diff --git a/README.md b/README.md
index dc00f091..b4a4b357 100644
--- a/README.md
+++ b/README.md
@@ -42,7 +42,7 @@ improvise import examples/demo.csv \
   --measure Revenue --measure Cost \
   --time Date --extract Date:Month --split-date Date \
   --axis Region:row --axis Product:row \
-  --axis Date_Month:column --axis Measure:column \
+  --axis Date_Month:column --axis _Measure:column \
   --axis Customer:page --axis Date:none \
   --formula "Profit = Revenue - Cost" \
   --name "Acme Sales Demo" \
diff --git a/context/code/cli.md b/context/code/cli.md
index 8fecbd1d..5add820a 100644
--- a/context/code/cli.md
+++ b/context/code/cli.md
@@ -392,3 +392,36 @@ The headless path matters most, because the code-maat pipeline
 (improvise-1xfb.5.1) runs there. `a_headless_import_of_a_folder_merges_its_csv_files_and_repeats_byte_for_byte`
 pins the views, the sparse Data lines and the byte-identical second run, with
 the first run's output sitting in the folder it reads.
+
+## 13 — a headless import refuses rather than saving less than was asked (improvise-8jk6.4, 8jk6.5, 8jk6.6)
+
+The user treats the headless command and model layer as a product of its own,
+and the scripted `import --no-wizard` is where it meets a pipeline. The
+repo-stats pipeline (improvise-1xfb.5.1) found three ways that import saved
+something other than what the flags asked for, and still exited 0. The
+reasoning for each is in `context/code/formula-and-io.md`, "A CSV field is
+text, and the field's role reads it" and "An import refuses what it cannot
+keep". The CLI's contract:
+
+- **`--category` keeps the CSV text.** A forced category's items are the
+  file's text, verbatim: `00565796` stays `00565796`, and `220944e2` stays
+  `220944e2`. Measures still read as numbers. A JSON number has no source
+  text, so it is named as JSON writes it (`1e3` is `1000.0`).
+- **An unparseable `--formula` is a refusal.** It exits non-zero and writes
+  nothing. Every bad formula is named on stderr, each with its parse error.
+  This is §10's rule: the request was not met, so the exit code says so. The
+  wizard refuses the same text at the formula step, with the same wording.
+- **An import that kept no record is a refusal.** It exits non-zero, writes
+  nothing, and says why: every record was skipped as not a JSON object, or
+  the source had none.
+- **An import that skipped *some* records exits 0 by default.** It writes the
+  model and warns with a count. With `--strict` (the user's ruling on PR #321)
+  any skipped record is a refusal instead: exit 1, the same count named, and
+  nothing written. The argument is in `formula-and-io.md`.
+
+Two refusals bail in `run_headless_import` before the build: the `--strict`
+skip check (`src/main.rs:305`) and the `--on-duplicate refuse` collision
+check. The rest come from `build_model_with_warnings` as an `Err`. All of
+them reach `main` as exit 1 before `persistence::save`. So "writes nothing"
+holds by statement order, not by where the refusal is raised. The headless
+tests check it by asserting that the output path does not exist.
diff --git a/context/code/formula-and-io.md b/context/code/formula-and-io.md
index 1d3bd443..9a8b88ad 100644
--- a/context/code/formula-and-io.md
+++ b/context/code/formula-and-io.md
@@ -711,6 +711,118 @@ one proves nothing.
 backend rather than only a file (improvise-fuj); the path-taking wrappers remain
 for the native CLI.
 
+### A CSV field is text, and the field's role reads it (improvise-8jk6.4)
+
+The parser used to type every field as it read it: text that parsed as an
+`i64` or a finite `f64` became a `Value::Number`. That is a lossy step, and it
+ran before anyone knew what the column was for. A column the user then forced
+to Category (`--category`, or the wizard's `c`) took its item names from the
+number's display, not from the file. So commit `00565796` became item `565796`,
+and `220944e2` was read as scientific notation and became `22094400.0`. The
+repo-stats pipeline (improvise-1xfb.5.1) found it.
+
+CSV has no types. Whether `00565796` is a number depends on whether its column
+is a measure, and the parser cannot know that. So the parser now yields every
+non-empty field as a `Value::String` of its text (blank is still `Null`), and
+two readers apply the old numeric rule, `csv_parser::number_in`, only where a
+number is wanted:
+
+- **Analysis** proposes a kind from a typed copy of the records
+  (`Scalars::typed_record`), so a numeric CSV column is still proposed as a
+  Measure, exactly as before.
+- **A measure** takes its value through `Scalars::number`.
+
+A category or a label is named from the text itself, so it is verbatim.
+
+`ImportPipeline` carries the reading as a private `scalars: Scalars`:
+`Typed` for JSON and `Text` for CSV. JSON cannot share the text rule, because
+a JSON string is text by the file's own say. `"02139"` in JSON is a zip code,
+and reading it as a number would propose it as a Measure. A JSON number
+forced to Category has no source text to keep, because serde_json does not
+keep one. It is named by `Value`'s own display, which is what writing it back
+as JSON would say: `42`, `1.5`, and `1000.0` for `1e3`.
+
+**The parser returns `CsvRecords`, not `Vec<Value>`.** It derefs to
+`[Value]`, so a caller can read it. It converts to a `Vec` only inside the
+crate, and the only public way into a pipeline is `ImportPipeline::from_csv`.
+This is not caution for its own sake. When the parser first changed, eleven
+tests in four files failed. Every one handed parser output to
+`ImportPipeline::new`, the JSON constructor, which would read every column as
+text and propose `Revenue` as a Category. Production had the same shape at all
+three CSV entry points: `main.rs`'s single file, `ImportPipeline::from_merge`
+(the merge path, from PR #314), and `ImportJsonHeadless`. `from_merge` merged
+in with `Self::new` and now calls `Self::from_csv`. Now that mistake does not
+compile. Two holes remain, and
+both take a deliberate step. `ImportPipeline::records` is still a public
+`Vec<Value>`, and a test helper once copied it out of a CSV pipeline into a
+JSON one. And `CsvRecords` derefs to a slice that `.to_vec()` will copy.
+
+### An import refuses what it cannot keep (improvise-8jk6.5, improvise-8jk6.6)
+
+Two failures used to end in an import that saved and reported success.
+
+**A formula that does not parse.** `build_model_with_warnings` ran
+`if let Ok(formula) = parse_formula(..)`, so a bad `--formula` vanished. The
+headless run then saved and exited 0, and the wizard kept the bad text in its
+list and dropped it the same way at build. Both now go through one
+function, `import_formula`, whose error names the formula and carries pest's
+parse error: `Formula error in 'Y = x +': …`. The prefix matches the TUI
+formula minibuffer's `Formula error:`.
+
+- `ImportPipeline::parsed_formulas` parses them all. If any fails, it refuses
+  and names **every** formula that failed, one per line (a message the user
+  needs is never truncated). The build calls it before creating anything, so a
+  refusal writes nothing.
+- The wizard checks at `confirm_formula`, the one place the user can still fix
+  the text (gotcha 22). A formula that does not parse stays in the buffer with
+  the error as the step's message, and never enters the list.
+
+**Nothing kept.** The one skip left after improvise-8jk6.2 is a record that
+is not a JSON object. When every record is skipped, there are no fields, so
+there are no proposals and no categories. The build did refuse that case, but
+by accident, with "At least one category must be accepted". That tells the
+user to change a field's kind when there are no fields. The ticket's own repro
+(`[1, 2, "three"]`) already exited 1 on the lpzu base, so the headline did not
+reproduce there. What did reproduce was the reason. `nothing_to_import` now
+states the rule and runs before the category check: `Nothing to import: all 3
+records were skipped as not JSON objects`, or `the source has no records`. The
+wizard asks it first on `ReviewProposals`, where another array path can still
+be chosen.
+
+One case meets PR #314's "a header-only file is still a file". A folder whose
+CSVs are **all** header-only has no records, only sources. Built from
+`from_merge`'s seeded proposals alone, it would save `File` items, empty
+columns and one empty records view per file. It is refused here, because it
+kept no record. A header-only file *beside* files with rows is unaffected,
+since some record was kept. The user confirmed this on PR #321: "Refuse (as
+#321 does)".
+
+**A partial skip stays a warning.** Some records kept and some skipped still
+builds, exits 0, and warns with a count: `Skipped 1 of 3 records that were not
+JSON objects`. The argument: a non-object in an array of records is not a
+record that failed to import. It is structurally not a record, such as a
+trailing `null` or a stray scalar in a hand-edited file. Every record that
+*is* a record was kept, and the count is on stderr for a script to read.
+Making it fatal would fail every import that carries one stray element.
+Making it a distinct non-zero exit that still writes gives a contract where
+`import … && next` stops even though the file was written.
+
+The user ruled on PR #321: "Add a --strict flag". The default stays exit 0
+with the counted warning, pinned by
+`an_import_that_skips_some_records_warns_with_a_count_and_still_writes`.
+`import --no-wizard --strict` turns any skipped record into a refusal:
+`--strict: Skipped 1 of 3 records that were not JSON objects`, exit 1.
+
+**A strict refusal writes nothing.** A refusal that still wrote the file would
+be the half-met contract argued against above: `&&` stops while the output
+sits there, and a script that checks only the file gets a model that `--strict`
+existed to prevent. So the check runs in `run_headless_import` before the
+build. It reads `ImportPipeline::skip_warning`, the same function the build's
+warning comes from, so the two cannot disagree about what was skipped.
+"Skipped" here means a record that is not an object, the only skip left after
+8jk6.2. An unparsed date or a short CSV row is a warning about a record that
+was kept, and `--strict` does not refuse it.
+
 ### A folder of CSVs is one sparse model, with a view per file (improvise-lpzu)
 
 `improvise import <folder>` exists for improvise-1xfb.5.1. code-maat writes one
diff --git a/context/test-evidence/cli.md b/context/test-evidence/cli.md
index 96ee4f9e..7f9b2233 100644
--- a/context/test-evidence/cli.md
+++ b/context/test-evidence/cli.md
@@ -448,6 +448,57 @@ Rivals R2, R7 and R9 each fail it. Since the review of `4e68372c` the fixture
 also holds a header-only `soc.csv`. Its view must carry `File` at `soc`, be a
 records view, and draw zero rows. Rivals F7 and F8 fail that half.
 
