chore(external-validity): retire design-era working material from the study tree

.gitignore

@@ -114,6 +114,27 @@ docs/research/provision-matching/probes/merged_labels.json
 # No trailing slash, per the #319 symlink reasoning above.
 docs/research/pdf-backend-bakeoff/holdout
 
+# External-validity probe evidence (A46). Running any xNN probe rewrites its own evidence
+# file, so after the A46 cleanup these reappear as untracked artifacts and a later `git add -A`
+# would silently re-commit the working material that was retired. Ignored rather than left
+# loose. They are regenerated by running the probe; git history holds the removed versions.
+#
+# The negations are the artifacts something actually READS or that a byte-frozen document
+# cites — dropping one un-tracks a live gate input, so only change this list together with
+# the consumer that justifies it. A new committed artifact needs a new negation here.
+docs/research/pdf-backend-bakeoff/validation/external-validity/results/x*.json
+# G2 reads this one; G6 defects ORACLE_INTEGRATION_NOT_VERIFIED without the other.
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x2_contract_assertions.json
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x26_control_oracle.json
+# Cited as MEASURED by PRE-REGISTRATION.md, which is byte-frozen and can never be repointed.
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x00_design_pilot.json
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x02_oracle_reference_defects.json
+# Named by retained METHODOLOGY_SURFACE files (cross_engine_control, run_hybrid, run_extended,
+# reconstruct_extended_corrected), which may not be edited for a cosmetic reason.
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x09_skeleton_cross_engine.json
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x11_provenance_chain.json
+!docs/research/pdf-backend-bakeoff/validation/external-validity/results/x13_x_arm.json
+
 # Build artifacts. The project became buildable in #398, so `uv build` now produces
 # these where it never used to. `build/` matters beyond tidiness: some backends copy
 # the package's own .py files into it, and a directory holding Python that is neither

AGENTS.md

@@ -89,13 +89,16 @@ A comment describes the code as it is **now**. History and rejected alternatives
 
 Probes, audits, spikes and study write-ups are **working material, not automatically permanent repository material.** At closure, retain an artifact only if it is needed to reproduce a consequential result, enforce an invariant, document a durable decision, or serve as a frozen input. **Delete the rest** — Git history preserves the investigative record. Retention is the exception that needs a reason, and "it was expensive to produce" is not one.
 
+Apply that strictly: "useful history", "might be interesting later", "shows how we got here" and "was once reviewed" are not retention reasons, because every one of them is satisfied by `git log`. Judge an artifact by the consequential role it plays **now**, never by its filename, its age, or how much history it carries. The inverse error is just as real, so decide by tracing consumers — under `docs/research/` a generated JSON that looks like a stale run log may be a committed expected-output oracle a gate reads, and a probe that looks superseded may be the only executable negative control of a rule still in force. Before removing one, check imports, direct path reads, subprocess calls, gate inputs, and any doc that names it as a *current* requirement.
+
 When deleting research, **remove or update live references to it**, and move any durable conclusion into its authoritative current home: an ADR, the architecture documentation, an executable test, or a frozen fixture. Do not write an archival summary, a tombstone map or a closure document whose only job is to record what was deleted — that is a new artifact with the same problem, and a pointer into history is one more thing that can be wrong.
 
-The failure this prevents is subtler than rot. A probe that no longer runs at least announces itself; a probe that still runs can publish a quantity the project has since disowned, and passing a check makes it look current. So the question at closure is not "does it work?" but "does it still answer a live question?"
+The failure this prevents is subtler than rot. A probe that no longer runs at least announces itself; a probe that still runs can publish a quantity the project has since disowned, and passing a check makes it look current. So the question at closure is not "does it work?" but "does it still answer a live question?" `tests/test_research_probes.py` runs the probes it declares runnable against a closed manifest, so a probe added later is either run or the gate fails.
 
