Request for review: SwarmBox-style analyzer projection in an external coordination benchmark

[minustwo]

Hello,

We prepared an independent public benchmark artifact for a semantic consistency kernel called MST-NF:

https://github.com/minustwo/mstnf-drone-swarm-artifact

This is a review request for the benchmark mapping, not a request to merge code upstream. The artifact is not a flight-control system and does not invoke ROS, PX4, radio links, autopilots, hardware, or physical-drone command paths.

The artifact uses SwarmBox as the analyzer/reporting shape. It projects committed semantic coordination decisions into the top-level result fields used by the SwarmBox analyzer:

• config
• flight_distance
• mission_duration
• utilization_rate
• idle_time
• workload_time
• workload_distance

Relevant artifact paths:

• docs/EXTERNAL_SWARM_SOURCE_MAP.md
• docs/REPRODUCIBILITY.md
• docs/THIRD_PARTY_REVIEW.md
• src/adapters/swarmbox/analyzer-projection.js
• reports/latest-external-standard-check.md

Reproduction command:

git clone https://github.com/minustwo/mstnf-drone-swarm-artifact.git
cd mstnf-drone-swarm-artifact
./scripts/reproduce.sh

Current artifact result summary:

• npm test: 10/10 pass
• main backend-attempt ratio: 10.81x
• seed-sweep median backend-attempt ratio: 10.821x
• SwarmBox-style result shape present: true
• MST-NF unsafe reservations: 0
• artifact assertions: pass

Questions for review:

1. Is this SwarmBox-style analyzer field projection useful for artifact comparison?
2. Are there additional SwarmBox metrics that should be included to make the benchmark review fairer?
3. Are the boundary clauses clear enough to prevent confusion with a full PX4/ROS SwarmBox run?

If issues are not the right place for this kind of review request, please feel free to close or redirect it. Thank you for any feedback on the benchmark mapping.

[minustwo]

Update: the public artifact has been tightened from a general benchmark artifact into a Layer B third-party evaluation artifact / evaluation harness.

The acceptance criteria are now explicit and reviewer-facing:

• reproducible: npm run verify runs the full artifact with fixed seeds
• externally anchored: Crazyswarm2-style collision fixture and SwarmBox-style analyzer projection are documented with source paths
• safety bounded: not flight control, not path planning, not collision-avoidance certification, not real drone control, no physical drone command path
• claim measurable: useful reservations preserved, backend attempts reduced, CAS/retry/duplicate backend attempts/priority inversions eliminated, replay divergence = 0, invariant failures = []
• third-party falsifiable: non-zero replay divergence, invariant failures, unsafe reservations, MST-NF external-fixture collision, or missing analyzer shape are failures

New docs:

• https://github.com/minustwo/mstnf-drone-swarm-artifact/blob/main/docs/ACCEPTANCE_CRITERIA.md
• https://github.com/minustwo/mstnf-drone-swarm-artifact/blob/main/docs/RELEASE_LEDGER.md

Latest public artifact commit:

884dde32ad1afc445ffa99ebf38f5d685a81a308