Last check-in: 2026-10-18T17:00:00Z.
Active campaigns
| Campaign | Recent progress | Next steps |
|---|---|---|
| HST imaging, clean slate: Nautilus base on A100, jax_cpu and numba_cpu at a pinned release, mass stage with linear lens light and with it fixed, then other samplers on the mass stage | 2026-10-19 (checkins/2026-10-19T171637Z.md, focused clean-up): 412539 (jax_cpu dense) COMPLETED. On the human's decision, 412541 (jax_cpu sparse) and 412565 (numba_cpu sparse) were cancelled with partial stages left on disk. jax_cpu sparse: seed 0 has slam_base complete; seed 1 has source_lp[1] and source_pix[1] complete. numba sparse: slam_base complete on both seeds. All lightfixed mass_total[1] stages are incomplete. autolens_inference#36 (flatten, closes #35) merged as c3ca4de 2026-10-19T15:53:33Z. flatten_cleanslate.py --apply ran on RAL (28 moves, 55 rewrites) and on the laptop mirror (28 moves, 38 rewrites). Stage output now lives at output/autolens/imaging/hst/slam/<variant>/<config>/seed_<n>/<stage>/, prepared problems at output/autolens/prepared/<config>/seed_<n>/<variant>/ and rows at results/slam/imaging/hst/<variant>/<config>/. RAL and laptop each hold 1974 files with identical paths and sizes. The wave-1 attempt-1 material (output/experiments, output/experiment_sources, results/searches/experiments) was moved aside to /mnt/ral/jnightin/inference/aside/autolens_inference_cleanslate_wave1_attempt1/ on RAL and to /mnt/c/Users/Jammy/Science/inference/aside/autolens_inference_cleanslate_wave1_attempt1/ on the laptop. Earlier (2026-10-18): A100 base complete; the sparse fixed-light mass_total[1] diverged on both seeds (Cortex observation, human reading pending). | Nothing is queued (scheduler empty after the human-directed cancel 2026-10-19 ~15:51Z). Each step needs human authorisation: (1) Confirm keeping output/autolens/prepared/ (agent proposal, not yet confirmed: wave.py needs output/autolens/prepared/<config>/seed_<n>/<variant>/manifest.json for every sampler wave and the divergence repro needs it; only re-running the base mass stage regenerates it). (2) File a Mind bug task and run a laptop repro of the jax_gpu sparse fixed-light mass_total[1] divergence (both seeds) before any CPU or sampler wave uses the sparse fixed-light problem. (3) Rerun the cancelled sparse legs under the flat layout: jax_cpu sparse seed 1 from source_pix[2], jax_cpu sparse lightfixed mass_total[1] (both seeds), numba_cpu sparse lightfixed mass_total[1] (both seeds). (4) GPU wave 1 attempt 2 (nautilus+emcee, --array=0-11%4). The RAL checkout is now at c3ca4de, which contains the #34 fix, and the attempt-1 arms are archived. (5) Commit the clean-slate rows through an autolens_inference PR. numba_cpu dense stays dropped (decisions/hst-cleanslate-drop-numba-dense.md). |
| Point-source sampler admission | 2 October project record reports five Nautilus seeds; no gradient-sampler admission measurement. | Assess a bounded gradient-sampler admission proposal with the human; no new job is authorized. |
| Graphical and EP inference coordination | Mind campaign map and two scoping contracts preserved verbatim; implementation work remains in Mind. | Reconcile current Cortex observations and shipped fixes, then select one bounded phase. |
| Analytic Gaussian ensemble and control calibration | Cortex records accepted wave-1 baseline and 30 September independent-review caveats. | Use Cortex planned diagnostic on passing, edge and worst seeds; retain the human conclusion there. |
| N=50 NUTS versus three EP repeats | Cortex records failed wave 1 (EMFILE/OOM), wave-2 preparation and no submission. | Present the existing wave-2 proposal with current dependency evidence for decision. |
| Hierarchical slope recovery at scale | Cortex retains graphical baseline, post-processing OOM and stale EP hierarchical factor; Mind records later moments fix. | Verify the moments fix release and project mirror; recover missing summaries from existing samples before a proposed rerun. |
| IC50 EP and graphical scale ladder | Cortex records N=5/10/25 results, failed N=50 EP rung and width-interpretation caveat. | Check implementation close-outs, then propose a seeded ladder with appropriate output and mean-field diagnostics. |
| PyAutoFit search benchmark on gaussian_x3 (autofit_inference) | Wave 2 closed (B5b-2, 2026-10-15, autofit_inference#12) — RAL 409930 pulled, 12,750 rows admitted at the pin df6d218f10cd with 0 missing; NSS 50-step leads both evidence rankings; the catalogue ranks only legs with success rate >= 50 % over >= 20 attempts (post-hoc presentation convention); fit records published as inference-summary@3 (101 record shards). | B5c-2b (draft/research/autofit_inference/search_ext_b5c2b_scored_wave_submit.md): realisations 2-3 and their exact references, the correlated_nd@2 scored-wave manifest (83 cells x 50 seeds x 3 realisations, 12 450 runs) and its probe for 8 probe-first cells, then the submit on RAL --partition=ral from a private pinned copy with the A2 memory requests; correlated_nd@2 is frozen (autofit_inference 44c5e77f1fb0); the family is catalogued apart from gaussian_x3 and ranks nothing across families. |
