prompt: record clipper-in-search-identifier as shipped (PyAutoFit#1493/#1494)

complete/2026/08/clipper-in-search-identifier.md

@@ -0,0 +1,199 @@
+# clipper-in-search-identifier (decided: option 2 — the clipper enters the MLE identifiers)
+
+- library-prs: https://github.com/PyAutoLabs/PyAutoFit/pull/1494
+- merge-commits: PyAutoFit `f3767a79743e6976accd8fd6957afc5f366654fb` (2026-08-18)
+- issue: PyAutoFit#1493 (closed by #1494)
+- summary: The prior-support `Clipper` did not enter the search identifier, so
+  two runs differing only in a result-affecting setting shared one output
+  directory and — stacked with the `.completed` short-circuit — a later run
+  could silently return the earlier one's numbers. **Decided option 2:**
+  `__identifier_fields__ = ("clipper",)` on `AbstractMultiStartGradient` and
+  `AbstractBFGS`, accepting that every existing multi-start and (L)BFGS output
+  directory re-keys. Nested samplers, MCMC searches and `Drawer` are deliberately
+  untouched and pinned byte-identical.
+- validation: 104 lines of new identifier tests (5 new cases) — hash-lists pinned
+  for `Nautilus`/`DynestyDynamic`/`Emcee`/`Zeus`/`Drawer`, a fork test over
+  `MultiStartAdam` and `LBFGS`, and a tripwire asserting the nested samplers
+  never grow a `clipper` attribute.
+- release: shipped in PyAutoFit **2026.8.20.1** (2026-08-20).
+
+## The decision, and why — this was chosen, not overlooked
+
+The prompt put four options up with none pre-selected and marked itself
+`Autonomy: human-required`, because either answer costs someone's stored
+results. The human chose **option 2: put it in, unconditionally, scoped to the
+searches that consume it.**
+
+Why not the others:
+
+- **Option 1 (status quo)** leaves a footgun that had already been stepped on
+  once — the phase-2 validation campaign's arms 1 and 2 differ in *nothing else*
+  and collided, and were only saved by per-arm unique names.
+- **Option 3 (include only when not `ClipperNone`)** buys back-compatibility
+  with a *conditional* identifier — a new concept in that machinery, and one
+  every consumer (database, aggregator, `.completed` discovery) would have to be
+  re-checked against. Rejected as a novel special case guarding a one-off.
+- **Option 4 (make the `.completed` short-circuit loud instead)** closes the
+  wider class of cached-result hazards and remains worth doing, but it does not
+  make the identifier tell the truth about what produced a result. It was not
+  taken here and is **not** carried forward by this record — file it fresh if
+  wanted.
+
+The cost was paid with eyes open: **stored multi-start and (L)BFGS results are
+orphaned, not deleted.** Nothing finds them by identifier any more. The
+re-key hits *every* such directory, including runs that never passed a clipper,
+because the pre-change hash list for these searches was the class name alone
+(`AbstractSearch.__identifier_fields__` is the empty tuple, and
+`Identifier._add_value_to_hash_list` then contributes no fields) — adding one
+field changes the hash for the default construction too.
+
+## The scoping constraint is the load-bearing part
+
+The clipper is resolved on `AbstractMLE`, and the fix declares the field on the
+two MLE subclasses that *consume* it — not on `NonLinearSearch`. Hoisting it
+would put the clipper within reach of the nested samplers' identifier machinery
+and silently re-key the entire nested-sampling archive, which is a far larger
+body of stored results and one where the setting cannot affect the answer at
+all. `test_nested_samplers_have_no_clipper` exists to fail loudly if anyone
+tries.
+
+`Drawer` is the same rule in miniature: it inherits the attribute from
+`AbstractMLE` but never consumes it, so it deliberately does **not** declare the
+field. **A setting that cannot change the result must not re-key stored
+results** — that is the general principle this task settled, and it is worth
+more than the specific answer.
+
+## Sequencing paid off
+
+The prompt argued this had to be decided **before** phase 3 (flipping the
+default to `ClipperPriorBox`), so the re-baseline and the re-keying would not be
+entangled. That held: the identifier change landed 2026-08-18 on its own, and
+phase 3 now has one variable rather than two.
+
+## Residuals — what this record does NOT claim is done
+
+1. **The release notes do not say plainly that stored results re-key.**
+   PyAutoFit 2026.8.20.1's notes carry the commit subject
+   (`feat: clipper enters the MLE search identifiers (#1494)`) and nothing more.
+   The prompt asked explicitly for a plain statement that existing multi-start
+   and (L)BFGS output directories are orphaned. Amend that release's notes, or
+   carry the sentence into the next release's.
+2. **"Check whether a migration is wanted" was never answered.** No migration
+   ships, and no decision that one is unwanted is on record. Users with stored
+   multi-start/(L)BFGS results have no path to re-key them.
+3. **The phase-2 campaign record's ARM-COLLISION TRAP section is now stale.**
+   `complete/2026/08/clipper-validation-campaign.md` states as fact that the
+   clipper does not enter the identifier, with three empirically-verified hashes.
