Move deferred and out-of-scope items to todo.md

AGENTS.md

@@ -8,11 +8,3 @@
 - A standing choice a reader would relitigate goes under Decisions in the README, not in a comment.
 - A README example is a `json` block that loads and renders as a category; `readme_test.go` runs every one.
 - Cyclomatic complexity is gated at 14: the table-shaped dispatches (`eachToken`, `calc.factor`, `walkPath`) sit at 13–14 and stay whole; anything else that reaches 14 is decomposed.
-
-# Deferred
-
-- **Shipped-data de-duplication (2026-09-02).** `email.json`'s `local` is a
-  drifted copy of `username.json`, and no shipped file yet uses a held path, an
-  operand or a reference. Fixed in the data fill before the first tag, when the
-  shipped set is rewritten anyway; premise: nothing depends on the shipped data's
-  shape until then. Not raised in review before that.

README.md

@@ -41,6 +41,8 @@ brace, a bracket or a quote, so the two cannot collide (see
 | `-s`, `--seed N` | reproducible output |
 | `-n`, `--repeat N` | render the value N times (up to 1048576), each an independent draw, streamed |
 | `--separator S` | between repeated values (default a newline) |
+| `--format F` | `text` (default), `json`, `ndjson`, `csv` or `sql` — a record's columns, one record per row (json frames them as an array) |
+| `--table T` | the INSERT target for `--format sql` (default: the path's last segment, or `records` for an inline template) |
 | `--list` | print every path, then exit |
 | `--version`, `-h`, `--help` | print, then exit |
 
@@ -73,6 +75,67 @@ fejkdata --seed 1 --data-path ./mydata sql
 fejkdata --repeat 100 --data-path ./mydata sql > seed.sql
 ```
 
+That `format` string is the free-form spelling — you hand-write the whole row.
+For structured output a record writes the row for you.
+
+### Records
+
+A record is a template seen as columns: its fields are the columns, its `format`
+the whole. `--format json|ndjson|csv|sql` writes the records; the library's
+`Record` (below) hands back the columns. Every column is a string — typed scalars
+are on the release checklist, see [`todo.md`](todo.md). Save
+`mydata/users.json`:
+
+```json
+{
+  "format": "{first} {last}",
+  "first": ["Ada", "Bo"],
+  "last": ["Lovelace", "Ek"]
+}
+```
+
+```sh
+fejkdata --seed 1 --data-path ./mydata --format json users            # [{"first":"Bo","last":"Lovelace"}]
+fejkdata --seed 1 --data-path ./mydata --format ndjson users          # {"first":"Bo","last":"Lovelace"}
+fejkdata --seed 1 --data-path ./mydata --format csv  users            # first,last  →  Bo,Lovelace
+fejkdata --seed 1 --data-path ./mydata --format sql  users            # INSERT INTO "users" ("first", "last") VALUES ('Bo', 'Lovelace');
+fejkdata --seed 1 --data-path ./mydata --format sql --table people users # INSERT into another table
+```
+
+`--repeat` streams that many records — `json` frames them as one array document,
+`ndjson` writes one object per line, `csv` a row after a header, `sql` one INSERT
+per line. Only a category-level template is a record; a field, choice or folder
+errors, and so does a `repeat` on
+the template itself, which composes the format into one string rather than
+projecting columns — ask for more records with `--repeat`. A `repeat` on a column
+is fine.
+
+A column carrying a newline keeps it inside the quoted CSV field or the SQL string
+literal, so a row can span physical lines: read the stream with a CSV or SQL
+parser rather than splitting it on newlines.
+
+The SQL is ANSI — identifiers in double quotes, a literal quote doubled (`''`),
+backslashes passed through — so a hyphenated field like `postal-code` stays a
+valid identifier. PostgreSQL and SQLite take it as written; MySQL and MariaDB need
+`ANSI_QUOTES` and `NO_BACKSLASH_ESCAPES` set first, or they read `"users"` as a
+string and a backslash as an escape.
+
+A record written only to emit columns still needs a `format` — the grammar's one
+required key — so `"format": ""` carries the fields with an inert format: it
+renders nothing by `Fake`, and is compiled only so the tree's fences still run.
+The columns are the point, and their facts stay together: two columns that read a
+path into one category — `{/currency.code}` and `{/currency.symbol}` — share one
+draw of it, so the record is internally consistent. A bare `{/currency}` names no
+field, so it keeps drawing on its own. That one draw is also why two
+columns may not read overlapping reference *paths* — `{/cat.a}` beside
+`{/cat.a.b}` is refused, naming the fields to write instead, as
+[One draw, one spelling](#one-draw-one-spelling) refuses that pair inside a single
+format. A column may not reference the record it belongs to by any spelling: `{/users.first}`
+or a bare `{/users}` inside `users` describes a draw other than the columns beside
+it, so put a value two columns share in its own category and reference that. A field hold, transform or
+operand ties fields together within one column as always (see
+[Correlated fields](#correlated-fields) and [Decisions](#decisions)).
+
 ## Library
 
 ```sh
@@ -89,6 +152,9 @@ paths := f.List()                  // every path Fake accepts, sorted
 v, err = f.FakeTemplate("name: {/sv_SE.person.last}")      // compile + render in one call
 t, err := f.NewTemplate(`{"format":"name: {x}","x":["bosse","lina"]}`) // compile once
 v = t.Fake()                                              // render many times, no re-parse
+r, err := f.Record("users")               // one record: each field a column
+s := r.JSON()                             // {"first":"Ada","last":"Lovelace"}
+r, err = f.FakeRecord(`{"format":"{x}","x":["a","b"]}`) // compile + render inline
 ```
 
 | Option | |
@@ -98,6 +164,11 @@ v = t.Fake()                                              // render many times,
 | `WithDataFS(fsys)` | layer an `fs.FS`, such as your own `embed.FS` |
 | `WithoutShippedData()` | load only what you give |
 
+A `*Record` carries its columns via `Columns()`, and serializes them with `JSON()`
+(one object), `CSVHeader()`/`CSVLine()`, or `SQLInsert(table)` — the shapes the
+CLI's `--format` writes. `Record` and `FakeRecord` take a record; a path or
+template that is not one — a bare string, a choice, or a folder — errors.
+
 A `*Generator` is safe for concurrent use; a seeded sequence is reproducible only
 when drawn from one goroutine. Changing how a value is composed shifts the seeded
 stream for that value and everything drawn after it.
@@ -454,6 +525,38 @@ tokens add cost in proportion to the output.
 - **Samples say what they emit, transforms what they do.** `{upper(2)}` is two
   letters, `{uppercase(x)}` is `x` upper-cased; one name for both would turn on
   whether the argument looks like a number.
+- **A record is a template seen as columns, not a second schema format.** A
+  template's `format` composes its fields into one string; `Record` and
+  `--format` project the same fields as columns. Two views of one dataset, so a
+  record author writes the same JSON they already know, and a column is the same
+  field `Fake` renders by dotted path. The `format` is inert to a record — a
+  record-only template writes `"format": ""` — but it is compiled and fenced, so
+  a template that loads renders as whichever shape is asked for.
+- **A record shares one reference draw per category.** Two columns that reference
+  one category — `{/currency.code}` beside `{/currency.symbol}` — read one draw of
+  it, so a record's facts agree the way a template's [correlated
+  fields](#correlated-fields) do. The draw is one per record, so it spans a
+  column's `repeat` and nested templates too (one record is one coherent unit);
+  a bare reference — `{/currency}`, no field — stays an independent draw every
+  time, the rule a format string already follows. Only references share: a sibling
+  field is local to its own column, so a `first` column does not silently bind to
+  a `first` in the column next to it.
+
+  The string view of that same template does not share. `Fake` renders each
+  sibling field as its own expansion, so a `{/currency.code}` field beside a
+  `{/currency.symbol}` field is two draws and may render `EUR $`; writing both
+  references in one `format` holds them together, as
+  [One draw, one spelling](#one-draw-one-spelling) says. The scope is what makes a
+  row coherent when the columns *are* the output, and there the caller cannot fall
+  back on one format string. Widening it to every render would change what `Fake`
+  has emitted since the start, for a correlation a single format already reaches.