+### a_csv_column_forced_to_category_keeps_its_text_verbatim
+
+improvise-8jk6.4. Before the fix, `commit,lines` with `00565796` and
+`220944e2` forced to Category imported as items `565796` and `22094400.0`.
+The CSV parser had typed both as numbers, and the category was named from the
+number. On the base this failed with the items line `- |565796|, |22094400.0|`.
+The test also asserts both Data lines, which proves the measure `lines` still
+reads 3 and 5 through the same records.
+
+- **R1**, a category named from the typed value, fails it.
+- **R2**, a CSV measure read as a JSON value (`as_f64` on the text), fails it
+  on the Data lines.
+
+### a_json_number_forced_to_category_is_named_as_json_writes_it
+
+improvise-8jk6.4, the JSON half. This is not a bug demonstration: it passed
+on the base, and it pins that the fix left JSON alone. JSON keeps no source
+text for a number, so `42`, `1.5` and `1e3` name items `42`, `1.5` and
+`1000.0`. `1e3` is the discriminating value, because its source text, its f64
+display (`1000`) and its JSON display (`1000.0`) all differ. **R10**, naming a
+number by its `f64` display, fails it on `1000.0`.
+
+### an_import_formula_that_does_not_parse_refuses_names_it_and_writes_nothing
+
+improvise-8jk6.5. Before the fix, `--formula 'Y = x +'` was dropped: the run
+printed `Saved to …` and exited 0. That is the base failure (`a dropped
+formula exited 0`). The test passes a good formula too (`Z = x`) and asserts
+it is not named, so a refusal that lists every formula cannot pass.
+`expected` comes from pest's error, which proves the parse error is carried
+and not only the formula text. **R4**, a build that drops what does not
+parse, fails it. Since improvise-8jk6.17 it passes two bad formulas, so
+**R5**, naming only the first, fails it here too. At `efaa5050` only the io
+unit test saw R5.
+
+### an_import_that_keeps_no_record_refuses_and_says_why
+
+improvise-8jk6.6. On the lpzu base `[1, 2, "three"]` already exited 1, so the
+exit-code half was not the bug. The reason was: `At least one category must
+be accepted`, which is true only because no fields exist, and which points
+the user at a fix that cannot help. The base failure was on the message.
+**R7**, removing `nothing_to_import` from the build, fails it the same way.
+
+### an_import_that_skips_some_records_warns_with_a_count_and_still_writes
+
+improvise-8jk6.6, the partial case. It passed on the base. It pins the
+default the user kept on PR #321 ("Add a --strict flag"), so that changing
+the default is a decision and not a drift. It asserts the warning's count (`1 of 3`), exit 0,
+and a Data line from the record *after* the skipped one. That last check
+shows the skip did not end the import early. **R9**, making any skip fatal,
+fails it.
+
 ### a_folder_of_one_csv_still_gets_a_file_category_and_its_view_but_a_lone_file_does_not
 
 improvise-lpzu, after the user's ruling on PR #314 ("we'd want file identity").
@@ -458,3 +509,42 @@ other half pins the direction the ruling did *not* change: the same CSV named
 directly imports without any `File`. The rival that merges every all-CSV
 import, lone files included, fails that half ("a lone file keeps today's
 shape"). So the test separates "a folder merges" from "everything merges".
+
+### a_strict_import_that_skips_a_record_refuses_and_writes_nothing
+
+improvise-8jk6.6, the user's ruling on PR #321. Written before `--strict`
+existed, it failed on clap's unknown-argument error. It asserts the refusal
+names the flag and the same count as the default warning, and that the output
+path does not exist.
+
+- **S1**, the flag parsed but ignored, fails it on the exit status.
+- **S3**, refusing only after `persistence::save`, fails it on the output
+  path. That is the "a strict refusal still writes" contract this PR decided
+  against.
+
+### a_strict_import_that_skips_nothing_still_writes
+
+The control for the entry above. Without it, a `--strict` that refuses
+**every** import passes that test. **S2** is that rival, and it fails here.
+It failed before the flag existed too, but only on clap's error, which is not
+the discriminating failure. The discriminating one is S2.
+
+### a_csv_decimal_measure_imports_as_its_number
+
+improvise-8jk6.17 (F3), from the adversarial review of #321. The fixture has
+decimal CSV measures (`1000.50`, `-2.25`). With `number_in`'s `f64` branch
+disabled (integers only), no headless test failed at `efaa5050`, and a decimal
+measure was silently dropped. **F3** now fails this test and the next one.
+This is coverage, not a bug: it passed when written.
+
+### import_json_on_a_csv_reads_its_numeric_column_as_a_measure
+
+improvise-8jk6.17 (F1). `improvise cmd 'import-json f.csv'` reaches
+`ImportJsonHeadless`, a third CSV entry point besides `import`. **F1**
+restores its CSV branch to the JSON constructor via
+`ImportPipeline::new(Value::Array(recs.to_vec()))`. That still compiles,
+because `CsvRecords` derefs to a slice. `Revenue` then becomes a category of
+text items, and this test fails on its Data lines. At `efaa5050` only
+`import_headless_refuses_records_that_collide` failed under F1, and only by
+accident. The `## Category: Revenue` assertion names the symptom directly.
+This is coverage: it passed when written.
diff --git a/context/test-evidence/formula-and-io.md b/context/test-evidence/formula-and-io.md
index 5523b05a..5284af34 100644
--- a/context/test-evidence/formula-and-io.md
+++ b/context/test-evidence/formula-and-io.md
@@ -882,6 +882,11 @@ The control: a **lone** `\r` is itself a record terminator to this reader, so
 it is not part of a CRLF pair, and it is why "two records, clean values" is a
 real observation rather than a coincidence.
 
+Since improvise-8jk6.4 the parser yields text, so the assertion is that
+`Revenue` is the text `1`. A kept `\r` would make it `1\r`, which fails that
+assertion, and which `number_in` would not read as a number where a measure
+wanted one.
+
 ### an_empty_file_is_no_records_and_no_warnings
 
 An empty file is no records and no complaint. The wizard's own preconditions
@@ -919,12 +924,17 @@ The consequence downstream is deliberate and worth stating: a column carrying
 Label rather than a Measure. That is the honest answer — the column is not
 numeric — and it is visible in the wizard, where a fabricated zero was not.
 
+Since improvise-8jk6.4 the numeric rule is `csv_parser::number_in`, applied
+through `typed` only where a number is wanted, and the parser itself yields
+text. So this test asserts that `typed` leaves each token as text. The
+`unwrap_or` fault would be injected into `number_in` now.
+
 ### an_ordinary_float_is_still_a_number
 
 The control for the entry above. Guarding non-finiteness by deleting the float
 branch entirely would also pass that test, so this one pins that `1000.50` is
 still a `Number` and that `0` — the value the bug invented — is still reachable
-when the file really says so.
+when the file really says so. Since improvise-8jk6.4 both go through `typed`.
 
 ### a_repeated_header_is_renamed_rather_than_overwriting_its_twin
 
@@ -1856,6 +1866,68 @@ document still has the three views in order, `File: filter` at the file's item,
 and the same drawn values. This is the first persisted, non-virtual use of
 `Axis::Filter` (gotcha 27). R7 and R9 fail it.
 
+### a_csv_column_read_as_a_category_keeps_its_text_and_as_a_measure_its_number
+
+improvise-8jk6.4, at the pipeline. It checks three things in one fixture,
+one per reader of a CSV value:
+
+- analysis still proposes the numeric-looking `commit` as a Measure (typed);
+- forced to Category, its items are `00565796` and `220944e2` (text);
+- `lines` is stored as the number 5 (typed).
+
+The rivals, each run:
+
+- **R1**, a category named from the typed value, fails it on the items.
+- **R2**, a CSV measure read with `as_f64` on its text, fails it on the cell.
+- **R3**, CSV analysed without typing, fails it on the proposal.
+
+R2 and R3 also fail the five lpzu merge tests in this module and five
+headless imports. That is how much of the suite leans on CSV measures reading
+as numbers.
+
+This test uses `from_csv` and so cannot compile on the base. The base failure
+for 8jk6.4 is the headless test
+`a_csv_column_forced_to_category_keeps_its_text_verbatim`.
+
+### a_formula_that_does_not_parse_stops_the_build_and_is_named
+
+improvise-8jk6.5. On the base the build returned `Ok`, keeping `Good` and
+silently dropping `Bad`. Two formulas fail to parse here, so the test proves
+every failure is named. **R5**, naming only the first, fails it on `Worse`,
+and since improvise-8jk6.17 the headless
+`an_import_formula_that_does_not_parse_refuses_names_it_and_writes_nothing`
+sees R5 too (it passes two bad formulas). It also asserts `Good` is not named, so a
+message that lists every formula cannot pass. **R4** (drop, as the base did)
+fails it.
+
+### confirming_a_formula_that_does_not_parse_keeps_it_out_and_the_text_in
+
+improvise-8jk6.5, the wizard half. On the base `confirm_formula` pushed
+`Bad = revenue +` into the list, which failed on `pipeline.formulas`. The test
+asserts the text is still in the buffer and the editor still open. A fix that
+refused but cleared the buffer would throw away what the user typed, the same
+hazard improvise-v4hg fixed in the TUI formula minibuffer. **R6**, confirming
+regardless, fails it. The headless suite cannot see R6.
+
+### a_build_that_keeps_no_record_refuses_and_says_why
+
+improvise-8jk6.6. On the base both halves got `At least one category must be
+accepted`, a refusal right only by accident. It asserts the exact message for
+the two cases `nothing_to_import` distinguishes: all skipped, and no records
+at all.
+
+- **R7**, the check removed from the build, fails it.
+- **R9**, `any` swapped for `all`, also fails it, on the empty source: `all`
+  is vacuously true over no records.
+
+### the_review_step_says_when_no_record_can_be_imported
+
+improvise-8jk6.6, the wizard half (gotcha 22: the check lives on the step
+where the user can still choose another array path). On the base the step
+refused with `No categories: … Press c to cycle a field's kind`, which offers
+a key that has no field to act on. **R8**, the step without the check, fails
+it. The build test does not see R8.
+
 ### a_header_only_file_keeps_its_file_item_its_columns_and_its_view
 
