Slate benchmark target registry migration

Objective: Design Slate benchmark target registry migration; done when plan gates define clean cutover and proof; plan docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md.

Flow mode: one-shot execution

Goal plan: docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md

Template: docs/plans/templates/major-task.md

Primary template: docs/plans/templates/major-task.md

Applied packs:

• docs (docs/plans/templates/packs/docs.md)
• agent-native (docs/plans/templates/packs/agent-native.md)

Major source:

• type: user architecture direction
• id / link: local thread request
• title: Clean Slate benchmark and Autoresearch target structure
• decision to make: decide the long-term benchmark ownership model after .tmp/slate-v2 is merged back into Plate
• decision criteria: one benchmark source of truth, no Evidence Kit / Autoresearch split-brain, target-based Autoresearch setup, runtime benchmark code near measured runtime code, report/history generation outside active loop state

Major lane:

• lane: benchmark and performance architecture
• output type: first implementation slice plus migration plan
• implementation expected: yes, add target-registry spine and agent commands without deleting Evidence Kit yet
• affected packages / surfaces: benchmarks/targets/**, tooling/scripts/bench-targets.mjs, package.json, .agents/rules/slate-autoresearch.mdc, generated skill mirrors
• dominant risk: creating a third benchmark control plane instead of a cutover path

Completion threshold:

• benchmarks/targets/slate-v2.json exists with the active Evidence Kit artifact rows imported as target definitions.
• tooling/scripts/bench-targets.mjs can list, validate, run, and create Autoresearch setup plans by target id.
• Root package scripts expose bench:targets:* and slate:ar:setup-target.
• slate-autoresearch tells future agents to use the target registry first and treat Evidence Kit as legacy input/reporting.
• Verification commands pass: target check, target list smoke, Autoresearch setup-target smoke, script syntax check, JSON parse check, generated skill sync check, diff whitespace check, and goal-plan check.

Verification surface:

• command: pnpm bench:targets:check
• command: pnpm bench:targets:list | sed -n '1,12p'
• command: pnpm slate:ar:setup-target -- react-active-typing-breakdown | sed -n '1,80p'
• command: node --check tooling/scripts/bench-targets.mjs
• command: JSON parse for package.json and benchmarks/targets/slate-v2.json
• source-audit: rg for target registry guidance in source and generated skills
• command: git diff --check
• command: node .agents/skills/autogoal/scripts/check-complete.mjs docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md

Constraints:

• Start from repo evidence before external claims.
• Keep helper stack proportional.
• Separate measured evidence, source evidence, inference, and recommendation.
• Do not delete Evidence Kit until target reports/history can replace current generated outputs.
• Do not move runtime benchmark implementations in this slice; the future merge changes their final home.

Boundaries:

• Source of truth: benchmarks/editor/research/benchmark-registry.json, benchmarks/editor/research/evidence-source-map.md, benchmarks/editor/iterations/003-evidence-control-plane.md, .tmp/slate-v2/package.json, and current package scripts.
• Allowed edit scope: benchmarks/targets/**, tooling/scripts/bench-targets.mjs, package.json, .agents/rules/slate-autoresearch.mdc, generated skill mirrors, this plan.
• External sources: N/A; repo evidence settles this architecture slice.
• Browser surface: N/A; no browser UI changed.
• Tracker sync: N/A; no tracker item.
• Non-goals: deleting Evidence Kit, moving .tmp/slate-v2 benchmark code, changing benchmark measurement semantics, running expensive browser benchmarks.

Output budget strategy:

• Read only targeted benchmark docs, registry rows, package scripts, and skill sections.
• Cap source reads with sed -n and target rg; do not stream generated perf wiki or full benchmark result JSON.
• Use CLI smoke output slices for setup/list commands.

Blocked condition:

• Block only if target import cannot preserve active artifact ids/commands, Autoresearch cannot accept target-derived setup plans, or generated skill sync fails and the source/generator owner is unclear.