+- **A record's column set is fixed before the first draw.** Only a category-level
+  template is a record: a path descending into a field, or naming a folder or a
+  choice, errors. A tail may pass through a choice whose variants carry different
+  fields, so the columns — and with them the CSV header written once ahead of every
+  row — would vary per draw. A fixed column set is what the CSV and `INSERT`
+  contracts rest on, so the restriction holds even where a particular choice would
+  happen to agree.
 - **The performance gate asserts allocations, not wall-clock time.** `AllocsPerRun`
   is deterministic across machines, so a ±10% ceiling does not flake under CI load,
   while time varies with the machine and its neighbours. A rendering slowdown
@@ -500,6 +603,7 @@ fejkdata.go     Generator, New, options, the embedded data set, List
 node.go         the node model and JSON -> node compilation
 path.go         the dotted-path walk, and proving a path resolves
 render.go       Fake and the recursive renderer (choices, format strings, expansions)
+record.go       records: Record, the JSON/CSV/SQL serializers, and their entry points
 inline.go       inline templates: Template, NewTemplate, FakeTemplate, and their compile and link
 template.go     the {token} grammar: scanning, tokens, operands, validation, compiling a format
 hold.go         the hold: one draw per expansion for paths and operands, and its fences

cmd/fejkdata/main.go

@@ -15,6 +15,7 @@ import (
 	"io"
 	"os"
 	"runtime/debug"
+	"sort"
 	"strconv"
 	"strings"
 
@@ -34,13 +35,19 @@ or a quote). Templates reach the data by reference from the root —
 argument carrying a bracket, a closing brace or a quote but no valid JSON names
 neither.
 
+With --format json, ndjson, csv or sql the argument must name a record — a
+template whose fields are its columns — and the rows are written as one JSON
+array, one JSON object per line, one CSV row (after a header), or one INSERT.
+
   -d, --data-path D      a data directory to layer over the shipped data (repeatable; last wins on a clash)
+      --format F         output form: text (default), json, ndjson, csv or sql
   -h, --help             print this help, then exit
       --list             list the paths the data offers, then exit
       --no-shipped-data  load only the --data-path directories
   -n, --repeat N         render the value N times, 1..1048576 (default 1)
   -s, --seed N           seed for reproducible output
       --separator S      string between repeated values (default newline)
+      --table T          the INSERT target for --format sql (default: the path's last segment, or records for an inline template)
       --version          print the version, then exit
 
 Flags may come before or after <path|template>; -- ends the flags. A short flag's
@@ -49,6 +56,8 @@ value attaches or follows (-n3, -n 3); short flags bundle (-hn 3).
 
 type invocation struct {
 	dirs         []string
+	format       string
+	formatSet    bool
 	help         bool
 	list         bool
 	noShipped    bool
@@ -59,6 +68,8 @@ type invocation struct {
 	seeded       bool
 	separator    string
 	separatorSet bool
+	table        string
+	tableSet     bool
 	version      bool
 }
 
@@ -71,6 +82,7 @@ type flagDef struct {
 
 var flagDefs = []flagDef{
 	{"data-path", "d", true, func(in *invocation, v string) error { in.dirs = append(in.dirs, v); return nil }},
+	{"format", "", true, func(in *invocation, v string) error { in.format, in.formatSet = v, true; return nil }},
 	{"help", "h", false, func(in *invocation, _ string) error { in.help = true; return nil }},
 	{"list", "", false, func(in *invocation, _ string) error { in.list = true; return nil }},
 	{"no-shipped-data", "", false, func(in *invocation, _ string) error { in.noShipped = true; return nil }},
@@ -91,6 +103,7 @@ var flagDefs = []flagDef{
 		return nil
 	}},
 	{"separator", "", true, func(in *invocation, v string) error { in.separator, in.separatorSet = v, true; return nil }},
+	{"table", "", true, func(in *invocation, v string) error { in.table, in.tableSet = v, true; return nil }},
 	{"version", "", false, func(in *invocation, _ string) error { in.version = true; return nil }},
 }
 
@@ -158,7 +171,7 @@ func splitFlags(arg string) ([]flagArg, error) {
 }
 
 func parseArgs(argv []string) (invocation, error) {
-	in := invocation{repeat: 1, separator: "\n"}
+	in := invocation{repeat: 1, separator: "\n", format: "text"}
 	for i := 0; i < len(argv); i++ {
 		arg := argv[i]
 		if arg == "--" {
@@ -196,14 +209,71 @@ func parseArgs(argv []string) (invocation, error) {
 	return in, nil
 }
 
+// recordFormat is one way to write a record out: the line each record renders,
+// the header that precedes the first one, and the open/close frame plus the
+// between-record separator a document form needs.
+type recordFormat struct {
+	header func(*fejkdata.Record) string
+	line   func(r *fejkdata.Record, table string) string
+	open   string
+	close  string
+	sep    string
+}
+
+// recordFormats is every --format that writes records. json frames the records
+// as one array document; ndjson is the same column, one object per line.
+var recordFormats = map[string]recordFormat{
+	"csv":    {header: (*fejkdata.Record).CSVHeader, line: func(r *fejkdata.Record, _ string) string { return r.CSVLine() }, sep: "\n"},
+	"json":   {line: jsonLine, open: "[", close: "]", sep: ",\n"},
+	"ndjson": {line: jsonLine, sep: "\n"},
+	"sql":    {line: func(r *fejkdata.Record, table string) string { return r.SQLInsert(table) }, sep: "\n"},
+}
+
+func jsonLine(r *fejkdata.Record, _ string) string { return r.JSON() }
+
+// writesRecords reports whether the format writes records rather than plain text.
+func (in invocation) writesRecords() bool {
+	return in.format != "text"
+}
+
+// formatNames lists the --format values, text included, for the misuse error.
+func formatNames() string {
+	names := []string{"text"}
+	for name := range recordFormats {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	return strings.Join(names[:len(names)-1], ", ") + " or " + names[len(names)-1]
+}
+
+// checkFlags rejects a flag value or combination that cannot run.
+func (in invocation) checkFlags() error {
+	if _, ok := recordFormats[in.format]; !ok && in.format != "text" {
+		return fmt.Errorf("--format takes %s, got %q", formatNames(), in.format)
+	}
+	if in.tableSet && in.table == "" {
+		return errors.New("--table names the INSERT target, so it cannot be empty")
+	}
+	if in.tableSet && in.format != "sql" {
+		return errors.New("--table names the INSERT target, so it needs --format sql")
+	}
+	if in.writesRecords() && in.separatorSet {
+		return errors.New("--separator joins text values, so it has no effect with --format " + in.format)
+	}
+	return nil
+}
+
 // check rejects a flag combination or an argument that cannot run, and reports
 // what the argument names, so its shape is settled before any data is read.
 func (in invocation) check() (argKind, error) {
+	if err := in.checkFlags(); err != nil {
+		return argPath, err
+	}
 	if in.list && len(in.paths) > 0 {
 		return argPath, errors.New("--list takes no path")
 	}
-	if in.list && (in.repeatSet || in.separatorSet) {
-		return argPath, errors.New("--list takes no --repeat or --separator")
+	if in.list && (in.repeatSet || in.separatorSet || in.formatSet || in.tableSet) {
+		return argPath, errors.New("--list takes no --repeat, --separator, --format or --table")
 	}
 	if in.list {
 		return argPath, nil
@@ -233,14 +303,12 @@ func (in invocation) options() []fejkdata.Option {
 // so a render failure comes before anything is written; a write failure surfaces
 // from Flush, bufio keeping the first one.
 func (in invocation) write(f *fejkdata.Generator, kind argKind, w io.Writer) error {
-	arg := in.paths[0]
-	draw := func() (string, error) { return f.Fake(arg) }
-	if kind == argTemplate {
-		t, err := f.NewTemplate(arg)
-		if err != nil {
-			return templateError{err}
-		}
-		draw = func() (string, error) { return t.Fake(), nil }
+	if in.writesRecords() {
+		return in.writeRecords(f, kind, w)
+	}
+	draw, err := in.textDraw(f, kind, in.paths[0])
+	if err != nil {
+		return err
 	}
 	out := bufio.NewWriter(w)
 	for i := 0; i < in.repeat; i++ {
@@ -257,6 +325,82 @@ func (in invocation) write(f *fejkdata.Generator, kind argKind, w io.Writer) err
 	return out.Flush()
 }
 
+// textDraw builds what one text render yields: the value, from a path or an
+// inline template.
