$ cat ./user-story --domain=caa-maa --topic=bvlos-swarm
DraftHackathon backlog seedApril 2026Author · Sentinel

CAA / MAAJOINT APPROVALWORKBENCH

Live BVLOS autonomous swarm flight trials sit at the intersection of three unsettled regulatory regimes. This is the user story for an assurance-grounded workbench that lets a joint review panel issue — or defensibly refuse — a Military Permit to Fly with traceable rationale at speed.

audience:MAA RPAS Reg Branch·MAA-CTS·CAA AAA·CAA FOI·CAA Innovation Hub·Duty Holders·Range Safety
// the story
As
As

a Lead Inspector in the UK MAA's RPAS Regulatory Branch — sitting on a joint review panel with CAA Airspace, ATM & Aerodromes and Flight Operations Inspectorate counterparts — responsible for granting or refusing a Military Permit to Fly and the associated Flight Trials Instruction for a BVLOS autonomous swarm operating across both segregated Danger Area and non-segregated airspace.

I want
I want

an assurance-grounded joint Approval Workbench that ingests the safety case, JARUS SORA, AMLAS/SACE autonomy assurance, DAA evidence, C2 link analysis, swarm-behaviour envelope and CAP 1616 artefacts; cross-checks against RA 1000 / 1600 / 2300, DEFSTAN 00-970 Pt 9, CAP 722/A/B, JSP 936 Pt 1, and AOP-15 / Article 36; flags residual risk and weak traceability; and produces a structured decision record in engineering English.

So that
So that

I can issue — or defensibly refuse — an MPTF for the live BVLOS swarm trial with confidence that ALARP has been demonstrated, that Meaningful Human Control and Duty Holder accountability are intact, that third-party airspace users and the public below are not exposed to unacceptable risk, and that a transparent audit trail exists for the RTSA, DASB, Ministers, and any future Service Inquiry or AAIB investigation.

cat ./context.txt

Autonomous swarm flight trials sit in the awkward intersection between three unsettled regulatory regimes: BVLOS in non-segregated airspace (CAA, CAP 722B), RPAS airworthiness and flight test (MAA, RA 1600 / 2335), and AI/autonomy assurance (JSP 936 Pt 1, AMLAS, SACE). No single rulebook covers a swarm cleanly. Approvals today are bespoke, slow, and consume senior-inspector bandwidth that the UK has too little of — particularly with the Protector RG Mk1 transition, GCAP/Tempest early airframe work, and Project Alvina's attritable-mass experimentation all converging through 2026–2028.

// ls ./personas

THE PANEL

PRIMARY
MAA Lead Inspector

Chairs the joint review panel; signs the decision record.

CO-PRIMARY
CAA Inspector (AAA + FOI)

Airspace integration, third-party risk, CAP 1616.

SECONDARY
Delivery / Operating Duty Holder

Owns residual safety risk once Permit issued.

SECONDARY
RAISO (JSP 936 Pt 1)

Accountable for AI-assisted onboard decision-making.

SECONDARY
Range Safety Officer

Owns live trial execution on the day.

SECONDARY
Independent Technical Advisor

MAA-CTS / QinetiQ / DAIC — autonomy specialism.

// scope --boundary

BOUNDARY CONDITIONS

In scope
  • Ingestion of submission pack
  • Clause-level cross-check vs regulatory anchors
  • Gap & ambiguity flagging
  • AI/autonomy assurance posture review
  • DAA evidence assessment
  • Swarm envelope & emergent-behaviour challenge
  • Airspace & third-party risk review
  • Drafting structured decision record
Out of scope
  • ×Issuing the MPTF itself (human signature)
  • ×Autonomous clearance of live flight
  • ×Displacement of Duty Holder accountability
  • ×Displacement of RAISO sign-off
  • ×Automated redaction of safety evidence
  • ×Cross-border approvals outside UK
// acceptance --given-when-then

ACCEPTANCE CRITERIA

8 criteria · liftable into a backlog or specification-by-example workshop

AC1
AC1
Submission completeness & clause coverage

Identify required artefacts, map to RA 1600/2300, DEFSTAN 00-970 Pt9, CAP 722/A/B. Produce a red/amber/green coverage matrix with page-level pointers.