 The ChatGPT Codex P2 on PR #314 at `4e68372c`, confirmed by the adversarial
@@ -1943,3 +2015,34 @@ to build one `Date` category and drop the `--axis` for the vanished
 `Date_Month`, exit 0. It now exits 1 naming the category and `--split-date`.
 This also refuses any `--axis` naming an absent category, which was silently
 ignored before this PR as well.
+
+### a_folder_of_only_header_only_files_keeps_no_record_and_refuses
+
+improvise-8jk6.6 meeting PR #314's rule that a header-only file is still a
+file. Written after merging `main` at `c91cc055`, where the two rules first
+met. The file seeds give the pipeline a `File` Category proposal even with no
+records. So without `nothing_to_import` the build succeeds and saves `File`
+items and empty views: the category check alone cannot refuse. The `setup`
+assertion proves that seeded proposal exists, so the test is not passing only
+because there was no category. **R7**, the check removed from the build, fails
+it. The user confirmed the refusal on PR #321: "Refuse (as #321 does)".
+
+### an_absent_axis_category_that_is_no_date_part_names_what_was_built
+
+improvise-8jk6.3 review (B2). The README's own recipe passed
+`--axis Measure:column`, a silent no-op because the category is `_Measure`; the
+first refusal blamed dates for it. The test uses a non-date absent name and
+asserts the message lists the built categories and carries no date hint.
+Fault injected: making the hint unconditional fails it.
+
+### the_readmes_import_recipe_runs
+
+improvise-8jk6.3 review (B2). Runs the README's recipe as printed. Fault
+injected: restoring `--axis Measure:column` in the README fails it with the
+refusal.
+
+### split_date_on_a_time_field_with_no_date_format_is_refused
+
+improvise-8jk6.3 review. `--time Revenue --split-date Revenue` made the field a
+time category with no date format and the split was a silent no-op. Without
+the `date_format.is_some()` condition the import exits 0 and this fails.
diff --git a/crates/improvise-command/src/ui/effect.rs b/crates/improvise-command/src/ui/effect.rs
index a0b674dd..17a947f7 100644
--- a/crates/improvise-command/src/ui/effect.rs
+++ b/crates/improvise-command/src/ui/effect.rs
@@ -1247,7 +1247,7 @@ pub struct ImportJsonHeadless {
 }
 impl Effect for ImportJsonHeadless {
     fn apply(&self, app: &mut App) {
-        use crate::import::analyzer::{FieldKind, analyze_records};
+        use crate::import::analyzer::FieldKind;
         use crate::import::wizard::{DuplicatePolicy, ImportPipeline};
 
         let is_csv = self
@@ -1255,10 +1255,10 @@ impl Effect for ImportJsonHeadless {
             .extension()
             .is_some_and(|ext| ext.eq_ignore_ascii_case("csv"));
 
-        // Parse the file exactly once: `records` feeds analysis, `raw` is reused
-        // as the pipeline's value (improvise-oaq). CSV warnings go to the status.
+        // Parse the file exactly once, into the pipeline that reads it
+        // (improvise-oaq, improvise-8jk6.4). CSV warnings go to the status.
         let mut warnings: Vec<String> = vec![];
-        let (records, raw) = if is_csv {
+        let mut pipeline = if is_csv {
             let bytes = match app.storage.read(&self.path) {
                 Ok(b) => b,
                 Err(e) => {
@@ -1269,8 +1269,7 @@ impl Effect for ImportJsonHeadless {
             match crate::import::csv_parser::parse_csv_bytes_with_warnings(&bytes) {
                 Ok((recs, w)) => {
                     warnings = w;
-                    let raw = serde_json::Value::Array(recs.clone());
-                    (recs, raw)
+                    ImportPipeline::from_csv(recs)
                 }
                 Err(e) => {
                     refuse_and_stop(app, format!("CSV error: {e}"));
@@ -1293,38 +1292,28 @@ impl Effect for ImportJsonHeadless {
                 }
             };
             match json_import_records(&value, self.array_path.as_deref()) {
-                Ok(recs) => (recs, value),
+                Ok(recs) => {
+                    let mut pipeline = ImportPipeline::new(value);
+                    pipeline.selected_path = self.array_path.clone().unwrap_or_default();
+                    pipeline.records = recs;
+                    pipeline.analyze();
+                    pipeline
+                }
                 Err(msg) => {
                     refuse_and_stop(app, msg);
                     return;
                 }
             }
         };
-
-        let proposals = analyze_records(&records);
-
-        let pipeline = ImportPipeline {
-            raw,
-            array_paths: vec![],
-            selected_path: self.array_path.as_deref().unwrap_or("").to_string(),
-            records,
-            proposals: proposals
-                .into_iter()
-                .map(|mut p| {
-                    p.accepted = p.kind != FieldKind::Label;
-                    p
-                })
-                .collect(),
-            model_name: self
-                .model_name
-                .as_deref()
-                .unwrap_or("Imported Model")
-                .to_string(),
-            formulas: vec![],
-            // Headless: nobody to ask, so a collision refuses (improvise-cgqt).
-            on_duplicate: DuplicatePolicy::Refuse,
-            sources: vec![],
-        };
+        pipeline.model_name = self
+            .model_name
+            .clone()
+            .unwrap_or_else(|| "Imported Model".to_string());
+        // Headless: nobody to ask, so a collision refuses (improvise-cgqt).
+        pipeline.on_duplicate = DuplicatePolicy::Refuse;
+        for p in &mut pipeline.proposals {
+            p.accepted = p.kind != FieldKind::Label;
+        }
 
         match pipeline.build_model_with_warnings() {
             Ok((new_workbook, build_warnings)) => {
diff --git a/crates/improvise-io/src/import/csv_parser.rs b/crates/improvise-io/src/import/csv_parser.rs
index 355b27fb..3a7cadfc 100644
--- a/crates/improvise-io/src/import/csv_parser.rs
+++ b/crates/improvise-io/src/import/csv_parser.rs
@@ -7,14 +7,32 @@ use serde_json::Value;
 
 use crate::storage::{PhysicalFs, Storage};
 
+/// Parsed CSV records. Every value is its source text, so they reach an import
+/// only through `ImportPipeline::from_csv` (improvise-8jk6.4).
+#[derive(Debug, Clone, Default, PartialEq)]
+pub struct CsvRecords(Vec<Value>);
+
+impl std::ops::Deref for CsvRecords {
+    type Target = [Value];
+    fn deref(&self) -> &[Value] {
+        &self.0
+    }
+}
+
+impl CsvRecords {
+    pub(crate) fn into_values(self) -> Vec<Value> {
+        self.0
+    }
+}
+
 pub fn csv_path_p(path: &Path) -> bool {
     path.extension()
         .is_some_and(|ext| ext.eq_ignore_ascii_case("csv"))
 }
 
-/// Parse a CSV file into a `serde_json::Value` array. Short-row warnings go to
-/// stderr; [`parse_csv_with_warnings`] captures them instead.
-pub fn parse_csv(path: &Path) -> Result<Vec<Value>> {
+/// Parse a CSV file into records. Short-row warnings go to stderr;
+/// [`parse_csv_with_warnings`] captures them instead.
+pub fn parse_csv(path: &Path) -> Result<CsvRecords> {
     let (records, warnings) = parse_csv_with_warnings(path)?;
     for warning in &warnings {
         eprintln!("warning: {warning}");
@@ -24,7 +42,7 @@ pub fn parse_csv(path: &Path) -> Result<Vec<Value>> {
 
 /// Parse a CSV file, returning records plus warnings. A short row keeps every
 /// header column, missing ones `Null`, each reported with a row count.
-pub fn parse_csv_with_warnings(path: &Path) -> Result<(Vec<Value>, Vec<String>)> {
+pub fn parse_csv_with_warnings(path: &Path) -> Result<(CsvRecords, Vec<String>)> {
     let bytes = std::fs::read(path)
         .with_context(|| format!("Failed to open CSV file: {}", path.display()))?;
     parse_csv_bytes_with_warnings(&bytes)
@@ -32,9 +50,9 @@ pub fn parse_csv_with_warnings(path: &Path) -> Result<(Vec<Value>, Vec<String>)>
 