+   True on 2026-08-16, false since 2026-08-18. A pointer note was added there in
+   this same change; the historical measurement is left intact.
+4. **Option 4 remains open** as the general fix for the `.completed`
+   short-circuit returning cached results when the search config differs.
+
+## Mind-side drift this closed
+
+The work shipped in PyAutoFit on 2026-08-18 while the prompt sat in
+`draft/feature/autofit/clipper_in_search_identifier.md` — never issued into
+`active/`, never registered, and never recorded. Four other records and drafts
+still pointed at that draft path as an *open* question. This record is the
+retrospective close; the code has been on `main` since #1494 merged.
+
+The transferable lesson: a task that ships straight from a decision — no issue,
+no branch registration — leaves no trace in the Mind unless the record is
+written by hand afterwards. `lifecycle.py check` does not catch it, because it
+only guards `active.md` against `complete/`; a prompt that never entered
+`active/` is invisible to the drift guard.
+
+## Original prompt
+
+# Decide whether the clipper belongs in the search identifier
+
+Type: feature
+Target: autofit
+Repos:
+- PyAutoFit
+Difficulty: medium
+Autonomy: human-required
+Priority: high
+Status: formalised
+
+Filed 2026-08-16. Follow-up 4 owed by the prior-support `Clipper`
+(`complete/2026/08/prior-support-clipper.md`, PyAutoFit#1477).
+
+**`Autonomy: human-required` on purpose.** Either answer has a cost that lands
+on someone's stored results, and the trade is a judgement about the project's
+data, not a technical detail an agent should settle.
+
+## The fact
+
+The clipper does **not** enter the search identifier. Verified against
+PyAutoFit `main` on 2026-08-16:
+
+```
+af.MultiStartAdam(name="x")                             -> 2bada4747f74bc46bf812605a762def9
+af.MultiStartAdam(name="x", clipper=ClipperNone())      -> 2bada4747f74bc46bf812605a762def9
+af.MultiStartAdam(name="x", clipper=ClipperPriorBox())  -> 2bada4747f74bc46bf812605a762def9
+```
+
+Two runs differing only in prior-support enforcement — which can change the
+answer, and demonstrably does — share one output directory.
+
+## Both answers cost something
+
+**Leave it out (status quo).** Existing on-disk results are never orphaned, and
+phase 1 stays back-compatible by construction. But runs that differ in a
+result-affecting setting collide, and stacked with the `.completed`
+short-circuit a later run silently returns the earlier one's numbers. That is
+live right now for the phase-2 campaign, whose arms 1 and 2 differ in *nothing
+else* — see
+`draft/feature/autofit/clipper_validation_campaign.md`, which works around it
+with unique per-arm names.
+
+**Put it in.** Arms separate naturally and the identifier tells the truth about
+what produced a result. But **every existing multi-start and BFGS output
+directory re-keys**, so stored results are orphaned — they are not deleted, but
+nothing finds them any more. That is the same comparability argument
+PyAutoFit#1472 made when deferring its own policy change, and it collides with
+phase 3, which wants a benchmark re-baseline it can compare against history.
+
+## Options, none pre-selected
+
+1. **Status quo**, with the collision documented wherever it bites. Cheapest;
+   leaves a footgun that has already been stepped on once.
+2. **Include the clipper in `__identifier_fields__`.** Honest; orphans stored
+   results.
+3. **Include it only when it is not `ClipperNone`.** Unclipped runs keep their
+   existing directories, clipped runs get their own. Back-compatible *and*
+   collision-free — but the identifier becomes conditional, which is a new
+   concept in that machinery and needs checking against how identifiers are
+   consumed (database, aggregator, `.completed` discovery).
+4. **Leave the identifier alone and make the collision loud** — refuse to
+   short-circuit on `.completed` when the search config differs from the one
+   recorded in the completed run's `search.json`. Fixes the general
+   cached-result hazard rather than this one instance.
+
+Option 3 is the obvious-looking compromise and option 4 is the one that closes
+the whole class; both need someone to confirm the consumers can take it.
+
+## Sequencing
+
+**Best decided before phase 3, not after.** Phase 3 flips the default to
+`ClipperPriorBox` and carries a re-baseline; if the identifier changes at the
+same time, the re-baseline and the re-keying are entangled and neither can be
+read cleanly. Deciding this first — either way — leaves phase 3 with one
+variable.
+
+## Whatever is chosen
+
+- State the decision and its reason in the record. A future reader hitting a
+  collision needs to know it was chosen, not overlooked.
+- If the identifier changes, say plainly in the release notes that stored
+  results re-key, and check whether a migration is wanted.
+- If it does not change, the phase-2 mitigation (unique `name` per arm) becomes
+  permanent advice for anyone comparing clipper settings, and belongs in the
+  `Clipper` docstring rather than only in a campaign prompt.
+
+## Out of scope
+
+- Flipping the default (phase 3).
+- The `.completed` short-circuit's other manifestations, unless option 4 is
+  chosen — in which case they are the point.