+func (in invocation) textDraw(f *fejkdata.Generator, kind argKind, arg string) (func() (string, error), error) {
+	if kind != argTemplate {
+		return func() (string, error) { return f.Fake(arg) }, nil
+	}
+	t, err := f.NewTemplate(arg)
+	if err != nil {
+		return nil, templateError{err}
+	}
+	return func() (string, error) { return t.Fake(), nil }, nil
+}
+
+// recordStream builds the record drawer for the argument, plus the INSERT table
+// a sql format names.
+func (in invocation) recordStream(f *fejkdata.Generator, kind argKind, arg string) (func() (*fejkdata.Record, error), string, error) {
+	record := func() (*fejkdata.Record, error) { return f.Record(arg) }
+	if kind == argTemplate {
+		t, err := f.NewRecordTemplate(arg)
+		if err != nil {
+			return nil, "", templateError{err}
+		}
+		record = func() (*fejkdata.Record, error) { return t.Fake(), nil }
+	}
+	table := in.table
+	if table == "" {
+		table = defaultTable(arg, kind)
+	}
+	return record, table, nil
+}
+
+// writeRecords streams a record per line in the chosen format, framing a document
+// form with its open/close brackets and a header preceding the first record.
+func (in invocation) writeRecords(f *fejkdata.Generator, kind argKind, w io.Writer) error {
+	record, table, err := in.recordStream(f, kind, in.paths[0])
+	if err != nil {
+		return err
+	}
+	format := recordFormats[in.format]
+	out := bufio.NewWriter(w)
+	if format.open != "" {
+		out.WriteString(format.open)
+		out.WriteByte('\n')
+	}
+	for i := 0; i < in.repeat; i++ {
+		r, err := record()
+		if err != nil {
+			return err
+		}
+		if i == 0 && format.header != nil {
+			out.WriteString(format.header(r))
+			out.WriteByte('\n')
+		}
+		if i > 0 {
+			out.WriteString(format.sep)
+		}
+		out.WriteString(format.line(r, table))
+	}
+	if format.close != "" {
+		out.WriteByte('\n')
+		out.WriteString(format.close)
+	}
+	out.WriteByte('\n')
+	return out.Flush()
+}
+
+// defaultTable names the INSERT target when --table is absent: the path's last
+// segment, or "records" for an inline template that sits in no folder.
+func defaultTable(arg string, kind argKind) string {
+	if kind == argTemplate {
+		return "records"
+	}
+	segments := strings.Split(arg, ".")
+	return segments[len(segments)-1]
+}
+
 // templateError marks a render failure that is the argument's own fault — an
 // inline template that does not compile. run reports it as misuse (exit 2, with a
 // pointer to --help), unlike an unknown path, which is a runtime error (exit 1).

cmd/fejkdata/main_test.go

@@ -2,6 +2,8 @@ package main
 
 import (
 	"bytes"
+	"encoding/csv"
+	"encoding/json"
 	"os"
 	"path/filepath"
 	"regexp"
@@ -450,6 +452,145 @@ func TestRunEmptyDataPathIsNamed(t *testing.T) {
 	}
 }
 
+func recordDir(t *testing.T) string {
+	t.Helper()
+	dir := t.TempDir()
+	content := `{"format":"{first} {last}","first":["Ada","Bo"],"last":["Lovelace","Ek"]}`
+	if err := os.WriteFile(filepath.Join(dir, "users.json"), []byte(content), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	return dir
+}
+
+func TestRunRecordJSON(t *testing.T) {
+	code, out, errb := runOut("--seed", "1", "--format", "json", "--repeat", "2", "--data-path", recordDir(t), "users")
+	if code != 0 {
+		t.Fatalf("run = %d, stderr=%q", code, errb)
+	}
+	var arr []map[string]string
+	if err := json.Unmarshal([]byte(out), &arr); err != nil {
+		t.Fatalf("json output is not one JSON array: %v\n%q", err, out)
+	}
+	if len(arr) != 2 || len(arr[0]) != 2 || arr[0]["first"] == "" || arr[0]["last"] == "" {
+		t.Fatalf("json output = %q, want an array of two records with first and last columns", out)
+	}
+}
+
+func TestRunRecordNDJSON(t *testing.T) {
+	code, out, errb := runOut("--seed", "1", "--format", "ndjson", "--repeat", "2", "--data-path", recordDir(t), "users")
+	if code != 0 {
+		t.Fatalf("run = %d, stderr=%q", code, errb)
+	}
+	lines := strings.Split(strings.TrimRight(out, "\n"), "\n")
+	if len(lines) != 2 {
+		t.Fatalf("ndjson output = %d lines %q, want one object per line", len(lines), out)
+	}
+	for _, line := range lines {
+		var m map[string]string
+		if err := json.Unmarshal([]byte(line), &m); err != nil || len(m) != 2 {
+			t.Fatalf("ndjson line %q is not one object: %v", line, err)
+		}
+	}
+}
+
+func TestRunRecordCSVRoundTrips(t *testing.T) {
+	code, out, errb := runOut("--seed", "1", "--format", "csv", "--repeat", "3", "--data-path", recordDir(t), "users")
+	if code != 0 {
+		t.Fatalf("run = %d, stderr=%q", code, errb)
+	}
+	lines := strings.Split(strings.TrimRight(out, "\n"), "\n")
+	if len(lines) != 4 {
+		t.Fatalf("csv output = %d lines %q, want a header and 3 rows", len(lines), out)
+	}
+	if lines[0] != "first,last" {
+		t.Fatalf("csv header = %q, want first,last", lines[0])
+	}
+	r, err := csv.NewReader(strings.NewReader(out)).ReadAll()
+	if err != nil {
+		t.Fatalf("csv output is not valid CSV: %v", err)
+	}
+	if len(r) != 4 || len(r[0]) != 2 {
+		t.Fatalf("csv parsed to %v, want 4 records of 2 fields", r)
+	}
+}
+
+func TestRunRecordSQL(t *testing.T) {
+	code, out, errb := runOut("--seed", "1", "--format", "sql", "--repeat", "2", "--table", "people", "--data-path", recordDir(t), "users")
+	if code != 0 {
+		t.Fatalf("run = %d, stderr=%q", code, errb)
+	}
+	lines := strings.Split(strings.TrimRight(out, "\n"), "\n")
+	if len(lines) != 2 || !strings.HasPrefix(lines[0], `INSERT INTO "people" ("first", "last") VALUES (`) {
+		t.Fatalf("sql output = %q, want two INSERT INTO people statements", out)
+	}
+}
+
+func TestRunRecordDefaultTable(t *testing.T) {
+	code, out, errb := runOut("--seed", "1", "--format", "sql", "--data-path", recordDir(t), "users")
+	if code != 0 || !strings.HasPrefix(out, `INSERT INTO "users" (`) {
+		t.Fatalf("default table = %d, %q, want the path's last segment; stderr %q", code, out, errb)
+	}
+}
+
+func TestRunRecordInlineTemplate(t *testing.T) {
+	code, out, errb := runOut("--seed", "1", "--format", "json", `{"format":"{x}","x":["a","b"]}`)
+	if code != 0 {
+		t.Fatalf("inline record = %d, stderr=%q", code, errb)
+	}
+	var arr []map[string]string
+	if err := json.Unmarshal([]byte(out), &arr); err != nil || len(arr) != 1 || (arr[0]["x"] != "a" && arr[0]["x"] != "b") {
+		t.Fatalf("inline record json = %q, want one record with a column x (err %v)", out, err)
+	}
+}
+
+func TestRunRecordMisuse(t *testing.T) {
+	for _, c := range []struct {
+		args []string
+		want string
+	}{
+		{[]string{"--format", "yaml", "users"}, "--format takes csv, json, ndjson, sql or text"},
+		{[]string{"--format", "json", "--separator", ",", "users"}, "--separator joins text values"},
+		{[]string{"--table", "t", "users"}, "--table names the INSERT target"},
+		{[]string{"--format", "json", "--table", "t", "users"}, "--table names the INSERT target"},
+		{[]string{"--format", "sql", "--table", "", "users"}, "--table names the INSERT target"},
+	} {
+		code, out, errb := runOut(c.args...)