AC2
AC2
SORA / operational risk assessment

Verify GRC, ARC, M1–M3 mitigations, derived SAIL, and all 24 OSOs. Any unmet OSO blocks a green rating without explicit panel rationale.

AC3
AC3
Detect and Avoid evidence

Sensor performance, closure geometries, Well Clear derivation, ASTM F3442 / EUROCAE ED-267 alignment. Tested envelope must enclose operational envelope.

AC4
AC4
C2 link & loss-of-link behaviour

Link budget, latency, RF/GNSS resilience, crypto posture. Loss-of-link behaviours that need swarm consensus post-loss are flagged high-risk.

AC5
AC5
Swarm: emergent behaviour & containment

Simulation + live-flight coverage of the collective state space. Geofencing at agent and swarm level. Extrapolation triggers panel decision.

AC6
AC6
Airspace integration & CAP 1616

NOTAM, electronic conspicuity, TDA/TRA validity, third-party risk over inhabited areas. Out-of-volume legs cannot turn green.

AC7
AC7
AI / autonomy assurance posture

AMLAS / SACE-aligned case with ODD, data provenance, drift monitoring. JSP 936 Pt 1 RAISO sign-off. Meaningful Human Control explicit at every layer.

AC8
AC8
Decision record, conditions & triggers

Structured record: regulatory basis, evidence, residual risk, ALARP, conditions, review triggers, signatures. Every claim traceable to artefact + paragraph.

// non-functional --requirements

NON-FUNCTIONAL REQS

Traceability
Every clause resolves to the publication revision in force on review date.
Audit
Immutable log of every panel action; export read-only PDF on close.
Classification
OFFICIAL-SENSITIVE → SECRET → STRAP, with no cross-class egress.
Latency
Coverage matrix < 10 min; AI flags < 30 min; full review ≤ 2 working days.
Explainability
Plain engineering English — never model internals.
Accessibility
WCAG 2.2 AA inspector UI.
Resilience
Workbench outage never blocks an in-progress live trial.
Data sovereignty
UK sovereign infrastructure; no cloud egress without authorisation.
// assurance --anchors

REGULATORY ANCHORS

RA 1000
MAA Regulatory Principles · Duty Holder construct (RA 1020)
RA 1600
RPAS airworthiness, operation, training
RA 2325 / 2335
Flight Test Permits · Flight Test & Evaluation
DEFSTAN 00-970 Pt 9
Design requirements for RPAS
CAP 722 / 722A / 722B
UAS ops · operating safety cases · BVLOS
CAP 1616
Airspace Change Process
JARUS SORA v2.5
Specific Operations Risk Assessment · 24 OSOs
JSP 936 Part 1
Dependable AI in Defence · RAISO accountability
AMLAS / SACE
AI/ML & Autonomy assurance — University of York / CfAA
ASTM F3442 / EUROCAE ED-267
DAA technical framework
AOP-15 · Article 36
Where weapons are carried on the swarm
// risks --with-mitigations

RISKS & MITIGATIONS

risk
Automation bias among inspectors
mitigation
Override-first UI · mandatory panel deliberation · calibration audits.
risk
Regulatory drift between MAA and CAA
mitigation
Shared clause library under joint config control · quarterly alignment.
risk
Over-reliance on simulation for emergent behaviour
mitigation
Explicit live-flight coverage matrix · mandatory extrapolation flags.
risk
Classification handling breach
mitigation
Class-aware pipeline · marking enforcement · two-person review on downgrade.
risk
Third-party risk underestimation
mitigation
CAA FOI review gate on every non-segregated leg.
risk
Scope creep into MPTF issuance
mitigation
Workbench stops at the draft decision record — signature stays human.
cat ./definition-of-done.txt
  • AC1–AC8 demonstrated end-to-end on a representative BVLOS swarm submission.
  • Live joint MAA/CAA panel completed ≥ 1 approval without manual fallback.
  • Decision record accepted by the Duty Holder without rework on structure or traceability.
  • Independent assurance review (MAA-CTS / CAA Innovation Hub) signed off the Workbench's own assurance case.
  • Independent red-team probe (automation bias · evidence-fabrication · classification leakage · prompt-injection) completed.
  • Drift monitoring on the Workbench's AI components in place with owner assigned.
  • UNCLASSIFIED lessons-learned note published for the wider UK aviation regulatory community.