Headwind

Proof

What has been shown

Live runs, controlled tests and their limits. All on sample agency data.

Live runs

Full weekly reviewrun_eac5efb75e90428e: 5 of 5 steps, 6 Crustdata credits, both cfo.ai readbacks matched, Main untouched.
Latest coderun_903b239b1dad49be: 6 credits, 1 model call, neatlogs trace finalized.
Reply that moves cashA simulated “on hold until January” reply lowered Reviewed; “back on” restored it. Both readbacks matched.
Crash mid-runThe worker resumed from the saved step with 0 duplicate cfo.ai writes.

Controlled tests

97 of 97 case notes passed grading by a separate model. Crustdata matched 35 of 40 real client companies across 4 more made-up agencies, with none matched wrongly; only 23 had 6+ months of hiring history. Details: evaluation report.

What this doesn't prove