labview-benchmark-actor — Specification Package

Status: Originating specification. The extension is now implemented and ships as ext-v* GitHub Releases (latest ext-v0.4.0). Standards baseline: repo-standards-review v0.2.19 (GitLab svelderrainruiz/repo-standards-review, commit d44f210d). Stamp rationale: this package is authored to move to a future labview-benchmark-actor repository; the standards-release stamp travels with it so the receiving repo can re-validate against the exact baseline it was written to.

Purpose

labview-benchmark-actor is a standalone, installable VS Code extension providing hooking and agentic benchmarking infrastructure (originally prototyped in vi-history-suite, since extracted into this independent repo — no runtime dependency on that prototype). Installed on a GitHub Codespace or a Vagrant golden VM, it lets an operator drive benchmarking through that agentic infrastructure and review results through a time-cursor benchmark viewer.

Five capabilities define the extension:

1. Benchmark time-cursor UI — a benchmark chart with a draggable vertical time cursor; dragging it left↔right selects a point in time, and the captured picture (frame) indexed at that time is shown directly below the chart, keeping metric and visual evidence synchronized.
2. TCP/UDP coordination bus — multiple Vagrant VMs, each running the extension, coordinate over a local TCP + UDP message bus instead of a GitHub Discussion, so benchmarking runs offline / air-gapped and in parallel across VMs. The bus carries inter-actor communication only — no run data ever crosses it.
3. VM cleanroom storage — a run's data (metrics and pictures) is stored locally in each VM via the mprr ring buffer model (absorbed in-repo, dependency-free — see ADR-0009); no run data crosses the bus — the whole ring buffer is VM-local (LBA-REQ-009, ADR-0005).
4. Own-run review + host ollama comparison — each actor reviews only its own previous runs; there is no cross-VM comparison. The operator concentrates runs onto the host (out-of-band) to improve an ollama comparison layer over previous runs (LBA-REQ-010, ADR-0006).
5. MCP tools for agent mode — the extension contributes a Model Context Protocol server (a dependency-free stdio JSON-RPC server) so Copilot agent mode can call the extension's own tools directly: host capabilities, the deterministic benchmark series, and the coordination bus (poll / post).

Absorbed model

• mprr — the bounded-RAM dual-packet ring buffer model (dual-packet policy from mprr ADR-0024) and the frozen TDMS-compatible 1.0 replay transport, absorbed in-repo as dependency-free mirrors under experiments/mprr-ring/ and exercised by experiments/verify-local-gates.mjs. The mprr name is retained for the local model; labview-benchmark-actor owns it and does not track an external repository (ADR-0009).

Standards coverage

Standard	Lane	Package artifact
ISO/IEC/IEEE 29148	Requirements	docs/requirements/srs.md
ISO/IEC/IEEE 42010	Architecture description	docs/architecture/overview.md
ISO/IEC/IEEE 29119-2/3	Test	docs/testing/test-plan.md
ISO 10007 / ISO/IEC/IEEE 12207	Configuration management & release	docs/cm/cm-plan.md
ISO/IEC/IEEE 26514:2022	Information for users	docs/information-for-users/user-guide.md
ISO/IEC/IEEE 15289	Information item map	docs/information-item-map.md

Cite standards as Std §clause throughout; keep observations separate from assumptions; show repo-relative evidence for load-bearing claims.

Requirement ID scheme

Software requirements use LBA-REQ-NNN; acceptance criteria are cited by position as LBA-REQ-NNN.M (matching the parent repo's derived-position convention).

Move / graduation note

When this package graduates to the labview-benchmark-actor repository:

• Re-run python3 scripts/pipeline.py validate-skill against repo-standards-review v0.2.19 (or bump this stamp and re-validate).
• Preserve the docs/ layout so the standards runner resolves each lane.
• Carry the requirement IDs unchanged so external traceability survives the move.