prompt: nfw-truncated-potential-accuracy shipped via PyAutoGalaxy#564 → complete

draft/bug/autogalaxy/nfw_truncated_potential_accuracy.md → complete/2026/08/nfw-truncated-potential-accuracy.md

@@ -1,3 +1,69 @@
+`NFWTruncatedSph.potential_2d_from` failed the `grad(psi)=alpha` finite-difference
+self-consistency check at med 7.1e-2 (vs ~8e-4 for every other NFW/gNFW/cNFW
+variant) — the MGE decomposition's sigma range (`radii_max = truncation_radius * 5`)
+was the prime suspect.
+
+**Shipped:** [PyAutoGalaxy#564](https://github.com/PyAutoLabs/PyAutoGalaxy/pull/564),
+merged 2026-08-09 (fixes PyAutoGalaxy#563). This ledger entry was reconciled
+retroactively on 2026-08-19 — the fix shipped outside the prompt's dev flow, so the
+draft prompt and `planned.md` entry had gone stale.
+
+## What shipped (and how it differed from the plan)
+
+The prompt's plan was to **widen/refine the MGE sigma range**. The shipped fix went
+further: the MGE approximation was **replaced entirely** by the exact analytic
+lensing potential — equation (18) of Baltz, Marshall & Oguri (2009) for the n=1
+smoothly truncated NFW — which the prompt's step 3 had flagged as a cross-check.
+The analytic form became the implementation:
+
+- `potential_func_sph_from(grid_radius, tau, xp)` — dimensionless analytic BMO
+  potential, with a sixth-order small-radius series below `x = 1e-1` to avoid
+  catastrophic cancellation in NumPy/JAX float32, zero-centred per the paper's
+  convention.
+- `NFWTruncatedSph.potential_2d_from` applies the `2 kappa_s r_s^2` normalisation.
+
+## Validation
+
+From the PR: 48 profile tests + 1054 full `test_autogalaxy/` pass; full-mode
+200×200 workspace checks across three tNFW truncation regimes 9 PASS / 0 FAIL;
+independent comparison against lenstronomy TNFW; JAX JIT/autodiff/float32, centre,
+`x=1` and large-truncation checks. The unit tests the prompt asked for exist in
+`test_autogalaxy/profiles/mass/dark/test_nfw_truncated.py`:
+`test__potential_2d_from__gradient_matches_analytic_deflections` (rel 1e-6),
+`__laplacian_matches_analytic_convergence` (rel 1e-5), an independent-implementation
+match, and an NFW-limit comparison at large truncation radius.
+
+Independent re-verification on `main` (49115ad, 2026-08-19), finite-differencing
+`grad(psi)` against `coord_func_m`-based deflections over r in [1e-3, 20] arcsec:
+
+| tau | median frac err | max |
+|-----|-----------------|-----|
+| 2.0 (defaults) | 1.4e-09 | 4.3e-07 |
+| 10.0 | 2.2e-09 | 4.8e-07 |
+| 0.5 | 1.2e-09 | 4.6e-07 |
+
+The med 7.1e-2 failure is gone — errors are now at the FD-scheme floor, ~5 orders
+below the other profiles' MGE-limited ~8e-4.
+
+## Traps / lessons
+
+- **Analytic beat MGE re-tuning.** The prompt's "widen the sigma range" plan would
+  have kept an approximation where a closed form exists. When a profile has a
+  published analytic potential (BMO 2009 here), port it — the MGE path is for
+  profiles without one.
+- **Ledger drift:** the fix merged 2026-08-09 but the prompt sat in `draft/` and
+  `planned.md` for ten more days, still surfacing on the dashboard as open
+  high-priority work. A fix shipped outside the prompt's own dev flow must
+  reconcile the Mind in the same breath.
+- Related but distinct: the Isothermal Ell/Sph potential disagreement
+  (`planned.md` § `isothermal-ell-sph-oversampling-at-the-cusp`) was once guessed
+  to share this root cause — retracted; MGE is not involved there.
+- `draft/feature/autogalaxy/piemass_potential.md` (PIEMass has *no* potential) is
+  the sibling task and remains open; the BMO-style analytic-port route used here
+  is the template for it.
+
+## Original prompt
+
 # `NFWTruncatedSph.potential_2d_from`: MGE potential fails `grad(psi)=alpha` self-consistency
 
 Type: bug

complete/index.md

@@ -6,7 +6,7 @@ Token-light navigation over the finished-work records (schema:
 only then grep a dated bucket. Curators: edit the band between the CURATED
 markers; everything below GENERATED is rebuilt.
 
