feat: register pending setup baseline campaign

campaigns.yaml

@@ -1,6 +1,15 @@
 version: 1
 last_checkin: '2026-10-04T17:18:38Z'
 campaigns:
+- id: setup-baseline
+  title: Setup baseline collection and scientific acceptance
+  status: needs-decision
+  updated: '2026-10-06'
+  next: 'Specification is draft (autolens_profiling#384); no baseline collection or acceptance authorized.
+    Resolve immutable software and dependency lock, dataset hashes/settings, hardware/threads/load and
+    timing/cache/memory methodology/witnesses; freeze the manifest before separately authorizing collection.
+    Compile probe builder remains unavailable; archived timings are context only.'
+  evidence: https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/wiki/campaigns/setup_baseline.md
 - id: point-source-image-cpu
   title: Point-source image plane · CPU
   status: active
@@ -83,6 +92,17 @@ campaigns:
     in Mind. Retain parked; read the project ledger before proposing new work.
   evidence: https://github.com/PyAutoLabs/autolens_profiling/blob/a93f37a7ade16fcb7673a7777dfde505183728ed/wiki/campaigns/certified_positive_solver.md
 tasks:
+- id: setup_baseline_campaign
+  campaign: setup-baseline
+  title: Freeze the setup baseline specification, then collect and review evidence
+  status: needs-decision
+  priority: high
+  path: tasks/setup_baseline_campaign.md
+  updated: '2026-10-06'
+  next: 'Human resolves draft manifest choices and freezes the specification; collection requires separate
+    compute authorization. The report prepares evidence for human scientific review and never promotes
+    baselines. CPU production arrays use ral only; compile capability gap remains open.'
+  evidence: https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/baseline/campaign.json
 - id: runtime_cell_single_jit_gpu_warmup
   campaign: point-source-source
   title: Runtime cells' A100 `single_jit` includes the post-compile warm-up — source-plane 0.642 ms vs a steady

dashboard.md

@@ -1,6 +1,6 @@
 # PyAutoPulse — profiling dashboard
 
-<!-- pulse-campaigns:77ca7c41ddc4b75d3a81b678afe72bb4a5d261fa62678fa94e5331b020b7b65c -->
+<!-- pulse-campaigns:cbea1eba629dc0064077ad5d1af613005d3a58f8a6d17b54b253eb6468e6d379 -->
 ## Profiling Check In
 
 <details><summary>Full check-in prompt</summary>
@@ -15,6 +15,18 @@ Last check-in: 2026-10-04T17:18:38Z (review date, not measurement freshness).
 
 ## Active campaigns
 
+<details><summary>Setup baseline collection and scientific acceptance — needs-decision</summary>
+
+Reviewed 2026-10-06. Specification is draft (autolens_profiling#384); no baseline collection or acceptance authorized. Resolve immutable software and dependency lock, dataset hashes/settings, hardware/threads/load and timing/cache/memory methodology/witnesses; freeze the manifest before separately authorizing collection. Compile probe builder remains unavailable; archived timings are context only.
+
+[Campaign evidence](https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/wiki/campaigns/setup_baseline.md)
+
+### Active tasks
+
+- [Freeze the setup baseline specification, then collect and review evidence](https://github.com/PyAutoLabs/PyAutoPulse/blob/main/tasks/setup_baseline_campaign.md) — needs-decision: Human resolves draft manifest choices and freezes the specification; collection requires separate compute authorization. The report prepares evidence for human scientific review and never promotes baselines. CPU production arrays use ral only; compile capability gap remains open.
+
+</details>
+
 <details><summary>Point-source image plane · CPU — active</summary>
 
 Reviewed 2026-10-04. Phases 1–4 shipped; PyAutoLens#764 (extent warning) released in 2026.10.4.1. Next: docs-only autolens_profiling PR writing campaign completion evidence and fixing stale wiki/index lines (#763 &#x27;in flight&#x27; → #764 merged). RAL leftover folders present; mirror sync unverified.