Active tasks
| Task | Campaign | Status | Priority | Next steps |
|---|---|---|---|---|
| Drop the 1e-3 prior centring for shear / multipole / ell_comps in the benchmark | slam-hst-base | blocked | normal | Verify the release containing Galaxy#634 and Lens#754 before changing priors; retain zero-centre gradient witness. Carried into hst-imaging-mass-samplers phase 1 as a prior-identity question for the frozen model. |
| EP campaign — phase map for the 2026 Q3 graphical/EP push | graphical-ep | needs-decision | high | Reconcile current Cortex records and human gates; this campaign map is never issued or bulk scheduled. |
| Expectation Propagation Scale-Up — Scoping | graphical-ep | needs-slicing | high | Reconcile completed wrapper work and Pulse profiling ownership before selecting a bounded remaining inference phase. |
| Graphical Model Scale-Up — Scoping | graphical-ep | needs-slicing | high | Reconcile current Cortex scale results and library dependencies; select one bounded inference phase with a witness. |
| NUTS/HMC trial on validated likelihoods | point-source-admission | needs-decision | normal | An unscoped idea, not permission to submit: reconcile point-source admission with imaging and weak-lensing candidate targets. |
| gaussian_x3 search benchmark — wave 1 (the pilot) | gaussian-x3-search-benchmark | blocked | normal | Pilot stopped at 223 of 520 runs and is not restarted; freeze gaussian_x3@2 from its 17-row calibration before any scored run; the deferred cells move to scored wave 2 (after A2 and A4). |
| gaussian_x3 search benchmark — wave 2 (scored) | gaussian-x3-search-benchmark | running | high | Wave 2 is closed and judged (B5b-2); the fit snapshot is captured from main under inference-summary@3 after the B5b-2 PRs merge. |
| HST imaging mass stage, clean slate: Nautilus base on A100 and jax_cpu at release 2026.10.17.1, linear and fixed lens light | hst-imaging-mass-samplers | needs-decision | high | A100 legs complete; jax_gpu sparse fixed-light mass_total[1] diverged (Cortex observation, human reading pending). 412539 (jax_cpu dense) COMPLETED 2026-10-18, elapsed 1-05:59:53 / 1-02:57:03 (sacct 2026-10-19 ~17:25Z, Cortex PR #73 e3ab356); its rows were not re-read in the 2026-10-19 record. 412541 (jax_cpu sparse) and 412565 (numba_cpu sparse) were CANCELLED 2026-10-19 ~15:51Z on the human's decision. Completed at cancel: jax_cpu sparse seed 0 slam_base; jax_cpu sparse seed 1 source_lp[1] and source_pix[1]; numba sparse slam_base on both seeds. Every lightfixed mass_total[1] is incomplete. The partial stage dirs stay under output/autolens/imaging/hst/slam/<variant>/<config>/seed_<n>/. A rerun of the remainder under the flat layout needs human authorisation. numba_cpu dense dropped (2026-10-18 decision). |
| HST imaging mass stage: sampler comparison, wave 1 | hst-imaging-mass-samplers | needs-decision | normal | Fix merged as autolens_inference#34; the RAL checkout was fast-forwarded to c3ca4de (includes #34 and the #36 flat layout) on 2026-10-19, so the pull blocker is cleared. Attempt 2 (nautilus+emcee, --array=0-11%4, NUTS/SMC deferred) is NOT submitted and needs human authorisation. It reads output/autolens/prepared/<config>/seed_0/<variant>/manifest.json. Attempt-1 material (output/experiments 341M, output/experiment_sources 76M, the 8 failed NUTS/SMC rows from results/searches/experiments) was moved aside 2026-10-19 on the human's decision to /mnt/ral/jnightin/inference/aside/autolens_inference_cleanslate_wave1_attempt1/ (RAL) and /mnt/c/Users/Jammy/Science/inference/aside/autolens_inference_cleanslate_wave1_attempt1/ (laptop), so attempt 2 starts clean. |
Inference Results
Decision History
Sampler candidates
Literature expectations, not measured PyAuto recommendations. Review a candidate and copy its investigation prompt.
dynesty
dynamic nested sampling · reviewed 2026-10-08
Expected uses
- Evidence and posterior baseline with adjustable effort allocation.
Limitations
- Prior transforms and likelihood-contour exploration need validation; no universal dimensional scaling advantage.
Integration: integrated: DynestyStatic and DynestyDynamic
Benchmark evidence: not_assessed: this literature review establishes no benchmark result
Investigation prompt
Nautilus
neural-network importance nested sampling · reviewed 2026-10-08
Expected uses
- Evidence and posterior estimation for expensive gradient-free likelihoods.
Limitations
- Learned proposal coverage and importance-weight stability need checking on unseen modes.
Integration: integrated: Nautilus
Benchmark evidence: not_assessed: this literature review establishes no benchmark result