 /// As [`parse_csv_with_warnings`] but over bytes, so import can come from any
 /// [`Storage`](crate::storage::Storage) backend, not only a file (improvise-fuj).
-pub fn parse_csv_bytes_with_warnings(bytes: &[u8]) -> Result<(Vec<Value>, Vec<String>)> {
+pub fn parse_csv_bytes_with_warnings(bytes: &[u8]) -> Result<(CsvRecords, Vec<String>)> {
     let (_, records, warnings) = parse_csv_bytes(bytes)?;
-    Ok((records, warnings))
+    Ok((CsvRecords(records), warnings))
 }
 
 /// The header (after renaming), the records and the warnings.
@@ -152,7 +170,7 @@ fn disambiguate_headers(raw: Vec<String>) -> (Vec<String>, Vec<String>) {
 
 /// Parse multiple CSV files and merge into a single JSON array.
 /// Each record gets a "File" field set to the filename stem (e.g., "sales" from "sales.csv").
-pub fn merge_csvs(paths: &[impl AsRef<Path>]) -> Result<Vec<Value>> {
+pub fn merge_csvs(paths: &[impl AsRef<Path>]) -> Result<CsvRecords> {
     let merged = merge_csvs_from(&PhysicalFs, paths)?;
     for warning in &merged.warnings {
         eprintln!("warning: {warning}");
@@ -173,7 +191,7 @@ pub struct SourceFile {
 /// Several CSV files as one record set, with each file's own columns kept.
 #[derive(Debug, Clone, PartialEq)]
 pub struct Merged {
-    pub records: Vec<Value>,
+    pub records: CsvRecords,
     /// One per distinct `File` item, in the order the files were read.
     pub sources: Vec<SourceFile>,
     /// Each file's parse warnings, prefixed with its path.
@@ -184,7 +202,7 @@ pub struct Merged {
 /// and two files with one stem share its item and its columns (improvise-lpzu).
 pub fn merge_csvs_from(fs: &dyn Storage, paths: &[impl AsRef<Path>]) -> Result<Merged> {
     let mut merged = Merged {
-        records: Vec::new(),
+        records: CsvRecords::default(),
         sources: Vec::new(),
         warnings: Vec::new(),
     };
@@ -212,7 +230,7 @@ pub fn merge_csvs_from(fs: &dyn Storage, paths: &[impl AsRef<Path>]) -> Result<M
             if let Value::Object(ref mut map) = record {
                 map.insert(FILE_CATEGORY.to_string(), Value::String(stem.clone()));
             }
-            merged.records.push(record);
+            merged.records.0.push(record);
         }
     }
     Ok(merged)
@@ -266,26 +284,33 @@ fn csv_documents_in(fs: &dyn Storage, dir: &Path) -> Result<Vec<PathBuf>> {
     Ok(csvs)
 }
 
+/// A field is its source text: which ones are numbers is decided by the role
+/// the import gives the column (improvise-8jk6.4, `context/code/formula-and-io.md`).
 fn parse_csv_field(field: &str) -> Value {
-    if field.is_empty() {
-        return Value::Null;
+    match field.is_empty() {
+        true => Value::Null,
+        false => Value::String(field.to_string()),
     }
+}
 
-    // Try to parse as number (integer or float)
-    if let Ok(num) = field.parse::<i64>() {
-        return Value::Number(serde_json::Number::from(num));
+/// The number a CSV field's text reads as, where a number is wanted.
+pub fn number_in(text: &str) -> Option<serde_json::Number> {
+    if let Ok(num) = text.parse::<i64>() {
+        return Some(serde_json::Number::from(num));
     }
-
     // NaN and the infinities have no JSON number. A fabricated 0 is an invented
-    // measurement (improvise-jc95), so the token falls through and stays text.
-    if let Ok(num) = field.parse::<f64>()
-        && let Some(number) = serde_json::Number::from_f64(num)
-    {
-        return Value::Number(number);
-    }
+    // measurement (improvise-jc95), so the token stays text.
+    text.parse::<f64>()
+        .ok()
+        .and_then(serde_json::Number::from_f64)
+}
 
-    // Otherwise treat as string
-    Value::String(field.to_string())
+/// A CSV value as a typed JSON value: numeric text becomes its number.
+pub fn typed(value: &Value) -> Value {
+    match value.as_str().and_then(number_in) {
+        Some(number) => Value::Number(number),
+        None => value.clone(),
+    }
 }
 
 #[cfg(test)]
@@ -310,8 +335,9 @@ mod tests {
         assert_eq!(records.len(), 2);
         assert_eq!(records[0]["Region"], Value::String("East".to_string()));
         assert_eq!(records[0]["Product"], Value::String("Shirts".to_string()));
+        assert_eq!(records[0]["Revenue"], Value::String("1000".to_string()));
         assert_eq!(
-            records[0]["Revenue"],
+            typed(&records[0]["Revenue"]),
             Value::Number(serde_json::Number::from(1000))
         );
     }
@@ -323,9 +349,10 @@ mod tests {
         let records = parse_csv(&path).unwrap();
 
         assert_eq!(records.len(), 2);
-        assert!(records[0]["Revenue"].is_f64());
+        assert_eq!(records[0]["Revenue"], Value::String("1000.50".to_string()));
+        assert!(typed(&records[0]["Revenue"]).is_f64());
         assert_eq!(
-            records[0]["Revenue"],
+            typed(&records[0]["Revenue"]),
             Value::Number(serde_json::Number::from_f64(1000.50).unwrap())
         );
     }
@@ -363,10 +390,10 @@ mod tests {
         assert_eq!(records.len(), 2);
         assert_eq!(records[0]["Name"], Value::String("Widget".to_string()));
         assert_eq!(
-            records[0]["Count"],
+            typed(&records[0]["Count"]),
             Value::Number(serde_json::Number::from(5))
         );
-        assert!(records[0]["Price"].is_f64());
+        assert!(typed(&records[0]["Price"]).is_f64());
         assert_eq!(records[0]["Active"], Value::String("true".to_string()));
     }
 
@@ -386,10 +413,7 @@ mod tests {
         assert_eq!(records[3]["File"], Value::String("expenses".to_string()));
         // Original fields preserved
         assert_eq!(records[0]["Region"], Value::String("East".to_string()));
-        assert_eq!(
-            records[2]["Revenue"],
-            Value::Number(serde_json::Number::from(50))
-        );
+        assert_eq!(records[2]["Revenue"], Value::String("50".to_string()));
     }
 
     #[test]
@@ -492,10 +516,10 @@ mod tests {
             ]
         );
         assert_eq!(
-            merged.records,
+            *merged.records,
             [
-                serde_json::json!({"entity": "src/a.rs", "coupled": "src/b.rs", "degree": 80, "File": "coupling"}),
-                serde_json::json!({"entity": "src/a.rs", "age-months": 2, "File": "age"}),
+                serde_json::json!({"entity": "src/a.rs", "coupled": "src/b.rs", "degree": "80", "File": "coupling"}),
+                serde_json::json!({"entity": "src/a.rs", "age-months": "2", "File": "age"}),
             ]
         );
     }
@@ -656,10 +680,7 @@ mod tests {
         assert!(warnings.is_empty());
         assert_eq!(records.len(), 2);
         assert_eq!(records[0]["Region"], Value::String("East".to_string()));
-        assert_eq!(
-            records[0]["Revenue"],
-            Value::Number(serde_json::Number::from(1))
-        );
+        assert_eq!(records[0]["Revenue"], Value::String("1".to_string()));
         assert_eq!(records[1]["Region"], Value::String("West".to_string()));
 
         let (control, _) = parse_csv_bytes_with_warnings(b"Region,Revenue\nEa\rst,1\n").unwrap();
@@ -714,7 +735,8 @@ mod tests {
                 token.parse::<f64>().is_ok_and(|n| !n.is_finite()),
                 "precondition: {token} parses as a non-finite f64"
             );
-            assert_eq!(parse_csv_field(token), Value::String(token.to_string()));
+            let text = Value::String(token.to_string());
+            assert_eq!(typed(&text), text);
         }
     }
 
@@ -723,10 +745,10 @@ mod tests {
     #[test]
     fn an_ordinary_float_is_still_a_number() {
         assert_eq!(
-            parse_csv_field("1000.50"),
+            typed(&Value::String("1000.50".into())),
             Value::Number(serde_json::Number::from_f64(1000.50).unwrap())
         );
-        assert_eq!(parse_csv_field("0"), Value::Number(0.into()));
+        assert_eq!(typed(&Value::String("0".into())), Value::Number(0.into()));
     }
 
     /// improvise-5hfh: two equal header names are one JSON key, so the later
@@ -735,9 +757,9 @@ mod tests {
     fn a_repeated_header_is_renamed_rather_than_overwriting_its_twin() {
         let (records, warnings) = parse_csv_bytes_with_warnings(b"A,B,A\n1,2,3\n").unwrap();
         let map = records[0].as_object().unwrap();
-        assert_eq!(map["A"], Value::Number(1.into()));
-        assert_eq!(map["B"], Value::Number(2.into()));
-        assert_eq!(map["A_2"], Value::Number(3.into()));
+        assert_eq!(map["A"], Value::String("1".into()));
+        assert_eq!(map["B"], Value::String("2".into()));
+        assert_eq!(map["A_2"], Value::String("3".into()));
         assert_eq!(warnings.len(), 1);
         assert!(warnings[0].contains("'A'") && warnings[0].contains("'A_2'"));
         assert!(warnings[0].contains("column 3"), "{}", warnings[0]);
@@ -750,8 +772,8 @@ mod tests {
         let (records, _) = parse_csv_bytes_with_warnings(b"A,A_2,A\n1,2,3\n").unwrap();
         let map = records[0].as_object().unwrap();
         assert_eq!(map.len(), 3, "{map:?}");
-        assert_eq!(map["A_2"], Value::Number(2.into()));
-        assert_eq!(map["A_3"], Value::Number(3.into()));
+        assert_eq!(map["A_2"], Value::String("2".into()));
+        assert_eq!(map["A_3"], Value::String("3".into()));
     }
 