Major state:

• task_type: major
• task_complexity: major
• current_phase: closeout
• current_phase_status: complete
• next_phase: none
• goal_status: complete after final goal-plan check passes

Current verdict:

• verdict: implement target-registry spine now, freeze Evidence Kit later
• confidence: 0.88
• next owner: benchmark target migration follow-up for reports/history replacement
• reason: target registry removes the active-control split while preserving current benchmark evidence during migration

Completion rule:

• Do not call update_goal(status: complete) until every completion threshold above is satisfied, final evidence is recorded, and node .agents/skills/autogoal/scripts/check-complete.mjs docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md passes.
• Do not create hook state for this goal. This file plus the active goal are the durable state.

Start Gates:

Gate	Applies	Evidence
major-task loaded	yes	Read .agents/skills/major-task/SKILL.md.
Active goal checked or created	yes	get_goal returned no goal; create_goal created this objective.
Source of truth read before analysis	yes	Read Evidence Kit source map, Evidence Kit control-plane note, active registry, health output, .tmp/slate-v2 scripts.
Major lane selected	yes	Benchmark and performance architecture.
Decision criteria stated	yes	One registry, target-based Autoresearch setup, no active Evidence Kit split-brain, runtime code follows runtime.
Existing repo patterns / prior decisions checked	yes	Evidence Kit memory and docs showed current active artifact registry and generated health/report flow.
Helper stack selected	yes	autogoal, major-task, docs-creator, performance-oracle, agent-native-reviewer; no external research needed.
External research decision recorded	yes	N/A: repo evidence is enough.
Implementation expectation recorded	yes	First implementation slice adds target registry and commands, not full deletion.
Workspace authority selected	yes	Root plate-2 owns control scripts and generated skills; .tmp/slate-v2 owns runtime benchmark commands.
Branch / PR expectation decided	yes	N/A: user did not ask for branch, commit, push, or PR.
Output budget strategy recorded	yes	See Output budget strategy.
Docs pack selected	yes	This plan and target README are docs/supporting surfaces.
docs-creator loaded	yes	Read .agents/skills/docs-creator/SKILL.md.
Docs lane selected	yes	Supporting docs under major-task, not docs-dominant.
Target docs and nearest sibling docs read	yes	Read Evidence Kit source map, Evidence Kit control-plane note, generated benchmark health excerpt.
Docs style doctrine read	yes	Read docs-creator voice/current-state rules.
Documented source owner identified	yes	Target registry README documents benchmarks/targets/slate-v2.json and tooling/scripts/bench-targets.mjs.
Agent-native pack selected	yes	Package scripts and .agents/rules/slate-autoresearch.mdc changed.
Agent-facing action surface identified	yes	pnpm bench:targets:*, pnpm slate:ar:setup-target, and slate-autoresearch skill text.
Source rule versus generated mirror boundary identified	yes	Edited .agents/rules/slate-autoresearch.mdc; ran pnpm install to regenerate .agents/skills/** and .claude/skills/**.
agent-native-reviewer loaded or waiver recorded	yes	Read .agents/skills/agent-native-reviewer/SKILL.md; action parity checked by command scripts and skill discoverability.

Work Checklist:

