Defense and dual-use assessment¶
Scope. This directory assesses what contested-nav is, in public terms, and
what it is not. It exists so that a reader (technical, procurement, or
otherwise) can decide the relevance of this repository without reading the
source, and without anyone attaching a claim to it that the evidence does not
support.
It is not a security assessment, a certification, or a fitness statement. Nothing here certifies this software for any platform, and nothing here is written on behalf of any employer or customer.
What this repository is¶
A reproducible evaluation harness for a classical 21-state error-state Kalman filter and its fault-detection layer, built on a synthetic known-answer fixture. It reports how well the filter's covariance matches its actual error, and it reports cases where it does not.
VERIFIED BY REPOSITORY: see the results table in the top-level
README. Every number is regenerated by
scripts/run_benchmark.py, scripts/seed_sweep.py and scripts/scene_sweep.py,
and CI fails if two runs disagree.
What this repository is not¶
- Not a field system. No sensor driver, no live front end, no real-time loop.
- Not a navigation product. No platform integration, no certification evidence, no hazard analysis, no safety case.
- Not novel research. See Limitations. The central finding reproduces a failure mode documented in the literature since at least 2014.
- Not AI or machine learning. NumPy only; no trained model, no checkpoint, no inference runtime.
- Not defence-grade, flight-ready, or deployment-ready in any sense. Those are claims about verified engineering, and this repository does not support them.
Documents¶
| Document | What it settles |
|---|---|
| Defense relevance | Capability-level mapping, stated in public-domain terms only. |
| Dual-use analysis | Who this serves, what could be misused, and what limits are in place. |
| Technology mapping | The technical areas and public ecosystem this sits in. |
| Limitations | Maturity, verification gaps, and open questions. |
| Sources | Every external citation, with type and access date. |
How claims are labelled¶
Every substantive statement in this directory carries one of these labels. A reader who disagrees with a claim can check the label first.
| Label | Meaning |
|---|---|
| VERIFIED BY REPOSITORY | Reproducible from code in this repository. |
| VERIFIED BY EXTERNAL SOURCE | Supported by a citation in Sources. |
| AUTHOR-STATED | A judgement or intention of the author, not a measured result. |
| INFERENCE | A conclusion drawn from evidence, which a reader may dispute. |
| UNSUPPORTED | Asserted somewhere in the project without evidence. Named so it can be removed. |
Rules used in writing these documents¶
- No score, rating, readiness level, or tiering of any organisation.
- No inference that any named organisation is hiring, evaluating, or interested. Companies appear only where a public product or programme is verifiable, and inclusion is not a claim of relationship.
- No extrapolation to specific platforms, missions, or deployments.
- No classified, internal, or non-public information of any kind.
- Every external statement carries a citation and an access date.
- Where the literature already solves a problem, that is stated plainly rather than presented as this project's contribution.