+		if code != 2 || out != "" || !strings.Contains(errb, c.want) {
+			t.Errorf("run(%v) = %d, %q, %q; want misuse naming %q", c.args, code, out, errb, c.want)
+		}
+	}
+}
+
+func TestRunRecordOnABareValueIsRuntimeError(t *testing.T) {
+	dir := t.TempDir()
+	if err := os.WriteFile(filepath.Join(dir, "word.json"), []byte(`["a","b"]`), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	code, _, errb := runOut("--format", "json", "--data-path", dir, "word")
+	if code != 1 {
+		t.Fatalf("record on a choice = %d, want exit 1 (runtime error), stderr %q", code, errb)
+	}
+}
+
+func TestRunRecordRejectsATopLevelRepeat(t *testing.T) {
+	dir := t.TempDir()
+	content := `{"format":"{a}-","repeat":3,"separator":"|","a":["x","y"]}`
+	if err := os.WriteFile(filepath.Join(dir, "rep.json"), []byte(content), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	code, _, errb := runOut("--format", "json", "--data-path", dir, "rep")
+	if code != 1 || !strings.Contains(errb, "carries repeat 3") {
+		t.Errorf("record on a repeating path = %d, %q; want exit 1 naming the repeat", code, errb)
+	}
+	code, _, errb = runOut("--format", "json", content)
+	if code != 2 || !strings.Contains(errb, "carries repeat 3") {
+		t.Errorf("record on a repeating inline template = %d, %q; want exit 2 naming the repeat", code, errb)
+	}
+	if code, out, _ := runOut("--seed", "1", "--data-path", dir, "rep"); code != 0 || strings.TrimSpace(out) != "x-|x-|x-" {
+		t.Errorf("text view = %d, %q; want the repeat still composed", code, out)
+	}
+}
+
 func TestRunNumbersFollowTheShell(t *testing.T) {
 	_, want, _ := runOut("--seed", "7", "--repeat", "3", "sv_SE.word")
 	code, got, errb := runOut("--seed", "007", "--repeat", "+3", "sv_SE.word")

fejkdata.go

@@ -45,6 +45,7 @@ type Generator struct {
 	mu         sync.Mutex
 	rand       *session
 	categories map[string]node
+	records    map[node]recordShape
 }
 
 // session is one generator's mutable render state: the seeded rng plus the {seq()}

hold.go

@@ -267,14 +267,20 @@ type draws struct {
 // gives one value, and a shown operand is the operand computed. Every other name is
 // drawn afresh, so {word} {word} still draws twice. checkTokens, checkPath and
 // linkRefs prove every step, so the walk cannot fail.
-func readField(s *session, t *template, held *draws, a arm) string {
+func readField(s *session, t *template, held, refScope *draws, a arm) string {
 	if !t.held[a.key] {
 		if len(a.tail) > 0 {
 			panic(fmt.Sprintf("fejkdata: %q reads a path into %q, which the expansion does not hold", a.name, a.key))
 		}
-		return render(s, t.fields[a.key])
+		return render(s, t.fields[a.key], refScope)
 	}
-	if v, read := held.value[a.path]; read {
+	// A reference that reads a path reads the caller's scope, so its draw outlives
+	// this expansion; a sibling, and a reference read whole, stay local to it.
+	d := held
+	if isRef(a.key) && refScope != nil && len(a.tail) > 0 {
+		d = refScope
+	}
+	if v, read := d.value[a.path]; read {
 		return v
 	}
 	var v string
@@ -286,16 +292,16 @@ func readField(s *session, t *template, held *draws, a arm) string {
 			if consumed := len(a.tail) - len(rest); consumed > 0 {
 				key = a.steps[consumed-1]
 			}
-			n, drew := held.variant[key]
+			n, drew := d.variant[key]
 			if !drew {
 				n = drawn(s, c)
-				held.variant[key] = n
+				d.variant[key] = n
 			}
 			return []node{n}, nil
 		},
-		leaf: func(n node) error { v = render(s, n); return nil },
+		leaf: func(n node) error { v = render(s, n, refScope); return nil },
 	})
-	held.value[a.path] = v
+	d.value[a.path] = v
 	return v
 }
 

hold_test.go

@@ -718,6 +718,6 @@ func TestReadFieldPanicsOnAnUnheldPath(t *testing.T) {
 		t.Fatal("not a template")
 	}
 	mustPanic(t, "unheld arm with a path", func() {
-		readField(engine(1).rand, tm, nil, arm{name: "w.x", key: "w", tail: []string{"x"}, path: "w.x"})
+		readField(engine(1).rand, tm, nil, nil, arm{name: "w.x", key: "w", tail: []string{"x"}, path: "w.x"})
 	})
 }

inline.go

@@ -20,7 +20,7 @@ type Template struct {
 func (t *Template) Fake() string {
 	t.g.mu.Lock()
 	defer t.g.mu.Unlock()
-	return render(t.g.rand, t.n)
+	return render(t.g.rand, t.n, nil)
 }
 
 // NewTemplate compiles an inline template — a format string or a JSON value — and

perf_test.go

@@ -51,6 +51,26 @@ func TestNoRenderAllocRegression(t *testing.T) {
 	}
 }
 
+// A record's fences read the compiled tree, so they belong to New, not to a draw.
+func TestNoRecordAllocRegression(t *testing.T) {
+	for _, s := range []struct{ name, json string }{
+		{"record 3 columns", `{"format":"","a":"x","b":"y","c":"z"}`},
+		{"record 50 columns", wideTokenJSON(50)},
+	} {
+		f, err := New(WithoutShippedData(), WithDataFS(fstest.MapFS{"x.json": {Data: []byte(s.json)}}))
+		if err != nil {
+			t.Fatalf("New(%s): %v", s.name, err)
+		}
+		if _, err := f.Record("x"); err != nil {
+			t.Fatalf("Record(%s): %v", s.name, err) // else the gate would measure the error path
+		}
+		const base = 4.0
+		if allocs := testing.AllocsPerRun(10000, func() { f.Record("x") }); allocs > base*1.10 {
+			t.Errorf("%s: %.1f allocs/op regressed past %.1f (baseline %.1f + 10%%); a record fence running per draw is the usual cause", s.name, allocs, base*1.10, base)
+		}
+	}
+}
+
 func BenchmarkNestedDepth25(b *testing.B)  { benchPath(b, tmpData(b, "deep", nestedJSON(25)), "deep") }
 func BenchmarkNestedDepth100(b *testing.B) { benchPath(b, tmpData(b, "deep", nestedJSON(100)), "deep") }
 func BenchmarkWideTokens100(b *testing.B) {

readme_test.go

@@ -43,6 +43,34 @@ func TestReadmeExamplesLoadAndRender(t *testing.T) {
 	}
 }
 
+func TestReadmeRecordExample(t *testing.T) {
+	src := readme(t)
+	section := src[strings.Index(src, "### Records"):]
+	section = section[:strings.Index(section, "## Library")]
+	block := jsonBlock.FindStringSubmatch(section)
+	if block == nil {
+		t.Fatal("README lost the Records example")
+	}
+	f, err := New(WithDataPath(writeData(t, map[string]string{"users": block[1]})), WithSeed(1))
+	if err != nil {
+		t.Fatal(err)
+	}
+	r, err := f.Record("users")
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := r.Columns()
+	if len(got) != 2 || got[0].Name != "first" || got[0].Value != "Bo" || got[1].Name != "last" || got[1].Value != "Lovelace" {
+		t.Fatalf("record columns = %v, want first=Bo, last=Lovelace with seed 1", got)
+	}
+	if r.CSVHeader() != "first,last" || r.CSVLine() != "Bo,Lovelace" {
+		t.Fatalf("csv = %q, %q, want first,last / Bo,Lovelace", r.CSVHeader(), r.CSVLine())
+	}
+	if r.SQLInsert("users") != `INSERT INTO "users" ("first", "last") VALUES ('Bo', 'Lovelace');` {
+		t.Fatalf("sql = %q, want the seeded INSERT", r.SQLInsert("users"))
+	}
+}
+
 func TestReadmeSQLExampleOutput(t *testing.T) {
 	src := readme(t)
 	src = src[strings.Index(src, "### Your own data"):]

record.go

@@ -0,0 +1,285 @@
+package fejkdata
+
+import (
+	"encoding/csv"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"sort"
+	"strings"
+)
+
+// Column is one rendered column of a record.