• Short objective plus outcome, completion threshold, verification surface, constraints, boundaries, and blocked condition are concrete.
• Major source records source type, id/link, title, decision type, expected outcome, decision criteria, likely files/packages/surfaces, browser surface, and highest-leverage owner.
• Current state is mapped before proposing a new architecture, migration, benchmark, or plan.
• Existing repo patterns, prior decisions, and nearby implementation constraints are recorded before external research.
• External docs or source are used only where repo evidence does not settle the question, or N/A reason is recorded.
• Options, recommendation, tradeoffs, blast radius, and rejection reasons are recorded.
• Facts, inference, and recommendation are separated.
• Review or pressure lenses are selected and completed, or marked N/A with reason.
• If implementation happens, touched-surface packs cover docs and agent-native surfaces.
• Workspace authority recorded: every proof command names the cwd/tool that owns the analyzed or changed behavior.
• Output budget discipline recorded and followed: broad searches are scoped, capped, counted, or artifacted instead of streamed into goal context.
• Accepted/actionable review findings are fixed or explicitly rejected with evidence.
• Docs pack: docs lane, target docs, nearest sibling docs, and source owner are recorded.
• Docs pack: every named API, import, option, route, component, transform, demo, and preview is source-backed or marked N/A with reason.
• Docs pack: docs use current-state reference voice, not changelog voice.
• Docs pack: links, anchors, and previews target real leaf pages or are marked N/A with reason.
• Agent-native pack: source-of-truth rule files are edited instead of generated skill mirrors.
• Agent-native pack: the changed agent action is discoverable from the skill/rule text.
• Agent-native pack: generated mirrors are synced when .agents/rules/** changed.
• Agent-native pack: accepted agent-native review findings are fixed or explicitly rejected with reason.

Completion Gates:

Gate	Applies	Required action	Evidence
Named verification threshold	yes	Run the repo audit and command smoke checks named in this plan	pnpm bench:targets:check, pnpm bench:targets:list, pnpm slate:ar:setup-target, syntax/JSON checks passed.
Current-state source audit	yes	Map current owner, boundaries, constraints, and affected surfaces	Findings record Evidence Kit active state, 23 active artifacts, 35 ignored historical artifacts, and .tmp/slate-v2 runtime ownership.
Decision criteria closure	yes	Mark each criterion satisfied, narrowed, rejected, or blocked with evidence	Target registry and CLI satisfy first-slice criteria; full Evidence Kit deletion intentionally deferred.
Options / tradeoffs / rejection record	yes	Record viable options, chosen recommendation, and why alternatives lose	Decisions and tradeoffs section records options A-D.
Review / pressure pass	yes	Run selected reviewer/lens or record N/A with reason	Performance-oracle lens applied manually to target registry complexity; agent-native reviewer loaded; no external reviewer subprocess available.
Review findings closure	yes	Fix or explicitly reject accepted/actionable findings and record closure proof	Fixed pnpm -- target parsing and Evidence Kit path-base import bug.
External-source audit	no	Cite official/local clone/external sources when used, or record N/A	N/A: local repo evidence is sufficient.
Implementation gates	yes	If code changed, close primary-template and touched-surface gates; otherwise N/A	Code/docs/agent gates closed in this plan.
Final handoff contract	yes	Record recommendation, evidence, caveats, residual risk, and next owner	Final handoff contract section filled.
Final lint	yes	Run scoped equivalent when files changed	git diff --check will run after final plan edit.
Output budget discipline	yes	Verify no unbounded high-volume command output was streamed, or record the accidental output and recovery	Used targeted sed, rg, JSON summaries, and capped command output.
Goal plan complete	yes	Run node .agents/skills/autogoal/scripts/check-complete.mjs docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md	Final command is the last closure gate and is recorded in Verification evidence.
Docs source-backed claim audit	yes	Verify docs claims against current source or record N/A	benchmarks/targets/README.md claims match bench-targets.mjs, package scripts, and target registry.
Docs links / routes / previews	no	Verify leaf links, routes, anchors, and preview names or record N/A	N/A: no route, anchor, or preview added.
Docs MDX/content parser	no	Run content build for MDX/content changes, or record N/A	N/A: only markdown plan/README changed.
Plugin page specifics	no	Apply docs-creator kit/manual/API rules; otherwise N/A	N/A: no plugin docs page.
Agent source / generated sync	yes	Run pnpm install when .agents/rules/** changed and verify generated mirrors	pnpm install regenerated Skiller output.
Agent action discoverability	yes	Source-audit the skill/rule path an agent will read	rg verifies Target Registry and pnpm slate:ar:setup-target in source and generated skills.
Agent-native review	yes	Load .agents/skills/agent-native-reviewer/SKILL.md and close accepted findings, or record N/A	Loaded; no orphan agent action remains because root scripts and skill text expose target setup/list/check.