-1026 records across 7 buckets.
+1027 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -91,6 +91,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [multistart-gradient-resume-fom-sanity-check](2026/08/multistart-gradient-resume-fom-sanity-check.md)
 - [multistart-nan-step-diagnostics](2026/08/multistart-nan-step-diagnostics.md)
 - [nautilus-1core-serial-pool](2026/08/nautilus-1core-serial-pool.md) — corrective for the Heart RED "release validation FAILED (stage
+- [nfw-truncated-potential-accuracy](2026/08/nfw-truncated-potential-accuracy.md)
 - [normalise-auto-simulate-guard-idiom](2026/08/normalise-auto-simulate-guard-idiom.md)
 - [notebook-setup-notebook-regen-drift](2026/08/notebook-setup-notebook-regen-drift.md)
 - [nss-remainline-mainline-blackjax](2026/08/nss-remainline-mainline-blackjax.md)

dashboard.md

@@ -1,6 +1,6 @@
 # PyAutoMind task dashboard
 
-<!-- generated by `pyauto-brain intake dashboard --apply` on 2026-08-18 — regenerate, do not hand-edit -->
+<!-- generated by `pyauto-brain intake dashboard --apply` on 2026-08-19 — regenerate, do not hand-edit -->
 
 Every task the Mind is holding, on one page: what is in flight, what is parked, and the whole backlog to pick from. Pick a line, then run `/start_dev <prompt-path>` to start it.
 
@@ -10,14 +10,14 @@ Tasks only — the organism's health lives with the Heart (`/health`), not here.
 |-------|------:|
 | [In flight](#in-flight) (`active/`) | 5 |
 | [Parked](#parked) (`parked.md`) | 3 |
-| [Planned](#planned) (`planned.md`) | 7 |
-| [Backlog](#backlog) (`draft/`) | 124 |
+| [Planned](#planned) (`planned.md`) | 6 |
+| [Backlog](#backlog) (`draft/`) | 123 |
 
 Live on GitHub: [open issues](https://github.com/search?q=org%3APyAutoLabs+is%3Aissue+is%3Aopen&type=issues) · [open pull requests](https://github.com/search?q=org%3APyAutoLabs+is%3Apr+is%3Aopen&type=prs)
 
 ## Start here
 
-**Highest priority** (filed as `high`) — showing 12 of 24
+**Highest priority** (filed as `high`) — showing 12 of 23
 
 - [TRIAGE: needs manual review before routing](draft/triage/jax_zero_contour.md) — medium · safe · high
 - [Decide whether the clipper belongs in the search identifier](draft/feature/autofit/clipper_in_search_identifier.md) — autofit · medium · human-required · high
@@ -29,8 +29,8 @@ Live on GitHub: [open issues](https://github.com/search?q=org%3APyAutoLabs+is%3A
 - [Deep research: Can we speed up Delaunay in PyAutoArray?](draft/research/autoarray/delaunay_research.md) — autoarray · too-large · supervised · high
 - [EP hierarchical parent-scale collapse: cure the basin, or document the](draft/bug/autofit/ep_scale_collapse_basin_cure_or_caveat.md) — autofit · too-large · human-required · high
 - [Census of priors and messages — confirmed bugs + redesign](draft/research/autofit/priors_and_messages_math_audit.md) — autofit · too-large · supervised · high
-- [`NFWTruncatedSph.potential_2d_from`: MGE potential fails `grad(psi)=alpha` self-consistency](draft/bug/autogalaxy/nfw_truncated_potential_accuracy.md) — autogalaxy · too-large · supervised · high
 - [`einstein_radius_jit_from`: replace static init_guess with a JAX-native seed finder](draft/refactor/autogalaxy/einstein_radius_jit_native_seed_finder.md) — autogalaxy · too-large · supervised · high
+- [Investigate eager `FitImaging.figure_of_merit` vs JIT/step-by-step divergence in rectangular pixelization](draft/bug/autolens/pixelization_eager_vs_jit_divergence.md) — autolens · too-large · supervised · high
 
