MesoNet: benchmark data reconciliation

these tables need run IDs. match split + settings + version before picking numbers from different drafts

There are at least two result-table versions. The native MesoNet Data sheet reports MNIST accuracy 98.11% and CIFAR-10 56.6% for MesoNet. JSHS Data Tables contains an earlier summary with MNIST 96.7% and CIFAR-10 53.47%. Other baseline values and energy figures also differ. Those differences require experiment/version identifiers; they should not be silently merged into one result.

parameter reference in the JSHS sheet: encoding lengths 300/350/450 for MNIST/CIFAR-10/CIFAR-100, membrane time constant 2 ms, synapse time constant 1 ms, and listed learning rates 0.01/0.005/0.001 in a separate table. The sheet does not consistently state units for every encoding-length field or explain how adaptive plasticity relates to those listed rates.

Energy comparison

Model comparison plots

• provenance check: JSHS Data Tables also contains text requesting generation of an operations graph and CSV beside the operations table. This does not establish which values are illustrative versus measured. Locate raw run outputs before presenting these columns as empirical computation counts.

• review questions: per-model commit, seed, split, evaluation/readout procedure, repeat count behind standard deviations, and energy-measurement method. The Evaluation Report extract contains chart headings but not the chart data. This card is for internal reconciliation; do not publish disputed measurements as established results.

Benchmark result tables