     #[test]
@@ -786,8 +808,9 @@ mod tests {
 
         assert_eq!(records.len(), 2);
         assert_eq!(records[0]["Date"], Value::String("03/31/2026".to_string()));
+        assert_eq!(records[0]["Amount"], Value::String("-50.00".to_string()));
         assert_eq!(
-            records[0]["Amount"],
+            typed(&records[0]["Amount"]),
             Value::Number(serde_json::Number::from_f64(-50.00).unwrap())
         );
         assert_eq!(records[0]["Flag"], Value::String("*".to_string()));
@@ -797,7 +820,7 @@ mod tests {
             Value::String("VENMO PAYMENT 260331".to_string())
         );
         assert_eq!(
-            records[1]["Amount"],
+            typed(&records[1]["Amount"]),
             Value::Number(serde_json::Number::from_f64(-240.00).unwrap())
         );
     }
diff --git a/crates/improvise-io/src/import/wizard.rs b/crates/improvise-io/src/import/wizard.rs
index 6c8cee29..5bac7ec5 100644
--- a/crates/improvise-io/src/import/wizard.rs
+++ b/crates/improvise-io/src/import/wizard.rs
@@ -7,8 +7,8 @@ use super::analyzer::{
     DateComponent, DateShape, FieldKind, FieldProposal, analyze_records, extract_array_at_path,
     extract_date_component, find_array_paths,
 };
-use super::csv_parser::{FILE_CATEGORY, Merged, SourceFile};
-use crate::formula::parse_formula;
+use super::csv_parser::{self, CsvRecords, FILE_CATEGORY, Merged, SourceFile};
+use crate::formula::{Formula, parse_formula};
 use crate::model::MAX_CATEGORIES;
 use crate::model::cell::{CellKey, CellValue};
 use crate::view::{Axis, VIRTUAL_VIEW_NAMES, View, ViewStep, records_steps};
@@ -33,6 +33,42 @@ pub struct ImportPipeline {
     /// The files a CSV merge read, each of which gets its own view. Empty for
     /// a single file or a JSON import (improvise-lpzu).
     pub sources: Vec<SourceFile>,
+    /// What the records' string values are, fixed by the constructor (8jk6.4).
+    scalars: Scalars,
+}
+
+/// How a record's values read. Why CSV is text: `context/code/formula-and-io.md`.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum Scalars {
+    /// JSON: a string is text and a number is a number.
+    Typed,
+    /// CSV: every value is source text, read as a number only where one is wanted.
+    Text,
+}
+
+impl Scalars {
+    /// The record as analysis sees it: typed, whatever its source.
+    fn typed_record(self, record: &Value) -> Value {
+        match (self, record) {
+            (Scalars::Text, Value::Object(map)) => Value::Object(
+                map.iter()
+                    .map(|(k, v)| (k.clone(), csv_parser::typed(v)))
+                    .collect(),
+            ),
+            (Scalars::Text, other) | (Scalars::Typed, other) => other.clone(),
+        }
+    }
+
+    /// The value a measure takes from `value`, if it is a number.
+    fn number(self, value: &Value) -> Option<f64> {
+        match self {
+            Scalars::Typed => value.as_f64(),
+            Scalars::Text => value
+                .as_str()
+                .and_then(csv_parser::number_in)
+                .and_then(|n| n.as_f64()),
+        }
+    }
 }
 
 /// Several source records that land on one coordinate tuple (improvise-cgqt).
@@ -57,7 +93,17 @@ pub enum DuplicatePolicy {
 }
 
 impl ImportPipeline {
+    /// A pipeline over a JSON value.
     pub fn new(raw: Value) -> Self {
+        Self::reading(raw, Scalars::Typed)
+    }
+
+    /// A pipeline over CSV records, whose values are their source text.
+    pub fn from_csv(records: CsvRecords) -> Self {
+        Self::reading(Value::Array(records.into_values()), Scalars::Text)
+    }
+
+    fn reading(raw: Value, scalars: Scalars) -> Self {
         let array_paths = find_array_paths(&raw);
         let mut pipeline = Self {
             raw: raw.clone(),
@@ -69,6 +115,7 @@ impl ImportPipeline {
             formulas: vec![],
             on_duplicate: DuplicatePolicy::Refuse,
             sources: vec![],
+            scalars,
         };
 
         // Auto-select if root is an array or there is exactly one candidate path.
@@ -84,7 +131,7 @@ impl ImportPipeline {
 
     /// A pipeline over a CSV merge, carrying its sources for the per-file views.
     pub fn from_merge(merged: Merged) -> Self {
-        let mut pipeline = Self::new(Value::Array(merged.records));
+        let mut pipeline = Self::from_csv(merged.records);
         pipeline.seed_sources(merged.sources);
         pipeline
     }
@@ -112,7 +159,52 @@ impl ImportPipeline {
         self.selected_path = path.to_string();
         if let Some(arr) = extract_array_at_path(&self.raw, path) {
             self.records = arr.clone();
-            self.proposals = analyze_records(&self.records);
+            self.analyze();
+        }
+    }
+
+    /// Propose a kind for every field, reading the records as `scalars` says.
+    pub fn analyze(&mut self) {
+        let typed: Vec<Value> = self
+            .records
+            .iter()
+            .map(|r| self.scalars.typed_record(r))
+            .collect();
+        self.proposals = analyze_records(&typed);
+    }
+
+    /// The warning for the records the build will skip, if it skips any.
+    pub fn skip_warning(&self) -> Option<String> {
+        let skipped = self.records.iter().filter(|r| !r.is_object()).count();
+        skip_warnings(self.records.len(), skipped)
+    }
+
+    /// Why there is nothing to build, when no record is an object to place.
+    pub fn nothing_to_import(&self) -> Option<String> {
+        let total = self.records.len();
+        match (total, self.records.iter().any(Value::is_object)) {
+            (_, true) => None,
+            (0, false) => Some("Nothing to import: the source has no records".into()),
+            (_, false) => Some(format!(
+                "Nothing to import: all {total} records were skipped as not JSON objects"
+            )),
+        }
+    }
+
+    /// The formulas, parsed. Every one that does not parse is named with its
+    /// error, never dropped (improvise-8jk6.5).
+    pub fn parsed_formulas(&self) -> Result<Vec<Formula>> {
+        let mut parsed = vec![];
+        let mut refused: Vec<String> = vec![];
+        for raw in &self.formulas {
+            match import_formula(raw) {
+                Ok(formula) => parsed.push(formula),
+                Err(e) => refused.push(e.to_string()),
+            }
+        }
+        match refused.is_empty() {
+            true => Ok(parsed),
+            false => Err(anyhow!(refused.join("\n"))),
         }
     }
 
@@ -244,6 +336,7 @@ impl ImportPipeline {
             labels: accepted(&[FieldKind::Label]),
             date_extractions,
             date_hierarchies,
+            scalars: self.scalars,
         }
     }
 
@@ -254,9 +347,13 @@ impl ImportPipeline {
         let categories = &plan.categories;
         let date_extractions = &plan.date_extractions;
 
+        if let Some(why) = self.nothing_to_import() {
+            return Err(anyhow!(why));
+        }
         if categories.is_empty() {
             return Err(anyhow!("At least one category must be accepted"));
         }
+        let formulas = self.parsed_formulas()?;
 
         let placements: Vec<Placement> = self.records.iter().map(|r| plan.place(r)).collect();
         if let Some(clash) = plan.spelling_clash(&placements) {
@@ -313,7 +410,6 @@ impl ImportPipeline {
             wb.add_item("_Measure", &m.field)?;
         }
 
-        let mut skipped_non_objects = 0usize;
         // Per-extraction tally of values the detected format could not parse. A
         // derived coordinate that is merely absent is improvise-yss5.
         let mut unparsed_dates = vec![0usize; date_extractions.len() + plan.date_hierarchies.len()];
@@ -329,7 +425,7 @@ impl ImportPipeline {
 
         for (at, placement) in placements.into_iter().enumerate() {
             match placement {
-                Placement::NotObject => skipped_non_objects += 1,
+                Placement::NotObject => {}
                 Placement::Placed {
                     mut coords,
                     measures,
@@ -381,22 +477,15 @@ impl ImportPipeline {
             wb.model.set_cell(cell.key, CellValue::Number(cell.total));
         }
 
-        // Parse and add formulas
-        // Formulas target the "_Measure" category by default.
-        let formula_cat: String = "_Measure".to_string();
-        for raw in &self.formulas {
-            if let Ok(formula) = parse_formula(raw, &formula_cat) {
-                wb.model.add_formula(formula);
-            }
+        for formula in formulas {
+            wb.model.add_formula(formula);
         }
 
         for view in source_views(&wb, &self.sources, &plan) {
             wb.views.insert(view.name.clone(), view);
         }
 
-        let mut warnings: Vec<String> = skip_warnings(self.records.len(), skipped_non_objects)
-            .into_iter()
-            .collect();
+        let mut warnings: Vec<String> = self.skip_warning().into_iter().collect();
         let (split, whole) = unparsed_dates.split_at(date_extractions.len());
         warnings.extend(unparsed_date_warnings(date_extractions, split));
         warnings.extend(
@@ -508,6 +597,7 @@ struct Plan<'a> {
     labels: Vec<&'a FieldProposal>,
     date_extractions: Vec<(&'a str, &'a str, DateComponent, String)>,
     date_hierarchies: Vec<DateHierarchy<'a>>,
+    scalars: Scalars,
 }
 