+type Column struct {
+	Name  string
+	Value string
+}
+
+// Record is one record rendered from a template: every direct field is a column,
+// listed in name order. Each column is its own expansion, so a sibling field is
+// local to it, while a reference that reads a path is drawn once for the whole
+// record.
+type Record struct {
+	columns []Column
+}
+
+// Columns returns the record's columns in name order.
+func (r *Record) Columns() []Column {
+	return append([]Column(nil), r.columns...)
+}
+
+// JSON renders the record as one JSON object, every column a string.
+func (r *Record) JSON() string {
+	m := make(map[string]string, len(r.columns))
+	for _, c := range r.columns {
+		m[c.Name] = c.Value
+	}
+	b, _ := json.Marshal(m)
+	return string(b)
+}
+
+// CSVHeader renders the column names as one CSV header line.
+func (r *Record) CSVHeader() string {
+	return csvLine(r.names())
+}
+
+// CSVLine renders the column values as one CSV row.
+func (r *Record) CSVLine() string {
+	return csvLine(r.values())
+}
+
+func (r *Record) names() []string {
+	out := make([]string, len(r.columns))
+	for i, c := range r.columns {
+		out[i] = c.Name
+	}
+	return out
+}
+
+func (r *Record) values() []string {
+	out := make([]string, len(r.columns))
+	for i, c := range r.columns {
+		out[i] = c.Value
+	}
+	return out
+}
+
+func csvLine(cols []string) string {
+	var b strings.Builder
+	w := csv.NewWriter(&b)
+	_ = w.Write(cols)
+	w.Flush()
+	line := strings.TrimSuffix(b.String(), "\n")
+	if line == "" {
+		return `""` // a blank line is a row every CSV reader drops
+	}
+	return line
+}
+
+// SQLInsert renders the record as one INSERT statement into table: identifiers in
+// ANSI double quotes, every value a single-quoted string literal.
+func (r *Record) SQLInsert(table string) string {
+	cols := make([]string, len(r.columns))
+	vals := make([]string, len(r.columns))
+	for i, c := range r.columns {
+		cols[i] = quoteIdent(c.Name)
+		vals[i] = "'" + strings.ReplaceAll(c.Value, "'", "''") + "'"
+	}
+	return fmt.Sprintf("INSERT INTO %s (%s) VALUES (%s);", quoteIdent(table), strings.Join(cols, ", "), strings.Join(vals, ", "))
+}
+
+func quoteIdent(s string) string {
+	return `"` + strings.ReplaceAll(s, `"`, `""`) + `"`
+}
+
+// Record renders a path as one record: the template it names, with each direct
+// field drawn as a column. Only a category-level template is a record — a path
+// that descends into a field, or that names a folder or a choice, is an error.
+func (f *Generator) Record(path string) (*Record, error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	_, n, tail, err := resolveCategory(f.categories, strings.Split(path, "."))
+	if err != nil {
+		return nil, fmt.Errorf("fejkdata: %s: %w", path, err)
+	}
+	if len(tail) > 0 {
+		return nil, fmt.Errorf("fejkdata: %s descends into %q, a field; only a category-level template is a record", path, tail[0])
+	}
+	shape := f.recordShapeOf(n)
+	if shape.err != nil {
+		return nil, fmt.Errorf("fejkdata: %s %w", path, shape.err)
+	}
+	return renderRecord(f.rand, shape.t, shape.columns), nil
+}
+
+// recordShape is what recordOf settled about a node: the template to project, its
+// columns, or why it is not a record.
+type recordShape struct {
+	t       *template
+	columns []string
+	err     error
+}
+
+// recordShapeOf fences a node once and remembers the answer. Callers hold the
+// generator's lock.
+func (f *Generator) recordShapeOf(n node) recordShape {
+	if shape, done := f.records[n]; done {
+		return shape
+	}
+	t, columns, err := recordOf(n)
+	shape := recordShape{t: t, columns: columns, err: err}
+	if f.records == nil {
+		f.records = map[node]recordShape{}
+	}
+	f.records[n] = shape
+	return shape
+}
+
+// RecordTemplate is an inline record compiled, referenced and validated once,
+// ready to render many times with [RecordTemplate.Fake].
+type RecordTemplate struct {
+	g       *Generator
+	t       *template
+	columns []string
+}
+
+// Fake renders the record with one draw.
+func (t *RecordTemplate) Fake() *Record {
+	t.g.mu.Lock()
+	defer t.g.mu.Unlock()
+	return renderRecord(t.g.rand, t.t, t.columns)
+}
+
+// NewRecordTemplate compiles an inline record — a JSON object with a format and
+// fields — and binds its references against the loaded tree.
+func (f *Generator) NewRecordTemplate(input string) (*RecordTemplate, error) {
+	t, err := f.NewTemplate(input)
+	if err != nil {
+		return nil, err
+	}
+	tm, columns, err := recordOf(t.n)
+	if err != nil {
+		return nil, fmt.Errorf("fejkdata: an inline record %w", err)
+	}
+	return &RecordTemplate{g: f, t: tm, columns: columns}, nil
+}
+
+// FakeRecord compiles and renders an inline record in one call.
+func (f *Generator) FakeRecord(input string) (*Record, error) {
+	t, err := f.NewRecordTemplate(input)
+	if err != nil {
+		return nil, err
+	}
+	return t.Fake(), nil
+}
+
+// recordOf is the fence both record entry points pass. The columns come back with
+// the template, fixed for every draw the caller goes on to make.
+func recordOf(n node) (*template, []string, error) {
+	t, ok := n.(*template)
+	if !ok {
+		return nil, nil, errors.New("names a choice, not a template; a record is a template whose fields are its columns")
+	}
+	if t.repeat != 1 {
+		return nil, nil, fmt.Errorf("carries repeat %d, which composes its format into one string; a record projects columns instead — drop the repeat and render the record again for more rows", t.repeat)
+	}
+	columns := recordColumns(t)
+	if len(columns) == 0 {
+		return nil, nil, errors.New("has no fields, so no columns")
+	}
+	if err := checkColumnRefs(t, columns); err != nil {
+		return nil, nil, err
+	}
+	return t, columns, nil
+}
+
+// checkColumnRefs rejects the reference reads a record's shared draw cannot answer
+// for: one column rendering a level another reads a path into, and a column
+// reading the record back through its own path.
+func checkColumnRefs(t *template, columns []string) error {
+	reads, err := columnRefs(t, columns)
+	if err != nil {
+		return err
+	}
+	sort.Slice(reads, func(i, j int) bool {
+		if reads[i].a.path != reads[j].a.path {
+			return reads[i].a.path < reads[j].a.path
+		}
+		return reads[i].column < reads[j].column
+	})
+	for i, level := range reads {
+		for _, into := range reads[i+1:] {
+			if strings.HasPrefix(into.a.path, level.a.path+".") {
+				return fmt.Errorf("column %q renders {%s}, a level column %q reads a path into with {%s}; name the fields you want instead",
+					level.column, level.a.name, into.column, into.a.name)
+			}
+		}
+	}
+	return nil
+}
+
+// columnRef is one held reference read, and the column whose render reaches it.
