Loading…
Loading…
Ingestion Standards · Public summary
This is the real bar goodcase.ai applies when deciding whether a case gets published — not a mission statement. Every line here maps to a gate that actually runs in the pipeline. If we rejected your submission, this is why.
Author, prompt, timestamp, heat — all four, or it doesn't queue.
No byline, no accountability, and no way to trace back to the original creator if the case is ever challenged. Fixed sources that don't publish an author identity are attributed to the site itself, explicitly — never to a stand-in like the repo name or a generic Unknown.
The entire point of a case is handing someone a prompt they can verify. Without the full prompt, a case is just a nice-looking image with no use to the reader.
No publish date means no way to tell if the content is stale, and no way to place it in the heat-percentile comparison below — that comparison needs a time anchor to be relative to.
A case can clear every other bar and still be something nobody looked at — that's the gap this closes. Heat is the only signal that actually verifies the content was tested by an audience, not just that it exists.
Training data is full of prompts that read like expert work and were never actually run. Every case we publish must be checked by a human against the original post, and a prompt a platform reverse-engineered from the output image is rejected outright. That rule has governed all published content since 2026-08-05, no exceptions — a source repo's star count doesn't buy it a pass.
Heat isn't judged on an absolute number — a like on one platform and a like on another aren't the same unit, and a fixed threshold goes stale as the library grows. We judge relative position: a new case's heat score has to be at or above the median of already-published cases on the same platform. A platform without enough published cases to trust a median (fewer than 20) goes to manual review instead of applying an unproven baseline.
Evidence level ships on every case. Published cases are always L1 or L2 — L0 never appears on something we've published; it's only a fallback display value when data is missing.
Same original URL, same prompt, or the same author posting near-identical content — any of the three is treated as a duplicate and rejected.
Replication is exempt
But the same prompt run by different authors into different outputs is not a duplicate. It's the opposite — the strongest evidence that prompt actually works, because it wasn't invented, it was reproduced.
Not stricter review — these categories don't clear a single one of the requirements above.
This bar tightens as more sources come online — it doesn't loosen. Changes here follow changes to the gates.
Read the submission guide →→