 /// A date column imported as ONE category: its items are the finest chosen
@@ -616,7 +706,7 @@ impl Plan<'_> {
         let measures = self
             .measures
             .iter()
-            .filter_map(|m| Some((m.field.clone(), map.get(&m.field)?.as_f64()?)))
+            .filter_map(|m| Some((m.field.clone(), self.scalars.number(map.get(&m.field)?)?)))
             .collect();
         Placement::Placed {
             coords,
@@ -805,6 +895,11 @@ fn unparsed_date_warnings(
         .collect()
 }
 
+/// An import formula, targeting `_Measure`, or the error naming it (improvise-8jk6.5).
+fn import_formula(raw: &str) -> Result<Formula> {
+    parse_formula(raw, "_Measure").map_err(|e| anyhow!("Formula error in '{raw}': {e}"))
+}
+
 /// The warning for records that could not be placed in the grid. A record
 /// missing a category value is placed, so only a non-object is skipped.
 fn skip_warnings(total: usize, skipped_non_objects: usize) -> Option<String> {
@@ -992,7 +1087,9 @@ impl ImportWizard {
     fn validation_error(&self) -> Option<String> {
         match self.step {
             WizardStep::ReviewProposals => self
-                .no_category_message()
+                .pipeline
+                .nothing_to_import()
+                .or_else(|| self.no_category_message())
                 .or_else(|| self.category_limit_message()),
             WizardStep::ConfigureDates => self
                 .category_limit_message()
@@ -1301,14 +1398,24 @@ impl ImportWizard {
         self.formula_buffer.pop();
     }
 
-    /// Commit the current formula buffer to the pipeline's formula list.
+    /// Commit the current formula buffer to the pipeline's formula list. One
+    /// that does not parse stays in the buffer, with the error (improvise-8jk6.5).
     pub fn confirm_formula(&mut self) {
         let text = self.formula_buffer.trim().to_string();
-        if !text.is_empty() {
-            self.pipeline.formulas.push(text);
+        match (text.is_empty(), import_formula(&text)) {
+            (false, Err(e)) => self.message = Some(e.to_string()),
+            (false, Ok(_)) => {
+                self.pipeline.formulas.push(text);
+                self.close_formula_edit();
+            }
+            (true, _) => self.close_formula_edit(),
         }
+    }
+
+    fn close_formula_edit(&mut self) {
         self.formula_buffer.clear();
         self.formula_editing = false;
+        self.message = None;
         self.cursor = self.pipeline.formulas.len().saturating_sub(1);
     }
 
@@ -2430,8 +2537,11 @@ mod tests {
         ),
     ];
 
-    fn code_maat_records() -> Value {
-        Value::Array(merged_pipeline(&CODE_MAAT).records)
+    /// The code-maat merge without its sources, so no per-file views.
+    fn code_maat_without_views() -> ImportPipeline {
+        let mut p = merged_pipeline(&CODE_MAAT);
+        p.sources.clear();
+        p
     }
 
     /// Every non-empty cell a view's grid draws, row by row.
@@ -2733,7 +2843,7 @@ src/a.rs,2
     /// 0, "" or a skip. Evidence: `context/test-evidence/formula-and-io.md`.
     #[test]
     fn files_with_different_headers_import_as_one_sparse_model() {
-        let p = ImportPipeline::new(code_maat_records());
+        let p = code_maat_without_views();
         let kinds: Vec<(&str, FieldKind)> = p
             .proposals
             .iter()
@@ -2775,7 +2885,7 @@ src/a.rs,2
     /// "". Evidence: `context/test-evidence/formula-and-io.md`.
     #[test]
     fn a_label_column_a_record_lacks_is_no_coordinate_not_an_empty_item() {
-        let mut p = ImportPipeline::new(code_maat_records());
+        let mut p = code_maat_without_views();
         p.proposals
             .iter_mut()
             .find(|f| f.field == "coupled")
@@ -3380,6 +3490,110 @@ src/a.rs,2
             w.message
         );
     }
+
+    /// improvise-8jk6.5. Evidence: `context/test-evidence/formula-and-io.md`.
+    #[test]
+    fn a_formula_that_does_not_parse_stops_the_build_and_is_named() {
+        let mut p = ImportPipeline::new(json!([{"region": "East", "revenue": 1.0}]));
+        p.formulas = vec![
+            "Good = revenue".into(),
+            "Bad = revenue +".into(),
+            "Worse = * 2".into(),
+        ];
+        let err = p.build_model().expect_err("an unparseable formula refuses");
+        let msg = err.to_string();
+        assert!(msg.contains("Formula error in 'Bad = revenue +'"), "{msg}");
+        assert!(msg.contains("Formula error in 'Worse = * 2'"), "{msg}");
+        assert!(!msg.contains("Good"), "{msg}");
+    }
+
+    /// improvise-8jk6.5: the wizard refuses where the text can still be
+    /// fixed. Evidence: `context/test-evidence/formula-and-io.md`.
+    #[test]
+    fn confirming_a_formula_that_does_not_parse_keeps_it_out_and_the_text_in() {
+        let mut w = sample_wizard();
+        w.start_formula_edit();
+        for c in "Bad = revenue +".chars() {
+            w.push_formula_char(c);
+        }
+        w.confirm_formula();
+        assert!(w.pipeline.formulas.is_empty(), "{:?}", w.pipeline.formulas);
+        assert_eq!(w.formula_buffer, "Bad = revenue +");
+        assert!(w.formula_editing);
+        let message = w.message.clone().unwrap_or_default();
+        assert!(
+            message.starts_with("Formula error in 'Bad = revenue +'"),
+            "{message}"
+        );
+    }
+
+    /// improvise-8jk6.4: the role reads the text. Evidence:
+    /// `context/test-evidence/formula-and-io.md`.
+    #[test]
+    fn a_csv_column_read_as_a_category_keeps_its_text_and_as_a_measure_its_number() {
+        use crate::import::csv_parser::parse_csv_bytes_with_warnings;
+        use crate::model::cell::{CellKey, CellValue};
+        let csv = b"commit,lines\n00565796,3\n220944e2,5\n";
+        let (records, _) = parse_csv_bytes_with_warnings(csv).unwrap();
+        let mut p = ImportPipeline::from_csv(records);
+        let kind = |p: &ImportPipeline, f: &str| {
+            p.proposals
+                .iter()
+                .find(|x| x.field == f)
+                .unwrap()
+                .kind
+                .clone()
+        };
+        assert_eq!(kind(&p, "commit"), FieldKind::Measure, "analysis types");
+        let commit = p.proposals.iter_mut().find(|x| x.field == "commit");
+        commit.unwrap().kind = FieldKind::Category;
+        let wb = p.build_model().unwrap();
+        let items: Vec<&String> = wb.model.category("commit").unwrap().items.keys().collect();
+        assert_eq!(items, ["00565796", "220944e2"]);
+        let key = CellKey::new(vec![
+            ("_Measure".into(), "lines".into()),
+            ("commit".into(), "220944e2".into()),
+        ]);
+        assert_eq!(wb.model.get_cell(&key), Some(&CellValue::Number(5.0)));
+    }
+
+    /// improvise-8jk6.6 meets lpzu's header-only rule: a folder of nothing but
+    /// headers kept no record. Evidence: `context/test-evidence/formula-and-io.md`.
+    #[test]
+    fn a_folder_of_only_header_only_files_keeps_no_record_and_refuses() {
+        let p = merged_pipeline(&[("m/a.csv", "entity,x\n"), ("m/b.csv", "entity,y\n")]);
+        assert!(p.proposals.iter().any(|f| f.field == "File"), "setup");
+        let msg = p.build_model().expect_err("refuses").to_string();
+        assert_eq!(msg, "Nothing to import: the source has no records");
+    }
+
+    /// improvise-8jk6.6: the wizard asks on the review step, where another
+    /// path can still be chosen. Evidence: the test-evidence doc.
+    #[test]
+    fn the_review_step_says_when_no_record_can_be_imported() {
+        let mut w = ImportWizard::new(json!([1, 2]));
+        assert_eq!(w.step, WizardStep::ReviewProposals, "setup");
+        w.advance();
+        assert_eq!(w.step, WizardStep::ReviewProposals);
+        assert_eq!(
+            w.message.as_deref(),
+            Some("Nothing to import: all 2 records were skipped as not JSON objects")
+        );
+    }
+
+    /// improvise-8jk6.6. Evidence: `context/test-evidence/formula-and-io.md`.
+    #[test]
+    fn a_build_that_keeps_no_record_refuses_and_says_why() {
+        let skipped = ImportPipeline::new(json!([1, "two", null]));
+        let msg = skipped.build_model().expect_err("refuses").to_string();
+        assert_eq!(
+            msg,
+            "Nothing to import: all 3 records were skipped as not JSON objects"
+        );
+        let empty = ImportPipeline::new(json!([]));
+        let msg = empty.build_model().expect_err("refuses").to_string();
+        assert_eq!(msg, "Nothing to import: the source has no records");
+    }
 }
 
