docs(tooling): the green run could not tell sparse from padded; record the rivals that now can (improvise-1xfb.5.1)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdfEGXA7aJ6ftMsCc9PJze

context/code/tooling.md

@@ -1084,6 +1084,13 @@ over an empty CSV was the other option. It was not taken, because it would test
 only how improvise treats a CSV with no records, which is not what this pipeline
 exists to exercise.
 
+**`--out` must hold no CSV the script did not write.** `improvise import .`
+merges every CSV in the directory, so a stale or foreign one would silently
+join the model. The script refuses one with exit 2 and names it. It refuses
+rather than deleting, because deleting files in a directory the caller named is
+the wrong default. Its own five CSVs are overwritten on every run, so a rerun
+into the same `--out` works.
+
 **No collisions, by construction.** A `churn.csv` record is keyed by commit and
 file, which `--no-renames` makes unique. code-maat's records are keyed by
 entity, by entity pair or by date. `File` separates the sources. So the import
@@ -1220,10 +1227,15 @@ known, runs the whole pipeline, code-maat and improvise included, and asserts
 improvise's exported tables. It runs in `nix-build`, which puts improvise and
 code-maat on `PATH`. Locally it uses improvise from `PATH`, or else
 `target/debug/improvise`, and it fails with a message when there is neither.
+The fallback does not check how fresh that binary is, so locally the suite can
+test a `target/debug/improvise` older than the tree. Rebuild after changing
+improvise.
 
 It does not build improvise itself. `checkers.sh` scrubs the environment to an
 allowlist, and on macOS that removes what the linker needs: a `cargo build`
-inside the suite failed with `library not found for -liconv`. Build first with
+inside the suite failed with `library not found for -liconv`. That is a bug in
+`checkers.sh`: its allowlist drops the Nix linker variables, and the
+orchestrator has filed it separately. Build first with
 `cargo build --bin improvise`; `cargo test --workspace` also leaves the binary
 there. The suite is the slowest in `checkers.sh`, at about 25 seconds, most of
 it JVM start-up for code-maat's four analyses per run.

context/test-evidence/tooling.md

@@ -1886,9 +1886,32 @@ agent docs. The coupling count is 16, not 15: `web/app.js` and
 
 Once #314 merged, `improvise import <dir>` replaced the file list, the `-`
 padding and the generated per-file views. The suite needed no change of
-expectations and still passes 41 of 41. That is the evidence that the sparse
-merge aggregates the same numbers the padded one did. The case "the model keeps
-a view per file and the report's four" counts 10 views: `Default`, one per file
-from the import, and the report's four. So a report view named like a file
-stem, which `create-view` would replace silently (improvise-31oe.8), shows up
-as 9.
+expectations and still passed 41 of 41. That shows only that the numbers
+agree. It does **not** show that the model is right: the suite could not tell
+sparse from padded. The delta review re-added `pad_categories` and the suite
+still passed 41 of 41, while the model then had a `|-|` item in 13 places and
+the `churn` view seated `agent-rules` on rows and paged everything else. The
+view case counted 10 views either way.
+
+**What separates them now:**
+- **"the sparse model has no padding item"**: no line of the model names `|-|`.
+- **"the churn view seats …"**: four cases, one per seat. `date` is on rows, and
+  `agent-rules`, `coupled` and `crosses-crate`, which are coupling-only, are on
+  `none`. The churn view puts `entity` on columns, not `area` as the review
+  guessed. What discriminates is `date` on rows and the coupling-only
+  categories on `none`, which a padded CSV cannot give, because it claims all
+  nine categories.
+
+**Rivals:**
+- **Padding re-added** (the review's own injection): 5 cases fail. They are
+  the `|-|` case, with 161 lines naming it in the fixture model, and all four
+  seat cases.
+- **A report view named like a file stem**: `By day` renamed to `churn` leaves
+  9 views, and the view-count case fails.
+- **The foreign-CSV guard removed**: "a foreign CSV in --out is refused with
+  exit 2" fails with exit 0. The guard is the answer to the review's
+  non-blocking finding that `import .` picks up any CSV in `--out`. It refuses
+  instead of clearing. Deleting files in a directory the caller named is the
+  wrong default, and the script's own five CSVs are overwritten on every run,
+  so a rerun into the same `--out` is still accepted. The same fixture run
+  writes JSON and then text into one directory, which shows that.