SlopNet
Generate-and-select for content, graded against measured response. Version 0 ships only the selecting half on purpose: a ledger, an instrument contract and gates that refuse. Generation does not exist yet, and it says so.
IN BUILD
What it is
SlopNet’s premise is old and cheap to state: make many variants of a piece of content, publish, keep what resonates. The hard part is the keep. Almost every system that claims to do this scores itself against numbers it made up, so SlopNet is built from the measuring end backwards.
The selection half: a prequential ledger that grades what has resolved before it calls the next thing, append-only, outranking any narrative about how things are going; an instrument contract that makes a measuring instrument state its own coverage; and two gates, where generation reports permission and capability separately and publication has no send path at all. Its first instrument is this site’s funnel.
The one idea worth stealing: given only counts, a day with no data is ambiguous. Nobody came, or nothing was running? An early version guessed, wrote fabricated measured zeros into an append-only ledger, and marked honest forecasts wrong. So an instrument must say how many hours it witnessed, and a day grades only when the whole day was witnessed or is provably outside coverage. Everything else stays open with its reason recorded.
What exists today
Private repository, last pushed 15 September 2026. 47 tests, standard library only, no install step. The ledger, the instrument contract and the gates are built and tested; a generator is absent and named as absent.
Not done
No generator, so nothing is selected yet.
The only live instrument is this site’s funnel, which counts little.
Next
A first generator behind the gate, once there is an honest signal to select against.
Built with
Python (stdlib) prequential ledger