 /// improvise-8jk6.3: a date column imports as ONE hierarchical category.
diff --git a/crates/improvise-io/src/persistence/roundtrip_tests.rs b/crates/improvise-io/src/persistence/roundtrip_tests.rs
index 4c2c7c69..a6c84ceb 100644
--- a/crates/improvise-io/src/persistence/roundtrip_tests.rs
+++ b/crates/improvise-io/src/persistence/roundtrip_tests.rs
@@ -1031,7 +1031,7 @@ fn a_csv_header_with_an_embedded_newline_survives_a_save() {
     let csv = "\"Region\nName\",Amount\nEast,1\nWest,2\n";
     let (records, _) =
         crate::import::csv_parser::parse_csv_bytes_with_warnings(csv.as_bytes()).unwrap();
-    let wb = crate::import::wizard::ImportPipeline::new(serde_json::Value::Array(records))
+    let wb = crate::import::wizard::ImportPipeline::from_csv(records)
         .build_model()
         .unwrap();
     assert!(
diff --git a/src/main.rs b/src/main.rs
index ab94e508..947b3999 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -105,6 +105,10 @@ struct ImportArgs {
     #[arg(long, value_enum, default_value_t = OnDuplicate::Refuse)]
     on_duplicate: OnDuplicate,
 
+    /// With --no-wizard, refuse (and write nothing) if any record is skipped
+    #[arg(long)]
+    strict: bool,
+
     /// Save to file instead of opening TUI
     #[arg(short, long)]
     output: Option<PathBuf>,
@@ -166,6 +170,7 @@ impl Runnable for ImportArgs {
             formulas: self.formula,
             name: self.name,
             on_duplicate: self.on_duplicate,
+            strict: self.strict,
         };
 
         if self.no_wizard {
@@ -201,6 +206,7 @@ struct ImportConfig {
     formulas: Vec<String>,
     name: Option<String>,
     on_duplicate: OnDuplicate,
+    strict: bool,
 }
 
 fn parse_colon_pairs(args: &[String]) -> Vec<(String, String)> {
@@ -305,6 +311,23 @@ fn apply_axis_overrides(wb: &mut Workbook, axes: &[(String, String)]) {
     }
 }
 
+/// Why `--axis` is refused: the categories the import did build, and a date
+/// hint only for a name shaped like one of the split date parts.
+fn axis_refusal(cat: &str, built: &[String]) -> String {
+    let date_part = ["_Year", "_Month", "_Quarter"]
+        .iter()
+        .any(|suffix| cat.ends_with(suffix));
+    let hint = if date_part {
+        " A date's parts are one category unless --split-date names it."
+    } else {
+        ""
+    };
+    format!(
+        "--axis names '{cat}', which this import did not build.{hint} Built: {}",
+        built.join(", ")
+    )
+}
+
 fn run_headless_import(
     mut pipeline: ImportPipeline,
     config: &ImportConfig,
@@ -312,11 +335,16 @@ fn run_headless_import(
     model_file: Option<PathBuf>,
 ) -> Result<()> {
     apply_config_to_pipeline(&mut pipeline, config);
+    // A strict skip refuses before the build, so nothing is written (8jk6.6).
+    if let (true, Some(skipped)) = (config.strict, pipeline.skip_warning()) {
+        anyhow::bail!("--strict: {skipped}");
+    }
     if let Some(field) = config.split_dates.iter().find(|f| {
-        !pipeline
-            .proposals
-            .iter()
-            .any(|p| &p.field == *f && p.kind == import::analyzer::FieldKind::TimeCategory)
+        !pipeline.proposals.iter().any(|p| {
+            &p.field == *f
+                && p.kind == import::analyzer::FieldKind::TimeCategory
+                && p.date_format.is_some()
+        })
     }) {
         anyhow::bail!("--split-date '{field}' is not a date column of this import");
     }
@@ -335,10 +363,10 @@ fn run_headless_import(
         .iter()
         .find(|(cat, _)| wb.model.category(cat).is_none())
     {
-        anyhow::bail!(
-            "--axis names '{cat}', which this import did not build \
-             (a date's parts are one category unless --split-date names it)"
-        );
+        anyhow::bail!(axis_refusal(
+            cat,
+            &wb.model.categories.keys().cloned().collect::<Vec<_>>()
+        ));
     }
     apply_axis_overrides(&mut wb, &config.axes);
 
@@ -416,7 +444,7 @@ fn get_import_data(args: &[PathBuf]) -> Option<ImportPipeline> {
             Ok(content) => {
                 if csv_path_p(path) {
                     match crate::import::csv_parser::parse_csv(path) {
-                        Ok(records) => Some(ImportPipeline::new(Value::Array(records))),
+                        Ok(records) => Some(ImportPipeline::from_csv(records)),
                         Err(e) => {
                             eprintln!("CSV parse error: {e}");
                             None
diff --git a/tests/bug_hunt_reproductions.rs b/tests/bug_hunt_reproductions.rs
index 68021682..2475931d 100644
--- a/tests/bug_hunt_reproductions.rs
+++ b/tests/bug_hunt_reproductions.rs
@@ -64,8 +64,8 @@ fn duplicate_csv_headers_not_silent() {
         2,
         "both columns must survive as separate fields: {fields:?}"
     );
-    assert_eq!(fields["A"].as_f64(), Some(1.0), "{fields:?}");
-    assert_eq!(fields["A_2"].as_f64(), Some(2.0), "{fields:?}");
+    assert_eq!(fields["A"], "1", "{fields:?}");
+    assert_eq!(fields["A_2"], "2", "{fields:?}");
     assert!(
         warnings
             .iter()
@@ -78,7 +78,7 @@ fn duplicate_csv_headers_not_silent() {
 fn blank_measure_still_infers_numeric() {
     use improvise_io::import::{analyzer::FieldKind, wizard::ImportPipeline};
     let (records, _) = parse_csv_bytes_with_warnings(b"Region,Revenue\nEast,10\nWest,\n").unwrap();
-    let p = ImportPipeline::new(serde_json::Value::Array(records));
+    let p = ImportPipeline::from_csv(records);
     let proposal = p.proposals.iter().find(|p| p.field == "Revenue").unwrap();
     assert_eq!(proposal.kind, FieldKind::Measure);
 }
@@ -87,7 +87,7 @@ fn repeated_import_coordinates_not_silently_overwritten() {
     use improvise_io::import::wizard::ImportPipeline;
     let (records, _) =
         parse_csv_bytes_with_warnings(b"Region,Revenue\nEast,10\nEast,20\n").unwrap();
-    let mut pipeline = ImportPipeline::new(serde_json::Value::Array(records));
+    let mut pipeline = ImportPipeline::from_csv(records);
     pipeline.on_duplicate = improvise_io::import::wizard::DuplicatePolicy::Sum;
     let (wb, warnings) = pipeline.build_model_with_warnings().unwrap();
     let values: Vec<f64> = wb
@@ -400,7 +400,7 @@ fn ordinary_controls() {
     let (records, warnings) =
         parse_csv_bytes_with_warnings(b"Region,Revenue\nEast,10\nWest,20\n").unwrap();
     assert!(warnings.is_empty());
-    let p = ImportPipeline::new(serde_json::Value::Array(records));
+    let p = ImportPipeline::from_csv(records);
     assert_eq!(
         p.proposals
             .iter()
diff --git a/tests/headless_cli.rs b/tests/headless_cli.rs
index 572740a2..1e0e11f4 100644
--- a/tests/headless_cli.rs
+++ b/tests/headless_cli.rs
@@ -703,6 +703,77 @@ fn an_axis_naming_a_category_the_import_did_not_build_is_refused() {
     );
 }
 
+/// `--time` on a column with no date format makes it a time category without
+/// dates, so there is nothing to split.
+#[test]
+fn split_date_on_a_time_field_with_no_date_format_is_refused() {
+    let scratch = Scratch::new("split_date_no_format");
+    let out_path = scratch.path("out.improv");
+    let out = run_import(
+        &dated_csv(&scratch),
+        &out_path,
+        &["--time", "Revenue", "--split-date", "Revenue"],
+    );
+    assert!(!out.status.success(), "{}", stderr_of(&out));
+    assert!(
+        stderr_of(&out).contains("--split-date"),
+        "{}",
+        stderr_of(&out)
+    );
+}
+
+/// A name that is no date part gets the categories that exist, never the date
+/// hint (the README's old `--axis Measure:column` was this).
+#[test]
+fn an_absent_axis_category_that_is_no_date_part_names_what_was_built() {
+    let scratch = Scratch::new("axis_unknown_plain");
+    let out_path = scratch.path("out.improv");
+    let out = run_import(
+        &dated_csv(&scratch),
+        &out_path,
+        &["--axis", "Measure:column"],
+    );
+    assert!(!out.status.success(), "{}", stderr_of(&out));
+    let err = stderr_of(&out);
+    assert!(
+        err.contains("'Measure'") && err.contains("_Measure"),
+        "{err}"
+    );
+    assert!(!err.contains("--split-date"), "no date hint: {err}");
+}
+
+/// The README's import recipe runs as printed.
+#[test]
+fn the_readmes_import_recipe_runs() {
+    let root = Path::new(env!("CARGO_MANIFEST_DIR"));
+    let readme = std::fs::read_to_string(root.join("README.md")).unwrap();
+    let start = readme.find("improvise import examples/demo.csv").unwrap();
+    let recipe = readme[start..]
+        .split("```")
+        .next()
+        .unwrap()
+        .replace("\\\n", " ");
+    let mut args: Vec<String> = vec![];
+    for (i, part) in recipe.split('"').enumerate() {
+        match i % 2 {
+            1 => args.push(part.to_string()),
+            _ => args.extend(part.split_whitespace().map(String::from)),
+        }
+    }
+    let at = args.iter().position(|a| a == "-o").unwrap();
+    args.truncate(at);
+    let scratch = Scratch::new("readme_recipe");
+    let out = Command::new(bin())
+        .arg("import")
+        .arg(root.join("examples/demo.csv"))
+        .args(args.iter().skip(3))
+        .arg("-o")
+        .arg(scratch.path("out.improv"))
+        .output()
+        .unwrap();
+    assert!(out.status.success(), "{}", stderr_of(&out));
+}
+
 /// code-maat's reports: three schemas, a stray note, and room for the output.
 fn code_maat_folder(scratch: &Scratch) -> PathBuf {
     let dir = scratch.path("maat");
@@ -799,6 +870,192 @@ fn a_headless_import_takes_files_and_folders_together_and_refuses_an_empty_folde
     assert!(!scratch.path("never.improv").exists());
 }
 
