Applications begin with questions.

Hypothesis / not demonstrated

An authenticated, sequence-numbered outer envelope could identify forged or replayed state reports within an agreed computation and bandwidth budget. AEA does not authenticate senders; AEA-AUTH/1 is not built.

Experiment

Implement the draft envelope with HMAC-SHA256 or Ed25519 reference vectors. Test reordered, delayed and forged records against a defined attack corpus.

Measure

False accepts of forged or replayed records, plus computation and link cost.

Failure condition

The envelope exceeds the target platform’s CPU or link budget. Hardware timing is not yet measured.

Hypothesis / not demonstrated

AEA does not detect or bound drift or perform dead-reckoning. H2 asks whether a separate estimator could detect abnormal sensor drift sooner by comparing propagated state with earlier trusted records.

Experiment

Use public GNSS-outage datasets with simulated sensor faults and bias. Add a separate future uncertainty/covariance profile; do not change the current record format.

Measure

Detection latency and false-alarm rate against innovation-gating baselines.

Failure condition

Detection latency or false-alarm rate is no better than the baseline.

Hypothesis / not demonstrated

AEA does not reject spoofed packets or provide anti-spoofing. H3 asks whether a future signed exchange and relative measurements could help several nodes identify an inconsistent or malfunctioning peer.

Experiment

Simulate three or more nodes with one emitting false or degraded position tracks. This requires the unbuilt H1 profile and relative range-measurement models.

Measure

Receiver operating characteristic area under the curve for identifying false peer reports.

Failure condition

No statistically significant improvement over local range-residual filters.

Hypothesis / not demonstrated

Mandatory reference-frame and time-scale tags might reduce reconciliation errors compared with untyped CSV or JSON workflows. This benefit has not been demonstrated.

Experiment

Inject frame and time-scale label errors into multi-vehicle logs. Compare detection and alignment with legacy workflows.

Measure

Frame and epoch error-detection rate and automated alignment speed.

Failure condition

Tags do not measurably reduce ingestion errors, or conversion overhead negates the benefit.

Hypothesis / not demonstrated

Record sealing and verification might fit the target platform’s timing and stack budgets. Hardware timing, stack depth and on-target measurements are not yet measured.

Experiment

Measure cycle counts and stack high-water marks on physical Cortex-M boards running a bare-metal C caller.

Measure

Execution latency, stack high-water mark and resource use against the predeclared budgets in H5.

Failure condition

Measured stack or execution latency exceeds the H5 limits. Numeric hardware targets remain in the source research plan, not a claimed result.

Hypothesis / not demonstrated

Local sealing might support post-test checks of stored records after communications loss. AEA does not maintain navigation through a blackout or detect SEUs.

Experiment

Use hardware-in-the-loop flash logging with simulated link loss, fault injection and power interruptions. Read records back after the test.

Measure

Detection rates for corrupted, truncated or incomplete records, plus dropped records and buffer occupancy.

Failure condition

Memory or write-latency overhead causes buffer overflows or dropped records at the proposed logging rate.