Phase / pass table:

Phase	Status	Evidence	Next
Intake and source read	complete	Read goal skill, major-task skill, Evidence Kit docs/registry/health, package scripts.	current-state map
Current-state map	complete	Evidence Kit owns active artifact list and generated reports; .tmp/slate-v2 owns runtime benchmark commands.	options
Options and recommendation	complete	Chose target registry as source; rejected Evidence Kit or Autoresearch as sole active owner.	implementation
Review / pressure pass	complete	Performance and agent-native lenses applied; CLI smoke found two real migration bugs.	verification
Implementation or plan artifact	complete	Added target registry, CLI, README, package scripts, and skill guidance.	verification
Verification	complete	Target check/list/setup and syntax/JSON checks passed; final diff/check-complete remains after last edit.	closeout
Closeout	complete	Final plan and final commands recorded.	final response

Findings:

• Fact: Evidence Kit currently records active benchmark claims in benchmarks/editor/research/benchmark-registry.json.
• Fact: Current Evidence Kit docs say unregistered benchmark JSON files are historical output and not active evidence.
• Fact: Current health output reports 23 active artifacts, 2 optional missing artifacts, and 35 ignored unregistered historical artifacts.
• Fact: .tmp/slate-v2/package.json owns runtime benchmark scripts such as bench:react:active-typing:local, bench:react:huge-document:browser-trace:local, and core benchmark scripts.
• Fact: The future merge means benchmark code should follow runtime/package ownership, not stay in .tmp/slate-v2 by identity.
• Inference: Evidence Kit is valuable as migration input and report archive but harmful as a second active control plane after Autoresearch target setup exists.
• Recommendation: make benchmarks/targets/slate-v2.json the source for benchmark decisions, with Autoresearch consuming target ids and generated reports/history reading the same target registry.

Decisions and tradeoffs:

• Option A: Keep Evidence Kit active. Rejected because it creates split-brain with Autoresearch and benchmark target commands.
• Option B: Make Autoresearch the only source of truth. Rejected because Autoresearch owns active loop state, not long-lived benchmark inventory/history.
• Option C: Make package scripts the source of truth. Rejected because package scripts do not carry question, metric, correctness, artifact, or docs metadata.
• Option D: Typed/structured target registry plus consumers. Accepted because it gives one product contract and lets benchmark runners, Autoresearch, and reports consume the same target id.
• First-slice tradeoff: import rows from Evidence Kit instead of deleting it. This preserves active evidence while creating the new spine.
• Later cutover: report generation should move from Evidence Kit-specific code to target/result history; then Evidence Kit can be frozen or deleted.

Implementation notes:

• Added tooling/scripts/bench-targets.mjs with list, check, run, autoresearch-setup, and import-evidence-kit.
• Added benchmarks/targets/slate-v2.json with 23 imported targets.
• Added benchmarks/targets/README.md documenting ownership, commands, and target contract.
• Added root scripts: bench:targets:list, bench:targets:check, bench:targets:run, bench:targets:import-evidence-kit, and slate:ar:setup-target.
• Updated .agents/rules/slate-autoresearch.mdc so future agents use target ids first and treat Evidence Kit as legacy input/reporting.
• Ran pnpm install so .agents/skills/slate-autoresearch/SKILL.md and .claude/skills/slate-autoresearch/SKILL.md match the source rule.

Review fixes:

• pnpm slate:ar:setup-target -- react-active-typing-breakdown initially treated -- as the target id. Fixed by stripping the pnpm separator in bench-targets.mjs.
• Evidence Kit import initially resolved ../../.tmp/slate-v2 relative to benchmarks/editor/research, producing benchmarks/.tmp/slate-v2. Fixed importer base to benchmarks/editor and regenerated the target registry.

