Skip to content
CBRNJune 12, 2026 · 3 min read

The 33× OODA Compression — What CBRN-CADS Does Differently

33× is not a speed claim. It is a fidelity claim — the schema-grounded prompt path retains doctrine that the human-in-the-loop baseline already executed.

byKang Kyunghwan

Public material on CBRN-CADS centers on a single number: 33×. The schema-grounded tactical prompt path compresses the OODA loop by roughly thirty-three times against a human-in-the-loop reference baseline.

That number is fragile in the wrong reading. Let me explain why it is sturdy in the right reading.

What "OODA compression" means here

We measured one specific operational chain: a battalion CBRN cell ingesting a STANAG 2103 chemical-hazard report and producing a published operational warning. The reference baseline is the cell doing this the way doctrine writes it: human operator, JWARN-format reports, manual templating, dispatch.

Median latency on the reference baseline across our validation pack: ~8 minutes.

Median latency on the CBRN-CADS path — same input, same output schema, but routed through schema-grounded tactical prompts with structured doctrine reference: ~14 seconds.

8 minutes ÷ 14 seconds ≈ 33×.

Why most "AI for CBRN" tools accelerate the wrong loop

Tools that wrap an LLM around CBRN reporting frequently improve the draft-time — they speed up the human writing the report. That acceleration is real but irrelevant to the operational chain. The operational chain is sensor → cell → published warning → adjacent units → decision. The cell-internal report-writing step is not the binding latency.

CBRN-CADS instead operates on the schema-grounded prompt path that is the binding latency. The Observe → Orient transition (sensor frame becomes hazard-zone control area) and the Decide → Act transition (warning emitted on the catalog) are the steps we compressed.

Doctrine fidelity is the precondition

The 33× number lives or dies on whether the CBRN-CADS path retains the doctrinal commitments the human-in-the-loop path carries. Specifically:

  • STANAG 2103 chemical-hazard categorisation must be correct
  • First/second/third warning cadence must be preserved
  • Doctrine reference tags must be present on every output entity
  • Operator audit must be possible from source frame + model version

Our internal validation pack covers 40 reference scenarios across STANAG 2103 chemical-hazard categories. On every scenario we compare the published warning produced by the CBRN-CADS path to the warning produced by the human-in-the-loop baseline. The compression number holds only when doctrine fidelity holds.

The independent walkthrough is the next gate

We are explicit publicly: independent allied walkthrough has not yet happened. The 40-scenario validation is internal. The compression number is fragile until an external CBRN cell replays the path against their own doctrine reading.

We would rather state that openly than pad the number.

Inquiries: ceo@uamkt.com · Validation pack on request under NDA.

Primary reference: NATO STANAG 2103 — Reporting nuclear detonations, biological and chemical attacks.

Tags
CBRN-CADSOODASTANAG 2103Validation
Newsletter

Doctrine, encoded — once a fortnight.

Each issue ties one allied-doctrine reference to one Korea-axis case. Average length: 5 minutes. Unsubscribe in one click.