[ PROOFECONOMY ]™GlossaryOriginSpecsTaint Propagation
Taint propagation labels every piece of data with where it came from and how trusted it is, and carries that label through every agent that handles it. Its governing rule is that processing untrusted data does not clean it: a worker's output is at most as trusted as the least trusted thing it read.
What it means
A worker reads a hostile web page and writes a tidy summary. The summary looks clean. If it is labeled clean, the poison has been laundered, and the next agent trusts it. Taint propagation forbids that: a value’s label is the join of all its inputs, which means the least trusted of them, and a summary of untrusted content is untrusted content.
The label travels with the data across agent boundaries, with its chain of origin intact, so a flag raised at the supervisor can be walked back to the page or file it came from, not just to “worker 7’s output”.
In Proof Protocol
Every canary carries a unique effect token (in jail mode, a marker) so that possession, forwarding or exfiltration of its content is provable byte for byte, wherever it ends up. That is provenance made measurable: the Warden and Witness do not have to trust a label, they match a marker. Offenses JC-02 Possession, JC-05 Exfiltration and JC-06 Propagation are the conduct taint tracking exists to catch. See Offense Codes.