Constellia Group

Companies print "biodegradable" on packaging and nobody can verify it. The published tests were run on 15-micron films. Real containers have walls two millimetres thick, and degradation is a surface process, so the two are not the same experiment. Ground Truth reads published PHA biodegradation studies and turns each measurement into a structured row carrying the verbatim sentence it came from. It then crosses wall thickness against environment and reports which combinations have no published data at all, and answers a geometry query: shape, wall thickness, destination environment in; what the literature supports, with citations, out. The first full corpus run produced 26 rows. Nine carried quotations that do not exist in the papers, assembled from fragments of different paragraphs. One recorded a specimen thickness of 0.315 mm taken from the sentence "Microcosms utilized a 315 micron mesh to contain the pellets" - the mesh bag, not the specimen. None of this was visible in the table; every number was plausible. So the rule moved out of the prompt and into the code. Before a row is written, the program re-opens the paper text and searches for every quote. A declared thickness needs its own verbatim quote containing the number with a unit. 26 rows went in, 10 came out, and the 16 that were dropped are listed with their reason. A separate auditor re-runs the same checks over