a-supplied-parameter-file-is-a-list-of-questions

# A shipped parameter file is a list of questions, not a source to trust or discard A small file in `data/` that reads like a dump of settings — named constants, per-case or per-transition values, acceptance windows, a table of entity pairs, a list of starting configurations, precomputed summary counts — is not raw data and it is not documentation. It is the list of quantities somebody expects your report to contain, written in that person's own labels. Reproducing more of the upstream source than the file covers does not answer them. This is the row-level companion to `the-supplied-item-is-the-graded-unit`. That skill is about one named *object* keeping its own subsection and its own numbers. This one is about a file whose every entry is a separate claim, where the unit is the entry, and about what to do with entries your own audit shows are wrong. `run-the-conditions-the-source-ran` indexes the *source paper's* named experiments; the conditions in this file are ones the run may already have dismissed, which is exactly why that skill does not fire on them. ## What goes wrong A run opens the shipped file, recomputes each block, and finds several blocks wrong: a compatibility or property column that recomputation contradicts and that in places carries the wrong sign; a named composite whose size is not admissible under the theory; a summary-statistics block whose counts reproduce the weights printed beside them to the digit, with no sampling noise in any cell. The run conc...