Iteration 30 — The sensitivity pair completes 2/2 CLEAN · 105 TOTAL FIRECRAWL UNREACHABLE — RECORDED

In plain words: the implemented-first streak holds — five clean entries across two firings, against two-of-five under the old order. Today's pair strengthens the STABLE boundary's most distinctive new capability: quantified skepticism about hidden confounders. The robustness value answers "what share of unexplained variance would a hidden variable need to overturn this conclusion?" — in regression units, complementing yesterday's E-value in epidemiology units; the pair now covers both scales an analyst might work in. The second entry is the implication enumerator: given a hypothesized invariance structure, it lists every conditional independence that structure MUST satisfy — turning one grand claim into many small falsifiable ones. Also on the record, less happily: the self-hosted crawler — the provenance ladder's designated last resort, waiting to hunt code for five parked instruments — was unreachable at its first live test (the host is off the tailnet). The ladder's bottom rung is theoretical until it returns; the five revisit notes stay parked, honestly.

 "Could a hidden confounder explain this away?"
   H-102 E-value          risk-ratio scale     (epidemiology units)   iter 29
   H-104 robustness value partial-R² scale     (regression units)     iter 30
   H-101 refutation battery adds the empirical probe (add-a-confounder simulation)
 Three complementary answers to the question no certified gate asks today.