+/// Runs `import <source> --no-wizard <extra> -o out.improv` on a source it writes.
+fn import_of(scratch: &Scratch, name: &str, body: &str, extra: &[&str]) -> (Output, PathBuf) {
+    let source = scratch.path(name);
+    std::fs::write(&source, body).unwrap();
+    let out_path = scratch.path("out.improv");
+    (run_import(&source, &out_path, extra), out_path)
+}
+
+/// improvise-8jk6.4: a forced category's values are item names, so they keep
+/// their source text. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn a_csv_column_forced_to_category_keeps_its_text_verbatim() {
+    let scratch = Scratch::new("forced_category_text");
+    let (out, out_path) = import_of(
+        &scratch,
+        "commits.csv",
+        "commit,lines\n00565796,3\n220944e2,5\n",
+        &["--category", "commit", "--measure", "lines"],
+    );
+    assert!(out.status.success(), "{}", stderr_of(&out));
+    let doc = std::fs::read_to_string(&out_path).unwrap();
+    assert!(
+        doc.lines().any(|l| l == "- |00565796|, |220944e2|"),
+        "the items are the source text:\n{doc}"
+    );
+    for line in [
+        "_Measure=lines, commit=|00565796| = 3",
+        "_Measure=lines, commit=|220944e2| = 5",
+    ] {
+        assert!(doc.lines().any(|l| l == line), "{line} missing from\n{doc}");
+    }
+}
+
+/// improvise-8jk6.4: JSON has no source text for a number, so a forced
+/// category names it as serde_json writes it. Evidence file.
+#[test]
+fn a_json_number_forced_to_category_is_named_as_json_writes_it() {
+    let scratch = Scratch::new("forced_category_json");
+    let (out, out_path) = import_of(
+        &scratch,
+        "commits.json",
+        r#"[{"commit": 42, "n": 1}, {"commit": 1.5, "n": 2}, {"commit": 1e3, "n": 3}]"#,
+        &["--category", "commit", "--measure", "n"],
+    );
+    assert!(out.status.success(), "{}", stderr_of(&out));
+    let doc = std::fs::read_to_string(&out_path).unwrap();
+    assert!(
+        doc.lines().any(|l| l == "- |42|, |1.5|, |1000.0|"),
+        "the items are the numbers' JSON text:\n{doc}"
+    );
+}
+
+/// improvise-8jk6.5: a formula the user named and that does not parse is a
+/// refusal, not a silent drop. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn an_import_formula_that_does_not_parse_refuses_names_it_and_writes_nothing() {
+    let scratch = Scratch::new("bad_formula");
+    let (out, out_path) = import_of(
+        &scratch,
+        "f.csv",
+        "a,x\nk,1\n",
+        &[
+            "--measure",
+            "x",
+            "--formula",
+            "Y = x +",
+            "--formula",
+            "Z = x",
+            "--formula",
+            "W = * 2",
+        ],
+    );
+    let err = stderr_of(&out);
+    assert!(!out.status.success(), "a dropped formula exited 0: {err}");
+    assert!(err.contains("'Y = x +'"), "names the formula: {err}");
+    assert!(err.contains("'W = * 2'"), "names every bad one: {err}");
+    assert!(err.contains("expected"), "carries the parse error: {err}");
+    assert!(!err.contains("'Z = x'"), "names only the bad one: {err}");
+    assert!(!out_path.exists(), "nothing is written");
+}
+
+/// improvise-8jk6.6: an import that kept no record says why, and exits
+/// non-zero writing nothing. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn an_import_that_keeps_no_record_refuses_and_says_why() {
+    let scratch = Scratch::new("kept_none");
+    let (out, out_path) = import_of(&scratch, "r.json", r#"[1, 2, "three"]"#, &[]);
+    let err = stderr_of(&out);
+    assert!(!out.status.success(), "{err}");
+    assert!(
+        err.contains("all 3 records were skipped as not JSON objects"),
+        "says why: {err}"
+    );
+    assert!(!out_path.exists(), "nothing is written");
+}
+
+/// improvise-8jk6.6: a partial skip warns with a count and exits 0, a choice
+/// put to the user. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn an_import_that_skips_some_records_warns_with_a_count_and_still_writes() {
+    let scratch = Scratch::new("kept_some");
+    let (out, out_path) = import_of(
+        &scratch,
+        "p.json",
+        r#"[{"a": "k", "x": 1}, 2, {"a": "m", "x": 4}]"#,
+        &[],
+    );
+    let err = stderr_of(&out);
+    assert!(out.status.success(), "{err}");
+    assert!(
+        err.contains("warning: Skipped 1 of 3 records that were not JSON objects"),
+        "{err}"
+    );
+    let doc = std::fs::read_to_string(&out_path).unwrap();
+    assert!(doc.contains("_Measure=x, a=m = 4"), "{doc}");
+}
+
+/// A JSON array with one stray scalar among its records.
+const ONE_STRAY: &str = r#"[{"a": "k", "x": 1}, 2, {"a": "m", "x": 4}]"#;
+
+/// improvise-8jk6.6, the user's ruling: `--strict` makes any skip a refusal
+/// that writes nothing. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn a_strict_import_that_skips_a_record_refuses_and_writes_nothing() {
+    let scratch = Scratch::new("strict_skip");
+    let (out, out_path) = import_of(&scratch, "p.json", ONE_STRAY, &["--strict"]);
+    let err = stderr_of(&out);
+    assert!(!out.status.success(), "{err}");
+    assert!(
+        err.contains("--strict: Skipped 1 of 3 records that were not JSON objects"),
+        "{err}"
+    );
+    assert!(!out_path.exists(), "a strict refusal writes nothing");
+}
+
+/// The control: `--strict` refuses a skip, not every import.
+#[test]
+fn a_strict_import_that_skips_nothing_still_writes() {
+    let scratch = Scratch::new("strict_clean");
+    let body = r#"[{"a": "k", "x": 1}, {"a": "m", "x": 4}]"#;
+    let (out, out_path) = import_of(&scratch, "p.json", body, &["--strict"]);
+    assert!(out.status.success(), "{}", stderr_of(&out));
+    let doc = std::fs::read_to_string(&out_path).unwrap();
+    assert!(doc.contains("_Measure=x, a=m = 4"), "{doc}");
+}
+
+/// improvise-8jk6.17 (F3): a CSV decimal measure is its number, not dropped
+/// text. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn a_csv_decimal_measure_imports_as_its_number() {
+    let scratch = Scratch::new("csv_decimal");
+    let body = "Region,Revenue\nEast,1000.50\nWest,-2.25\n";
+    let (out, out_path) = import_of(&scratch, "f.csv", body, &[]);
+    assert!(out.status.success(), "{}", stderr_of(&out));
+    let doc = std::fs::read_to_string(&out_path).unwrap();
+    for line in [
+        "Region=East, _Measure=Revenue = 1000.5",
+        "Region=West, _Measure=Revenue = -2.25",
+    ] {
+        assert!(doc.lines().any(|l| l == line), "{line} missing from\n{doc}");
+    }
+}
+
+/// improvise-8jk6.17 (F1): `import-json` reads a CSV as CSV, so its numeric
+/// column is a measure. Evidence: `context/test-evidence/cli.md`.
+#[test]
+fn import_json_on_a_csv_reads_its_numeric_column_as_a_measure() {
+    let scratch = Scratch::new("import_json_csv");
+    let csv = scratch.path("f.csv");
+    std::fs::write(&csv, "Region,Revenue\nEast,1000.50\nWest,7\n").unwrap();
+    let doc = scratch.path("m.improv");
+    let out = run_cmd(&[&format!("import-json {}", csv.display())], &doc);
+    assert!(out.status.success(), "{}", stderr_of(&out));
+    let text = std::fs::read_to_string(&doc).unwrap();
+    assert!(!text.contains("## Category: Revenue"), "{text}");
+    for line in [
+        "Region=East, _Measure=Revenue = 1000.5",
+        "Region=West, _Measure=Revenue = 7",
+    ] {
+        assert!(
+            text.lines().any(|l| l == line),
+            "{line} missing from\n{text}"
+        );
+    }
+}
+
 /// improvise-lpzu, the user's ruling: a folder always merges, so even one CSV
 /// keeps its file identity. Evidence: `context/test-evidence/cli.md`.
 #[test]
diff --git a/tests/records_wide_smoke.rs b/tests/records_wide_smoke.rs
index c121b25a..38cfbbe3 100644
--- a/tests/records_wide_smoke.rs
+++ b/tests/records_wide_smoke.rs
@@ -18,7 +18,7 @@ fn imported_model_shows_one_row_per_record_with_measure_columns() {
     .expect("write csv");
 
     let records = improvise::import::csv_parser::parse_csv(&csv).expect("csv parses");
-    let pipeline = ImportPipeline::new(serde_json::Value::Array(records));
+    let pipeline = ImportPipeline::from_csv(records);
     let workbook = pipeline.build_model().expect("model builds");
 
     let saved = dir.path().join("w.improv");