complete/2026/08/clipper-validation-campaign.md

@@ -143,6 +143,14 @@ rather than live hazards — but do **not** delete the mitigations:
 
 ### ⚠ THE ARM-COLLISION TRAP — read before running anything
 
+> **SUPERSEDED 2026-08-18 — this trap no longer exists.** The clipper now
+> enters the identifier for the multi-start and (L)BFGS searches
+> (PyAutoFit#1493 / #1494, `f3767a7`), so arms differing only in clipper
+> separate by themselves. The hashes and the reasoning below are left
+> intact as the measurement that motivated the change — they were true on
+> 2026-08-16 and are false from 2026-08-18. Record:
+> `complete/2026/08/clipper-in-search-identifier.md`.
+
 **The clipper does NOT enter the search identifier.** Verified empirically
 against PyAutoFit `main` on 2026-08-16:
 
@@ -171,9 +179,10 @@ Mitigation — do all three:
 
 Not a bug in itself — it is what keeps existing runs' directories stable — but
 it is precisely wrong for this campaign. Whether the clipper *should* enter the
-identifier is now filed as
-`draft/feature/autofit/clipper_in_search_identifier.md`, and phase 3's
-re-baseline is the reason it matters.
+identifier was filed as a follow-up and **decided on 2026-08-18: it does**
+(PyAutoFit#1493 / #1494; record
+`complete/2026/08/clipper-in-search-identifier.md`). Phase 3's
+re-baseline was the reason it mattered.
 
 ### Harness traps that already cost a session
 

complete/2026/08/mge-lane-death.md

@@ -102,8 +102,10 @@ Downstream, all 2026-08-16:
   `draft/research/autolens_profiling/ell_comps_trapping_unmasked.md`. It runs on top
   of phase 1 and can share phase 2's arms.
 - **Clipper phase 3**, which was never drafted. If it is ever written it must argue
-  hygiene, not accuracy; `draft/feature/autofit/clipper_in_search_identifier.md` is
-  its prerequisite decision.
+  hygiene, not accuracy. Its prerequisite decision is now **made** — the clipper
+  entered the MLE identifiers on 2026-08-18 (record
+  `complete/2026/08/clipper-in-search-identifier.md`), so phase 3's
+  re-baseline carries one variable rather than two.
 
 ## Unexplained, left open
 

complete/2026/08/prior-support-clipper.md

@@ -194,9 +194,11 @@ So **follow-ups 1 and 2 in the list below are now FIXED**; they are left in plac
 because the reasoning that found them is the record. Their own records are
 `complete/2026/08/save-json-numpy-scalar-typeerror.md` and
 `complete/2026/08/crashed-run-poisons-resume.md`. Follow-ups 3, 4 and 5 remain
-open and 3 and 4 are now filed as prompts:
-`draft/bug/autofit/loggaussian_prior_declares_own_support.md` and
-`draft/feature/autofit/clipper_in_search_identifier.md`.
+open and 3 and 4 were filed as prompts:
+`draft/bug/autofit/loggaussian_prior_declares_own_support.md` and the
+identifier question — the latter **shipped 2026-08-18** (PyAutoFit#1493 /
+#1494, the clipper now enters the MLE identifiers; record
+`complete/2026/08/clipper-in-search-identifier.md`).
 
 Two corrections that came out of doing that follow-up work:
 

complete/2026/08/uniform-prior-bounds-numpy-path.md

@@ -36,8 +36,9 @@
   overrode the worktree-conflict guard (PyAutoFit was claimed by stored-sample-reconstruction-guard
   and version-stamp-sync-guards; change surface disjoint from both).
 - follow-ups (filed/known at completion):
-  1. Shape C deliberately deferred: default `LBFGS` to `ClipperPriorBox` (draft
-     `draft/feature/autofit/clipper_in_search_identifier.md` is adjacent). With strict priors and
+  1. Shape C deliberately deferred: default `LBFGS` to `ClipperPriorBox` (the
+     adjacent identifier question was decided 2026-08-18 — record
+     `complete/2026/08/clipper-in-search-identifier.md`). With strict priors and
      `ClipperNone`, an LBFGS step out of the box now sees an infinite objective — C gives scipy
      declarative bounds instead.
   2. Workspace smoke literal risk: `*_workspace_test` scripts that pin literal values from

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.
 
-1050 records across 7 buckets.
+1051 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -38,6 +38,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [circular-ell-comps-image-gradient](2026/08/circular-ell-comps-image-gradient.md) — Replaced the Sersic image path's Cartesian-to-polar ellipticity conversion with an algebraically equivalent Ca…
 - [cli-noise-autonerves-batch](2026/08/cli-noise-autonerves-batch.md)
 - [cli-noise-pyautofit-batch](2026/08/cli-noise-pyautofit-batch.md)
+- [clipper-in-search-identifier](2026/08/clipper-in-search-identifier.md) — decided: option 2 — the clipper enters the MLE identifiers
 - [clipper-usage-in-search-summary](2026/08/clipper-usage-in-search-summary.md) — `search.summary` now reports prior-support clipping. `n_clipped_lane_steps`
 - [clipper-validation-campaign](2026/08/clipper-validation-campaign.md)
 - [compile-axis-campaign-coverage](2026/08/compile-axis-campaign-coverage.md)