tracking change, occasionally a tips box aimed at the next attempt rather than this one. The tool is built around the assumption that one result is a single draw and that what you actually want is a trend.

That assumption is defensible. A single rating is closer to an anecdote than to evidence, and a tool that structurally supports a series is doing something a one-shot tool cannot: giving you a comparable second, third and fourth number under conditions the tool itself is keeping consistent on its side.

There is a middle case worth naming too: a tool that keeps an account and a history but does nothing to encourage a second submission - no retake prompt, no framing around change, just a passive log sitting behind a login. That design gets the infrastructure of an iterative tool without the intent of one, and it is common enough to be its own category rather than a rare edge case. It still supports a series if you build one yourself, it just will not nudge you toward doing so the way a tool designed around retakes will.

Which one supports comparability

Comparability is not free just because a tool keeps history - the tool holds its own side constant (the model, the rubric, the calibration, at least until it changes them), but you still have to hold your side constant too. Standardising what you submit each time is work the tool cannot do for you regardless of which design it uses, so an iterative tool is a better environment for that discipline, not a replacement for it.

An iterative tool also makes one specific test easy that a one-shot tool makes awkward: submitting the same file twice to measure the tool's own noise before you trust any change between two different submissions. A one-shot design, with no saved account and no easy resubmission path, makes that check more friction than most people will bother with, which quietly hides how noisy the tool actually is.

Neither history nor a retake button on its own proves a tool is well built for a series; both can be present and still not do the job if the tool changes its rubric or calibration between visits without telling you. The infrastructure only pays off if what sits underneath it stays stable enough for a past result and a present one to belong on the same axis. That stability cannot be assumed of a hosted model: Chen, Zaharia and Zou (2023) found GPT-4's accuracy at identifying prime numbers dropped from 84% to 51% between its March and June 2023 versions.

The trade each design is making

A one-shot tool trades away the series in exchange for simplicity and, often, privacy - nothing about you or your submissions sits in an account after the session ends, which some readers will prefer regardless of what it costs in comparability. An iterative tool trades that simplicity for the infrastructure a real series needs: an account, a history you can read without reconstructing it from screenshots, and retakes that do not feel like starting over.

Rate Cock leans iterative, keeping public entries dated and comparable rather than treating each submission as disposable - a design choice worth naming because it is checkable on any public entry, not a claim you have to take on faith. Neither design is the honest one and the other the dishonest one; a one-off reading and a tracked series answer different questions, in roughly the same way a single conversation with a person and an ongoing one do, and a physical measurement taken once versus tracked over months does too.