tasks/setup_baseline_campaign.md

@@ -0,0 +1,71 @@
+# Freeze the setup baseline specification, then collect and review evidence
+
+Campaign: setup-baseline
+Task: setup_baseline_campaign
+Reviewed: 2026-10-06
+
+Current scheduling state lives in `campaigns.yaml`. This is a pending scientific
+campaign task, not authorization to run jobs or accept measurements. The project
+specification is being implemented under
+[autolens_profiling#384](https://github.com/PyAutoLabs/autolens_profiling/issues/384);
+those specification changes do not complete baseline collection or scientific
+acceptance.
+
+## Project specification and evidence
+
+- [Draft campaign manifest](https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/baseline/campaign.json)
+- [Baseline procedure and report contract](https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/baseline/README.md)
+- [Project campaign ledger](https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/wiki/campaigns/setup_baseline.md)
+- [Compile route capability](https://github.com/PyAutoLabs/autolens_profiling/blob/62a6ad1e82735e360c7cf25be62253a286493b17/catalogue/script_routes.json)
+- [Pending compile-builder task](jax_compile_probe_needs_own_cell_builder.md)
+- [Measurement noise audit](timing_noise_audit.md)
+
+The specification links pin the reviewed Phase 6 implementation commit. This
+is still a draft, not the later frozen campaign. Record the separately reviewed
+frozen manifest identity and project revision before collection. Existing archived measurements remain historical context, without
+establishing a fresh campaign baseline or a temporal performance trend. Pulse
+keeps its existing receipts and snapshots; this task creates no new measurements.
+
+## Human decisions before freezing
+
+1. Choose full immutable revisions for the profiling project and every library,
+   and an immutable dependency lock including Python, JAX/jaxlib and numerical
+   dependencies. Record how the executable environment reproduces that lock.
+2. Specify each selected setup's dataset content hashes and full scientific
+   settings, including model, mask/grid, precision, solver, sampling and all
+   relevant defaults. Decide which cells and evidence axes are required.
+3. Name the exact CPU/GPU hardware and backend, thread allocation, process
+   placement and host-load conditions for each collection leg. Production CPU
+   arrays must use `--partition=ral` only. GPU collection needs its own explicit
+   device/resource allocation and authorization.
+4. Fix timing, synchronization, repetition and uncertainty methodology;
+   distinguish compile, warm-up and steady runtime, declare cache state, memory
+   metric/measurement boundaries and the necessary correctness/parity witnesses.
+   Keep time, compile cost, memory and correctness as separate evidence axes.
+5. Decide how to handle unavailable capabilities before collection. The compile
+   probe's per-cell builder is explicitly unavailable in `catalogue/script_routes.json`.
+   Archive coverage does not make the probe runnable. Repair it through its
+   separate bounded development task, or record the gap and obtain a human scope
+   decision; never mark compile evidence complete from archived records.
+
+## Next bounded steps and acceptance
+
+1. Have the human resolve the choices above in the project's draft manifest and
+   procedure, review its validation and record the frozen manifest identity plus
+   the immutable project specification revision. Preserve the decisions in the
+   project campaign ledger. A draft or a freeze alone authorizes no compute.
+2. Request separate human compute authorization for the named cells, resources
+   and measurement methods. Then use the project's drivers to collect against
+   that frozen specification, preserving result identities, provenance and
+   incomplete/invalid evidence. Record actual execution outcomes in the project.
+3. Run the project's report against that frozen manifest and collected evidence.
+   Treat its output as readiness for human scientific review; report generation
+   never promotes a baseline, moves a pin or grants scientific acceptance.
+4. Have the human judge the matched evidence and record acceptance, rejection or
+   requested follow-up for each selected setup and evidence axis in the project
+   ledger. Only then update this pending task from the recorded decision. Any
+   later baseline/pin promotion requires its own explicit authorization.
+
+Until those steps are complete, retain `needs-decision`: no campaign is reported
+as running or accepted, and specification implementation remains distinct from
+the later collection and review deliverables.