 **Quick wins** (small enough, and safe enough to run unattended)
 
@@ -69,27 +69,25 @@ Started or scoped, not currently in flight — resume by moving the row back to
 Scoped but not started; some are not yet prompt files. Full detail in [`planned.md`](planned.md).
 
 <details>
-<summary><b>7</b> task(s)</summary>
+<summary><b>6</b> task(s)</summary>
 
 - **isothermal-ell-sph-oversampling-at-the-cusp** — planned — NOT yet a prompt file; file one via `/intake` before starting
 - **remote-mcp-deployment-tiers** — [issue #20](https://github.com/PyAutoLabs/autofit_assistant/issues/20) — DESIGN-COMPLETE, build BLOCKED-ON-DEMAND — issue #20 holds the full auth/transport/hosting design + Richard/PyAutoMCP…
 - **samples-parameter-paths** — [issue #1327](https://github.com/PyAutoLabs/PyAutoFit/issues/1327) — parked
 - **jax-point-source-point-smoke-sentinel** — planned
-- **nfw-truncated-potential-accuracy** — planned
 - **piemass-potential** — planned
 - **latent-nan-guard-honest-run**
 
 </details>
 
 ## Backlog
 
-**124** filed prompts, not started. Each section is sorted most-pickable first (priority, then size).
+**123** filed prompts, not started. Each section is sorted most-pickable first (priority, then size).
 
 <details>
-<summary><b>bug</b> — 32</summary>
+<summary><b>bug</b> — 31</summary>
 
 - [EP hierarchical parent-scale collapse: cure the basin, or document the](draft/bug/autofit/ep_scale_collapse_basin_cure_or_caveat.md) — autofit · too-large · human-required · high
-- [`NFWTruncatedSph.potential_2d_from`: MGE potential fails `grad(psi)=alpha` self-consistency](draft/bug/autogalaxy/nfw_truncated_potential_accuracy.md) — autogalaxy · too-large · supervised · high
 - [Investigate eager `FitImaging.figure_of_merit` vs JIT/step-by-step divergence in rectangular pixelization](draft/bug/autolens/pixelization_eager_vs_jit_divergence.md) — autolens · too-large · supervised · high
 - [Fix Autofit release sampler and database regressions](draft/bug/health_fixes/autofit_sampler_database.md) — health_fixes · too-large · supervised · high
 - [Fix release JAX runtime compatibility and likelihood parity](draft/bug/health_fixes/jax_runtime_and_parity.md) — health_fixes · too-large · supervised · high

planned.md

@@ -68,23 +68,6 @@
     Sibling smoke scripts to check while triaging: image_plane.py,
     source_plane.py in the same dir — they share the seed dataset.
 
-## nfw-truncated-potential-accuracy
-- prompt: draft/bug/autogalaxy/nfw_truncated_potential_accuracy.md
-- status: planned
-- filed: 2026-06-05
-- classification: library (accuracy bug)
-- suggested-branch: feature/nfw-truncated-potential-accuracy
-- summary: |
-    Pre-existing accuracy bug surfaced while shipping dark-matter-potentials.
-    NFWTruncatedSph.potential_2d_from (MGE) fails grad(psi)=alpha self-
-    consistency in autolens_workspace_test/scripts/mass/dark.py (med 7.1e-2 vs
-    ~8e-4 for every other NFW/gNFW/cNFW variant). Deflections pass, only the
-    potential is off — likely the MGE sigma range (radii_max = truncation_radius
-    * 5) is too narrow. Reproduce on clean main first.
-- affected-repos:
-  - PyAutoGalaxy
-
-
 ## piemass-potential
 - prompt: draft/feature/autogalaxy/piemass_potential.md
 - status: planned