+type columnRef struct {
+	column string
+	a      arm
+}
+
+// columnRefs lists every reference that reads a path, anywhere a column renders,
+// following the same edges expand does. A reference landing back on the record
+// itself is reported rather than collected, whether it reads a path or the record
+// whole: either way the column describes a draw other than its neighbours'.
+func columnRefs(t *template, columns []string) ([]columnRef, error) {
+	var out []columnRef
+	var err error
+	for _, name := range columns {
+		seen := map[node]bool{}
+		var walk func(n node)
+		walk = func(n node) {
+			if n == nil || seen[n] || err != nil {
+				return
+			}
+			seen[n] = true
+			if tm, ok := n.(*template); ok {
+				for _, ref := range refTokens(tm.format) {
+					a := splitArm(ref, tm.refs)
+					if tm.fields[a.key] == node(t) {
+						err = fmt.Errorf("column %q reads {%s}, which points back at this record; a column cannot read another column — move the shared value into its own category and reference that", name, a.name)
+						return
+					}
+					if len(a.tail) > 0 {
+						out = append(out, columnRef{name, a})
+					}
+				}
+			}
+			for _, e := range renderEdges(n) {
+				walk(e.to)
+			}
+		}
+		walk(t.fields[name])
+	}
+	return out, err
+}
+
+// renderRecord draws each column once, in the name order recordOf fixed, over one
+// reference scope shared across them.
+func renderRecord(s *session, t *template, columns []string) *Record {
+	scope := &draws{variant: map[string]node{}, value: map[string]string{}}
+	r := &Record{columns: make([]Column, len(columns))}
+	for i, name := range columns {
+		r.columns[i] = Column{Name: name, Value: render(s, t.fields[name], scope)}
+	}
+	return r
+}
+
+// recordColumns is the sorted non-reference field names — the columns a record
+// projects. A {/path} binding is carried in fields under its root path, so only a
+// name that is not a reference is a column.
+func recordColumns(t *template) []string {
+	var names []string
+	for _, name := range sortedNames(t.fields) {
+		if !isRef(name) {
+			names = append(names, name)
+		}
+	}
+	return names
+}

record_test.go

@@ -0,0 +1,408 @@
+package fejkdata
+
+import (
+	"encoding/csv"
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+func recordCat(t *testing.T) *Generator {
+	t.Helper()
+	dir := writeData(t, map[string]string{
+		"users": `{
+			"format": "{first} {last}",
+			"first": ["Ada", "Bo"],
+			"last": ["Lovelace", "Ek"]
+		}`,
+	})
+	return newGenerator(t, dir, WithSeed(1))
+}
+
+func TestRecordProjectsFieldsAsColumns(t *testing.T) {
+	f := recordCat(t)
+	r, err := f.Record("users")
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := r.Columns()
+	if len(got) != 2 || got[0].Name != "first" || got[1].Name != "last" {
+		t.Fatalf("Record.Columns() = %v, want columns first, last in name order", got)
+	}
+	if (got[0].Value != "Ada" && got[0].Value != "Bo") || (got[1].Value != "Lovelace" && got[1].Value != "Ek") {
+		t.Fatalf("columns = %v, want the field values", got)
+	}
+}
+
+func TestRecordIsDeterministic(t *testing.T) {
+	a, b := recordCat(t), recordCat(t)
+	for i := 0; i < 20; i++ {
+		x, _ := a.Record("users")
+		y, _ := b.Record("users")
+		if x.JSON() != y.JSON() {
+			t.Fatalf("same seed diverged: %s != %s", x.JSON(), y.JSON())
+		}
+	}
+}
+
+func TestRecordSkipsReferenceBindings(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"name": `{"format": "{first} {last}", "first": "Ada", "last": "Lovelace"}`,
+		"user": `{"format": "they are {/name}", "id": "1"}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	r, err := f.Record("user")
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := r.Columns()
+	if len(got) != 1 || got[0].Name != "id" {
+		t.Fatalf("Record.Columns() = %v, want only the id column (a {/path} is not a column)", got)
+	}
+}
+
+func TestRecordErrors(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"bare":      `"hello"`,
+		"pick":      `["a", "b"]`,
+		"group/cat": `"x"`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	for _, c := range []struct {
+		path string
+		want string
+	}{
+		{"bare", "has no fields, so no columns"},
+		{"pick", "names a choice"},
+		{"group", "names a folder"},
+		{"nope", "no entry"},
+	} {
+		if _, err := f.Record(c.path); err == nil || !strings.Contains(err.Error(), c.want) {
+			t.Errorf("Record(%q) = %v, want an error containing %q", c.path, err, c.want)
+		}
+	}
+}
+
+func TestRecordJSON(t *testing.T) {
+	f := recordCat(t)
+	r, err := f.Record("users")
+	if err != nil {
+		t.Fatal(err)
+	}
+	var m map[string]string
+	if err := json.Unmarshal([]byte(r.JSON()), &m); err != nil {
+		t.Fatalf("Record.JSON() is not valid JSON: %v\n%s", err, r.JSON())
+	}
+	if len(m) != 2 || (m["first"] != "Ada" && m["first"] != "Bo") {
+		t.Fatalf("Record.JSON() = %s, want two addressable columns", r.JSON())
+	}
+	for _, f := range r.Columns() {
+		if m[f.Name] != f.Value {
+			t.Fatalf("JSON column %q = %q, want %q", f.Name, m[f.Name], f.Value)
+		}
+	}
+}
+
+func TestRecordCSV(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"note": `{"format": "{word}, {word}", "word": ["a", "b,c", "d\"e", "f\n"]}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	seen := map[string]bool{}
+	for i := 0; i < 200 && len(seen) < 4; i++ {
+		r, err := f.Record("note")
+		if err != nil {
+			t.Fatal(err)
+		}
+		if got := r.CSVHeader(); got != "word" {
+			t.Fatalf("CSVHeader() = %q, want word", got)
+		}
+		rec, err := csv.NewReader(strings.NewReader(r.CSVLine() + "\n")).Read()
+		if err != nil {
+			t.Fatalf("CSVLine() is not valid CSV: %v\n%q", err, r.CSVLine())
+		}
+		want := r.Columns()[0].Value
+		if len(rec) != 1 || rec[0] != want {
+			t.Fatalf("CSVLine() = %v, want the field value %q round-tripped", rec, want)
+		}
+		seen[want] = true
+	}
+	if len(seen) != 4 {
+		t.Fatalf("round-tripped %d of the 4 values; the comma, quote and newline shapes must each survive", len(seen))
+	}
+}
+
+func TestRecordCSVEmptyValueStaysARow(t *testing.T) {
+	dir := writeData(t, map[string]string{"blank": `{"format": "", "note": ""}`})
+	f := newGenerator(t, dir, WithSeed(1))
+	r, err := f.Record("blank")
+	if err != nil {
+		t.Fatal(err)
+	}
+	rows, err := csv.NewReader(strings.NewReader(r.CSVHeader() + "\n" + r.CSVLine() + "\n")).ReadAll()
+	if err != nil {
+		t.Fatalf("csv: %v", err)
+	}
+	if len(rows) != 2 || len(rows[1]) != 1 || rows[1][0] != "" {
+		t.Fatalf("one empty column parsed to %v, want a header and one row of one empty field", rows)
+	}
+}
+
+func TestRecordRejectsOverlappingReferenceColumns(t *testing.T) {
+	cat := `{"format":"","a":[{"format":"A={b}","b":"1"},{"format":"A={b}","b":"2"}]}`
+	for _, c := range []struct{ name, row string }{
+		{"sibling columns", `{"format":"","whole":"{/cat.a}","inner":"{/cat.a.b}"}`},
+		{"through a nested template", `{"format":"","whole":"{/cat.a}","inner":{"format":"{/cat.a.b} {x}","x":"1"}}`},
+		{"through a column repeat", `{"format":"","whole":"{/cat.a}","inner":{"format":"{/cat.a.b}","repeat":2,"separator":"-"}}`},
+		{"through a choice variant", `{"format":"","whole":"{/cat.a}","inner":[{"format":"{/cat.a.b} {x}","x":"1"},{"format":"{/cat.a.b}! {x}","x":"2"}]}`},
