Why we keep a record of our dead ends

Science publishes what works. Results showing that a material will not do a job often stay in notebooks and internal reports, so the next team can try the same thing again and lose the same months.

Lattice Graph keeps a table of the negative findings its own pipelines turn up. They include simulations that found a compound unstable, decisions to drop a candidate with the reason, corrections to earlier screening results, and hits from automated prior-art pre-screens, which are preliminary screens and not legal opinions. On 5 October 2026 our public health endpoint reported 11,601 entries.

Graded, and allowed to be wrong

A finding is only as good as its source. So each entry carries a grade, from A (strongest) to D, and a status: standing, confirmed, or rebutted by a later check. The grade lets an API user weigh a single calculation differently from a result several methods agree on. We have not yet published how grades are assigned.

A finding is evidence rather than a final verdict. When a later check rebuts one, the entry says so rather than disappearing.

What it is for, and its limits

For a materials team the value is time: a quick check against known dead ends before a candidate reaches the lab. The record comes from our own pipelines, not from a public database. The entries themselves sit behind an API key, and we have not yet published a sample of them.