-So the gate executes (which catches the first two), and retirement is judged on whether the artifact still answers a live question (which catches the third). `tests/test_research_probes.py` runs the probes it declares runnable against a closed manifest, so a probe added later is either run or the gate fails.
+Two rules that decide the ambiguous cases:
 
-When retiring an artifact, record in the document that cited it **which executable test inherited its question**, and pin the commit it stays readable at. A reproduction claim pointing at a deleted file is the defect this is meant to avoid, not the price of avoiding it.
+- **Prove claimed regenerability rather than assuming it.** If an artifact is being deleted because surviving machinery can reproduce it, re-run the producer and confirm that claim. Failure to regenerate is a reason to re-check the four criteria above; it is not, by itself, a retention reason — a one-off diagnostic that nothing needs is just as disposable when its producer is gone.
+- **A dangling reference is acceptable only where the referring document can be appended to.** If a byte-frozen document (a pre-registration, a sealed manifest) cites the artifact, it can never be given a pointer to git history — so retain the artifact instead. An append-only ledger can simply record where the removed file resolves, which is the one place a history pointer earns its keep.
 
 ## Architecture decisions
 

docs/research/pdf-backend-bakeoff/validation/external-validity/HARNESS-PLAN.md

@@ -45,8 +45,8 @@ never re-derives an earlier stage's decisions.
 
 ## 2. `build_frames.py` — **BUILT (A31)**
 
-Implemented and tested on SYNTHETIC + DEVELOPMENT material only; evidence in
-`results/x17_build_frames.json`, **56/56 controls passing**. No canonical `results/frames.json`
+Implemented and tested on SYNTHETIC + DEVELOPMENT material only; evidence is regenerated by
+running `x17_build_frames.py`, **56/56 controls passing**. No canonical `results/frames.json`
 exists. The D-frame is emitted as a **complete census** — the A10/A27.3 budget is not applied
 here. Anchor equality uses the **whole emitted production `Anchor` value**; no reduced
 signature was invented. No runner instrumentation was added: each arm's anchors come from the
@@ -372,7 +372,7 @@ a pre-committed sentence, and must refuse to emit any comparative-accuracy claim
 
 ### RESOLVED by measurement — 7.4 anchor→neutral bridge
 
-**Proven, not argued** (`x14_anchor_bridge.py`, `results/x14_anchor_bridge.json`):
+**Proven, not argued** (run `x14_anchor_bridge.py`):
 
 ```
 Anchor(page_number, line_number)   <- line_number is the GPO PRINTED margin number
@@ -426,7 +426,7 @@ enforce it.
 | §4.5 P-head was a caller obligation | **A30.4** — `filter_keys` now takes `(key, kind, population)` and applies both restrictions itself |
 | blind-ID uniqueness was only synthetic | **A30.5** — asserted over the **realized** set; a collision aborts the build |
 
-**Proven, not argued** (`x16_occurrence_identity.py`, `results/x16_occurrence_identity.json`):
+**Proven, not argued** (run `x16_occurrence_identity.py`):
 
 ```
 instrumented mirror == production extract_anchors    every DEVELOPMENT page consumed

docs/research/pdf-backend-bakeoff/validation/external-validity/PRE-EXECUTION-AMENDMENTS.md

@@ -6071,6 +6071,74 @@ opening any document and nothing measured, since the gate reads keys only.
 