+		{"as a builtin operand", `{"format":"","whole":"{uppercase(/cat.a)}","inner":"{/cat.a.b}"}`},
+	} {
+		f := newGenerator(t, writeData(t, map[string]string{"cat": cat, "row": c.row}), WithSeed(1))
+		_, err := f.Record("row")
+		if err == nil || !strings.Contains(err.Error(), "reads a path into") {
+			t.Errorf("%s: Record = %v, want the overlap rejected the way one format is", c.name, err)
+			continue
+		}
+		if !strings.Contains(err.Error(), `"whole"`) || !strings.Contains(err.Error(), `"inner"`) {
+			t.Errorf("%s: error %q names neither column; it must name both", c.name, err)
+		}
+		if _, err := f.Fake("row"); err != nil {
+			t.Errorf("%s: Fake(row) = %v, want the string view untouched", c.name, err)
+		}
+	}
+}
+
+func TestInlineRecordRejectsOverlappingColumns(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"cat": `{"format":"","a":[{"format":"A={b}","b":"1"},{"format":"A={b}","b":"2"}]}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	_, err := f.FakeRecord(`{"format":"","whole":"{/cat.a}","inner":"{/cat.a.b}"}`)
+	if err == nil || !strings.Contains(err.Error(), "reads a path into") {
+		t.Fatalf("inline record over an overlapping pair = %v, want the inline entry point to refuse it too", err)
+	}
+}
+
+func TestRecordRejectsAColumnReadingItsOwnRecord(t *testing.T) {
+	for _, c := range []struct{ name, column string }{
+		{"a path into itself", `"full":"{/person.first} {/person.last}"`},
+		{"the record read whole", `"whole":"{/person}"`},
+		{"the record as an operand", `"up":"{uppercase(/person)}"`},
+	} {
+		person := `{"format":"{first} {last}","first":["Ada","Bo"],"last":["Lovelace","Ek"],` + c.column + `}`
+		f := newGenerator(t, writeData(t, map[string]string{"person": person}), WithSeed(1))
+		if _, err := f.Record("person"); err == nil || !strings.Contains(err.Error(), "points back at this record") {
+			t.Errorf("%s: Record = %v, want it refused; the column would contradict the columns beside it", c.name, err)
+		}
+	}
+}
+
+func TestRecordBareReferenceStaysIndependentAsAnOperand(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"cur": `[{"format":"{code}","code":"aud"},{"format":"{code}","code":"eur"}]`,
+		"row": `{"format":"","up":"{uppercase(/cur)}","low":"{lowercase(/cur)}"}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	sawMismatch := false
+	for i := 0; i < 200 && !sawMismatch; i++ {
+		r, err := f.Record("row")
+		if err != nil {
+			t.Fatal(err)
+		}
+		m := map[string]string{}
+		for _, c := range r.Columns() {
+			m[c.Name] = c.Value
+		}
+		sawMismatch = !strings.EqualFold(m["up"], m["low"])
+	}
+	if !sawMismatch {
+		t.Error("two bare-reference operand columns never disagreed; a bare reference draws on its own, as the plain spelling does")
+	}
+	for i := 0; i < 50; i++ {
+		v, err := f.FakeTemplate("{/cur}|{uppercase(/cur)}")
+		if err != nil {
+			t.Fatal(err)
+		}
+		if parts := strings.Split(v, "|"); !strings.EqualFold(parts[0], parts[1]) {
+			t.Fatalf("one format rendered %q; within an expansion a bare reference is still one draw", v)
+		}
+	}
+}
+
+func TestRecordSQLInsert(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"person": `{"format": "{last}", "last": "O'Brien"}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	r, err := f.Record("person")
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := r.SQLInsert("people")
+	if got != `INSERT INTO "people" ("last") VALUES ('O''Brien');` {
+		t.Fatalf("SQLInsert() = %q, want quoted identifiers and the single quote doubled", got)
+	}
+}
+
+func TestRecordRejectsFieldDescent(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"cat": `{"format":"{sub}","sub":{"format":"{x}","x":"1"}}`,
+		"row": `[{"format":"{x}","x":"1"},{"format":"{x}","x":"2"}]`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	for _, path := range []string{"cat.sub", "row.x"} {
+		if _, err := f.Record(path); err == nil || !strings.Contains(err.Error(), "field") {
+			t.Errorf("Record(%q) = %v, want a 'descends into a field' error", path, err)
+		}
+	}
+}
+
+func TestRecordSharesAReferenceAcrossColumns(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"currency": `[{"format":"{code}","code":"AUD","symbol":"$"},{"format":"{code}","code":"EUR","symbol":"€"}]`,
+		"price":    `{"format":"","code":"{/currency.code}","symbol":"{/currency.symbol}"}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	for i := 0; i < 100; i++ {
+		r, err := f.Record("price")
+		if err != nil {
+			t.Fatal(err)
+		}
+		m := map[string]string{}
+		for _, c := range r.Columns() {
+			m[c.Name] = c.Value
+		}
+		switch m["code"] {
+		case "AUD":
+			if m["symbol"] != "$" {
+				t.Fatalf("record %q: code AUD but symbol %q, want one currency draw across columns", r.JSON(), m["symbol"])
+			}
+		case "EUR":
+			if m["symbol"] != "€" {
+				t.Fatalf("record %q: code EUR but symbol %q, want one currency draw across columns", r.JSON(), m["symbol"])
+			}
+		default:
+			t.Fatalf("record %q has unexpected code %q", r.JSON(), m["code"])
+		}
+	}
+}
+
+func TestRecordSharesAReferenceIntoAColumnRepeat(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"currency": `[{"format":"{code}","code":"AUD"},{"format":"{code}","code":"EUR"}]`,
+		"order":    `{"format":"","codes":{"format":"{/currency.code}","repeat":3,"separator":"-"}}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	for i := 0; i < 50; i++ {
+		r, err := f.Record("order")
+		if err != nil {
+			t.Fatal(err)
+		}
+		parts := strings.Split(r.Columns()[0].Value, "-")
+		if len(parts) != 3 || parts[0] != parts[1] || parts[1] != parts[2] {
+			t.Fatalf("codes column = %q, want one shared draw across its repeat", r.Columns()[0].Value)
+		}
+	}
+}
+
+func TestRecordBareReferenceStaysIndependent(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"currency": `[{"format":"{code}","code":"AUD"},{"format":"{code}","code":"EUR"}]`,
+		"order":    `{"format":"","whole":"{/currency}","code":"{/currency.code}"}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	sawMismatch := false
+	for i := 0; i < 100; i++ {
+		r, err := f.Record("order")
+		if err != nil {
+			t.Fatal(err)
+		}
+		m := map[string]string{}
+		for _, c := range r.Columns() {
+			m[c.Name] = c.Value
+		}
+		if m["whole"] != m["code"] {
+			sawMismatch = true
+			break
+		}
+	}
+	if !sawMismatch {
+		t.Fatal("a bare {/currency} column never disagreed with a tailed {/currency.code} column; a bare reference should draw independently")
+	}
+}
+
+func TestRecordSQLQuotesIdentifiers(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"row": `{"format": "", "postal-code": "1", "street-number": "2"}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	r, err := f.Record("row")
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := r.SQLInsert("my-table")
+	if got != `INSERT INTO "my-table" ("postal-code", "street-number") VALUES ('1', '2');` {
+		t.Fatalf("SQLInsert() = %q, want hyphenated identifiers and table quoted", got)
+	}
+}
+
+func TestFakeRecordAndTemplate(t *testing.T) {
+	f := recordCat(t)
+	in := `{"format":"{x} {y}","x":["1","2"],"y":["3","4"]}`
+	want, err := f.FakeRecord(in)
+	if err != nil {
+		t.Fatalf("FakeRecord: %v", err)
+	}
+	if len(want.Columns()) != 2 {
+		t.Fatalf("FakeRecord columns = %v, want two columns", want.Columns())
