Impact
We do not have a wall of customer logos. Inventing one would contradict the thing we sell.
What we have is a record: what shipped, what broke, what the root cause turned out to be, and what the evaluations say including where they are weak. Named studies come from Mission Sprints, with permission negotiated up front. Until then, this is the honest version.
Study 01
Global Health IQ · Verifiable from code and release history
A fatal cluster was being dropped silently at ingest.
The failure
A fatal Legionnaires' cluster in New York City was not appearing in the feed. Not misranked — absent. It never reached the detection layer at all.
Root cause
Four disease dictionaries had been maintained separately and drifted apart. A condition recognised by one was unrecognised by another, and anything the ingest dictionary did not know was discarded before it could be scored.
The correction
We collapsed detection into one shared module rather than patching the dictionary that failed. Verified against live feeds; nine correctly geolocated signals were recovered.
What it says about us
The tempting fix was one dictionary entry — a day's work, invisible, and the same class of failure waiting in the other three. We took the architectural fix on a system where the failure mode is people not being warned.
Study 02
PRODIGY · Generated against simulated conditions
An exercise an emergency manager would actually run.
A dashboard proves you can draw. A tabletop exercise proves you understand how the response is organised — which agency owns which decision, and what breaks when two of them are degraded at once. This one was generated from a single typed scenario.
-
Inject 1 · T+00:45
The SCADA blind spot
Drainage pumps fail to engage via remote telemetry as tidal waters peak. Crews attempting manual operation are locked out of the local interface, which now displays a ransom message.
Expected action · Pivot to manual field overrides; reroute traffic without signal synchronisation; triage the intrusion.
-
Inject 2 · T+02:15
Communications degradation
A lightning strike disables a primary microwave relay carrying municipal radio, while a denial-of-service attack halts outbound emergency alerts to the blackout zones.
Expected action · Stand up interoperable comms; fall back to runners and satellite phones; canvass door to door.
-
Inject 3 · T+04:00
The generator threshold
A Tier-1 cooling center loses its backup generator to a corrupted fuel-pump controller. Internal temperature reaches 98°F with 150 vulnerable residents inside.
Expected action · Dispatch transport with auxiliary cooling or evacuate; prioritise feeder restoration against competing demand.
Generated against simulated conditions. Not drawn from, and not representing, any live agency deployment.
Study 03
OpenAya · Public evaluation harness
Evaluation artifact pending publication
The superseded dashboard capture has been removed. This surface will show the current public audit trace when the evaluation artifact is approved.
Publishing the evaluation includes publishing where it is thin.
OpenAya runs a public harness that writes reasoning traces to an inspectable endpoint and aggregates them into an audit view. Anyone can look.
What we will publish
Run counts, traces, and the audit aggregate — as they are.
What we will not publish yet
A headline score. Without route-level comparison data, a single number invites a reading the evidence doesn't support. It goes up when the comparison exists.
Shipping log
Receipts, not announcements
How the next study gets made
Every Mission Sprint is a study in waiting.
A publishable study is a named deliverable of the engagement, with permission agreed at the start rather than requested afterwards.