 ---
 
+## A46 — SUBSTANTIVE. Design-era working material is retired from the tree
+
+```json
+{"id": "A46", "class": "SUBSTANTIVE",
+ "commits": ["4b808e2"],
+ "confirmatory_output_at_time": "none",
+ "affects_membership": false, "affects_scoring_rule": false,
+ "files_touched": ["probes/x06_m6_feasibility.py", "probes/x19_raster_edge_diagnostic.py",
+                   "HARNESS-PLAN.md"],
+ "supersedes_text_in": "none -- NO frozen rule changes. No metric, threshold, denominator, control fixture, population, adjudication route, decision rule or F/G gate moves. G5's surface stays 15 files and no member of it is touched",
+ "status": "COMPLETE. Boundary ABSENT, execution FORBIDDEN"}
+```
+
+**The retention rule being applied.** Research artifacts are working material, not
+automatically permanent repository material. At closure, an artifact is retained only if it
+is needed to reproduce a consequential result, enforce an invariant, document a durable
+decision, or serve as a frozen input — plus, here, anything the still-active Run 2 execution
+path needs. Git history preserves the investigative record. "Useful history", "shows how we
+got here" and "was once reviewed" are not retention reasons.
+
+**Two probes are removed**, and neither owns a live invariant:
+
+- **`x06_m6_feasibility.py`** measured whether M6 could be adjudicated inside a 6–10 line
+  region. **A20 STRUCK M6**, and HARNESS-PLAN implements "§6 (M0–M9 minus M6)", so the
+  question is no longer in the methodology. It also drives the superseded `pdfium_hybrid` /
+  `reconstruct_hybrid` arms rather than the frozen H/X pair, so it does not exercise any
+  current component. A17.1's table transcribes its finding inline.
+- **`x19_raster_edge_diagnostic.py`** asked whether the visible left edge is displaced from
+  the geometric `x0`. It **states no pass threshold** by construction, and the ruling it
+  informed — A30.3 stands, no tolerance — is frozen in A32. Its 16 development raster PNGs
+  are referenced only by its own evidence file.
+
+**Twenty generated evidence artifacts are removed.** Their only role was recording that an
+earlier probe run passed; nothing reads them. Each is regenerated by a probe that remains in
+the tree, and regenerability was measured rather than assumed: every committed probe artifact
+in this study was re-run before the deletion and rewrote **byte-identically**, leaving the
+tree clean.
+
+**Earlier links in this ledger resolve in git history.** A17/A19/A21/A22/A23/A24.2 and A41
+link `results/x06_m6_feasibility.json`, `x07_neutral_geometry.json`,
+`x08_neutral_identity.json`, `x10_reconstruction_signature.json`,
+`x12_skeleton_eligibility.json` and `x27_score_metrics.json` as MEASURED evidence. Those
+amendments are **not rewritten** — the ledger is an append-only record, and editing a past
+entry to hide a removal would be the defect, not the fix. Read those links at **`4b808e2~1`**.
+Each amendment already transcribes its numbers inline, so the reasoning stands without them.
+
+**What was deliberately NOT removed, and why.** The line is that a dangling reference is
+acceptable only where the referring document can be appended to, as this ledger can:
+
+| retained | why it could not go |
+|---|---|
+| `holdout/`, `control_fixtures/`, `holdout_membership.json`, `contamination.json`, `design_exposure.json`, `control_fixtures.json` | frozen inputs to F1/F2/F3/F7/F8/F10/F11, G4 and G6 |
+| `results/x2_contract_assertions.json` | G2 reads it |
+| `results/x26_control_oracle.json` | G6 defects `ORACLE_INTEGRATION_NOT_VERIFIED` without it |
+| `results/design_runs/*.log` | `x05` **derives** the design-exposure list from them; a primary source, not a log dump |
+| `results/x00_design_pilot.json`, `results/x02_oracle_reference_defects.json` | cited as MEASURED by **`PRE-REGISTRATION.md`**, which is byte-frozen and can never carry a pointer to git history |
+| `results/x09_skeleton_cross_engine.json`, `x11_provenance_chain.json`, `x13_x_arm.json` | named by retained METHODOLOGY_SURFACE files; repointing those would edit a result-bearing file for a cosmetic reason |
+| every other `xNN` probe | each is the executable control over a retained live component |
+
+**Why SUBSTANTIVE and not TOOLING or CLERICAL.** All three protected files are already
+declared under SUBSTANTIVE amendments (A17/A18/A20, A32, A27/A28/A30/A31/A33/A34), and x04
+refuses a file declared under both a SUBSTANTIVE and a TOOLING amendment. CLERICAL would
+misdescribe the removal of two executable probes as a typo fix. The change alters no scoring
+rule, so `affects_scoring_rule` is **false** — but removing study machinery is a material act
+and is recorded as one, which also means it would be forbidden after the boundary opens.
+
+---
+
 ## A18 — the commit ↔ file accounting of record
 
 ```json