Error attempts:

Error / failed attempt	Count	Next different move	Resolution
pnpm separator treated as target id	1	Strip leading -- from CLI args	Fixed and verified with pnpm slate:ar:setup-target -- react-active-typing-breakdown.
Evidence Kit path base resolved from wrong directory	1	Resolve imported paths from benchmarks/editor	Fixed and regenerated benchmarks/targets/slate-v2.json.

Verification evidence:

• /Users/zbeyens/git/plate-2: pnpm install -> passed and regenerated Skiller output.
• /Users/zbeyens/git/plate-2: pnpm bench:targets:check -> benchmark-targets ok: 23 targets.
• /Users/zbeyens/git/plate-2: pnpm bench:targets:list | sed -n '1,12p' -> printed target ids and commands.
• /Users/zbeyens/git/plate-2: pnpm slate:ar:setup-target -- react-active-typing-breakdown | sed -n '1,80p' -> returned Autoresearch setup-plan JSON with typing_seconds and target benchmark command.
• /Users/zbeyens/git/plate-2: node --check tooling/scripts/bench-targets.mjs -> passed.
• /Users/zbeyens/git/plate-2: JSON parse for package.json and benchmarks/targets/slate-v2.json -> passed.
• /Users/zbeyens/git/plate-2: generated skill source audit for target registry guidance -> passed; Target Registry, pnpm slate:ar:setup-target, and Evidence Kit legacy guidance are present in source and generated skills.
• /Users/zbeyens/git/plate-2: git diff --check -- tooling/scripts/bench-targets.mjs benchmarks/targets package.json .agents/rules/slate-autoresearch.mdc .agents/skills/slate-autoresearch/SKILL.md .claude/skills/slate-autoresearch/SKILL.md docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md -> passed.
• /Users/zbeyens/git/plate-2: node .agents/skills/autogoal/scripts/check-complete.mjs docs/plans/2026-06-01-slate-benchmark-target-registry-migration.md -> [autogoal] complete.

Final handoff contract:

• Recommendation: keep target registry as the benchmark product contract; make runners, Autoresearch, and reports consume it.
• Confidence: 0.88.
• Evidence: 23 imported active targets, passing registry check, setup-target smoke, source/generated skill guidance.
• Tests / commands: target check/list/setup, syntax, JSON parse, generated skill audit, diff check, goal-plan check.
• Browser proof: N/A; no browser surface changed.
• PR / tracker: N/A.
• Caveats: benchmark commands still mostly use wrapper timing until individual targets emit native METRIC/ARTIFACT lines.
• Next owner: replace Evidence Kit report generation with target/result history, then freeze/delete Evidence Kit active package.

Timeline:

• 2026-06-01T10:20:20.368Z Major-task goal plan created.
• 2026-06-01T10:21:00Z Active goal created with short objective.
• 2026-06-01T10:24:00Z Read Evidence Kit source map, control-plane note, registry, and health output.
• 2026-06-01T10:30:00Z Added benchmark target CLI and imported 23 active targets.
• 2026-06-01T10:33:00Z Added root scripts and target README.
• 2026-06-01T10:36:00Z Updated slate-autoresearch source rule and regenerated skill mirrors.
• 2026-06-01T10:39:00Z Fixed pnpm separator handling and Evidence Kit path-base import.
• 2026-06-01T10:43:00Z Ran target, generated-skill, syntax, JSON, and diff checks.
• 2026-06-01T10:44:00Z Ran final autogoal completion check successfully.

Reboot status:

Question	Answer
Where am I?	Closeout complete after first implementation slice.
Where am I going?	Complete the goal if final goal-plan check passes.
What is the goal?	Establish the clean long-term benchmark target registry migration spine and proof.
What have I learned?	Evidence Kit can be imported as legacy input, but should not remain the active control plane.
What have I done?	Added target registry, CLI, commands, docs, and agent guidance.

Open risks:

• Individual target metrics still need native METRIC/ARTIFACT output for best Autoresearch fidelity.
• Evidence Kit deletion is intentionally deferred until reports/history consume the target registry.