+	}
+	reusable, err := f.NewRecordTemplate(in)
+	if err != nil {
+		t.Fatalf("NewRecordTemplate: %v", err)
+	}
+	for i := 0; i < 20; i++ {
+		if got := reusable.Fake(); len(got.Columns()) != 2 {
+			t.Fatalf("RecordTemplate.Fake() = %v, want two columns", got.Columns())
+		}
+	}
+}
+
+func TestInlineRecordErrors(t *testing.T) {
+	f := recordCat(t)
+	for _, c := range []struct {
+		input string
+		want  string
+	}{
+		{`"hello"`, "has no fields, so no columns"},
+		{`["a","b"]`, "a record is a template whose fields are its columns"},
+		{`{"format":"{a}-","repeat":3,"separator":"|","a":["x","y"]}`, "carries repeat 3"},
+	} {
+		if _, err := f.FakeRecord(c.input); err == nil || !strings.Contains(err.Error(), c.want) {
+			t.Errorf("FakeRecord(%q) = %v, want an error naming %q", c.input, err, c.want)
+		}
+	}
+}
+
+func TestRecordRejectsATopLevelRepeat(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"rep": `{"format":"{a}-","repeat":3,"separator":"|","a":["x","y"]}`,
+	})
+	f := newGenerator(t, dir, WithSeed(1))
+	if _, err := f.Record("rep"); err == nil || !strings.Contains(err.Error(), "carries repeat 3") {
+		t.Errorf("Record on a repeating template = %v, want an error naming the repeat", err)
+	}
+	if v, err := f.Fake("rep"); err != nil || v != "x-|x-|x-" {
+		t.Errorf("Fake(rep) = %q, %v, want the repeat still composed for the string view", v, err)
+	}
+}
+
+func TestRecordTemplateRejectsATopLevelRepeat(t *testing.T) {
+	f := recordCat(t)
+	in := `{"format":"{a}-","repeat":3,"separator":"|","a":["x","y"]}`
+	if _, err := f.NewRecordTemplate(in); err == nil || !strings.Contains(err.Error(), "carries repeat 3") {
+		t.Errorf("NewRecordTemplate on a repeating template = %v, want an error naming the repeat", err)
+	}
+	if _, err := f.NewTemplate(in); err != nil {
+		t.Errorf("NewTemplate on the same input = %v, want the string view to still compile", err)
+	}
+}

reference.go

@@ -93,7 +93,7 @@ func linkTemplateRefs(folder []string, path string, t *template, root map[string
 		if err != nil {
 			return fmt.Errorf("%s: reference {%s}: %w", path, name, err)
 		}
-		head, target, tail, err := resolveRef(root, segments)
+		head, target, tail, err := resolveCategory(root, segments)
 		if err != nil {
 			return fmt.Errorf("%s: reference {%s}: %w", path, name, err)
 		}
@@ -146,11 +146,11 @@ func eachTemplate(root map[string]node, fn func(folder []string, path string, t
 	return inFolder(nil, root)
 }
 
-// resolveRef walks a reference path through the folders to the category it names,
-// returning that head, the node, and the tail left to read into it. A descent of
-// its own rather than a walkPath: it walks groups only and returns where they end,
-// not a leaf.
-func resolveRef(root map[string]node, segments []string) (head []string, target node, tail []string, err error) {
+// resolveCategory walks a dotted path through the folders to the category it
+// names, returning that head, the node, and the tail left to read into it. A
+// descent of its own rather than a walkPath: it walks groups only and returns
+// where they end, not a leaf.
+func resolveCategory(root map[string]node, segments []string) (head []string, target node, tail []string, err error) {
 	var n node = &group{children: root}
 	i := 0
 	for ; i < len(segments); i++ {

render.go

@@ -27,7 +27,7 @@ func (f *Generator) Fake(path string) (string, error) {
 	if _, ok := n.(*group); ok {
 		return "", fmt.Errorf("fejkdata: %s names a folder, not a value", path)
 	}
-	return render(f.rand, n), nil
+	return render(f.rand, n, nil), nil
 }
 
 // descend walks named fields to the node a path names. It is the one render-side
@@ -50,17 +50,18 @@ func descend(s *session, root node, segments []string) (node, error) {
 }
 
 // render evaluates a compiled node to a string. compile validates every node up
-// front, so rendering a compiled tree cannot fail.
-func render(s *session, n node) string {
+// front, so rendering a compiled tree cannot fail. refScope carries the draws a
+// reference shares beyond its own expansion; nil keeps every reference local.
+func render(s *session, n node, refScope *draws) string {
 	switch n := n.(type) {
 	case *choice:
-		return render(s, pick(s, n))
+		return render(s, pick(s, n), refScope)
 	case *template:
 		if n.repeat == 1 {
 			if n.fixed {
 				return n.lit
 			}
-			return expand(s, n)
+			return expand(s, n, refScope)
 		}
 		var b strings.Builder
 		b.Grow(n.repeat * (n.grow + len(n.separator)))
@@ -68,7 +69,7 @@ func render(s *session, n node) string {
 			if i > 0 {
 				b.WriteString(n.separator)
 			}
-			b.WriteString(expand(s, n))
+			b.WriteString(expand(s, n, refScope))
 		}
 		return b.String()
 	default:
@@ -90,11 +91,12 @@ func pick(r rng, c *choice) node {
 
 // expand renders a template's compiled ops. compile validated every token, so this
 // cannot fail.
-func expand(s *session, t *template) string {
+func expand(s *session, t *template, refScope *draws) string {
 	var b strings.Builder
 	b.Grow(t.grow)
 	// One draw per held name, for this expansion only: a nested template and each
-	// repeat iteration get their own, since each is its own expansion.
+	// repeat iteration get their own, since each is its own expansion. A reference
+	// reads the caller's scope instead, whenever one was supplied.
 	var held *draws
 	if len(t.held) > 0 {
 		held = &draws{
@@ -108,7 +110,7 @@ func expand(s *session, t *template) string {
 		case 'l':
 			b.WriteString(o.lit)
 		case 'f':
-			b.WriteString(readField(s, t, held, o.arms[s.IntN(len(o.arms))]))
+			b.WriteString(readField(s, t, held, refScope, o.arms[s.IntN(len(o.arms))]))
 		case 'b':
 			// Read before the call, so the value a calc computes is the value the
 			// format showed. calcVars fixed the order op.operands holds.
@@ -116,7 +118,7 @@ func expand(s *session, t *template) string {
 			if len(o.operands) > 0 {
 				operands = make([]string, len(o.operands))
 				for j, a := range o.operands {
-					operands[j] = readField(s, t, held, a)
+					operands[j] = readField(s, t, held, refScope, a)
 				}
 			}
 			b.WriteString(o.call(s, b.String(), operands)) // b.String() is the output so far

shipped_data_test.go

@@ -108,6 +108,10 @@ func TestFakeIsSafeForConcurrentUse(t *testing.T) {
 					return
 				}
 				f.List()
+				if _, err := f.Record("sv_SE.person"); err != nil {
+					t.Error(err)
+					return
+				}
 				tmpl, err := f.NewTemplate("{/sv_SE.person.last}")
 				if err != nil {
 					t.Error(err)

template_test.go

@@ -38,7 +38,7 @@ func compiled(t *testing.T, s string) node {
 
 func mustRender(t *testing.T, f *Generator, s string) string {
 	t.Helper()
-	return render(f.rand, compiled(t, s))
+	return render(f.rand, compiled(t, s), nil)
 }
 
 func TestStringIsAFormat(t *testing.T) {
@@ -315,7 +315,7 @@ func TestGrowIsALowerBound(t *testing.T) {
 			t.Fatalf("format %q did not compile to a template", format)
 		}
 		for i := 0; i < 50; i++ {
-			if got := len(expand(f.rand, tmpl)); got < tmpl.grow {
+			if got := len(expand(f.rand, tmpl, nil)); got < tmpl.grow {
 				t.Errorf("format %q: expand emitted %d bytes, below grow %d", format, got, tmpl.grow)
 			}
 		}