diff --git a/tools/jsontypes/cmd/jsoncoalesce/main.go b/tools/jsontypes/cmd/jsoncoalesce/main.go
new file mode 100644
index 0000000..684abc3
--- /dev/null
+++ b/tools/jsontypes/cmd/jsoncoalesce/main.go
@@ -0,0 +1,38 @@
+package main
+
+import (
+	"bufio"
+	"fmt"
+	"os"
+
+	"github.com/therootcompany/golib/tools/jsontypes"
+)
+
+func main() {
+	var input *os.File
+	if len(os.Args) > 1 && os.Args[1] != "-" {
+		f, err := os.Open(os.Args[1])
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "error: %v\n", err)
+			os.Exit(1)
+		}
+		defer f.Close()
+		input = f
+	} else {
+		input = os.Stdin
+	}
+
+	var lines []string
+	scanner := bufio.NewScanner(input)
+	for scanner.Scan() {
+		lines = append(lines, scanner.Text())
+	}
+	if err := scanner.Err(); err != nil {
+		fmt.Fprintf(os.Stderr, "error reading input: %v\n", err)
+		os.Exit(1)
+	}
+
+	for _, line := range jsontypes.Coalesce(lines) {
+		fmt.Println(line)
+	}
+}
diff --git a/tools/jsontypes/coalesce.go b/tools/jsontypes/coalesce.go
new file mode 100644
index 0000000..22d33da
--- /dev/null
+++ b/tools/jsontypes/coalesce.go
@@ -0,0 +1,341 @@
+package jsontypes
+
+import (
+	"fmt"
+	"regexp"
+	"sort"
+	"strings"
+)
+
+// Coalesce takes streamer output lines and returns coalesced lines with:
+//   - Types deduplicated by key set (same keys = same type)
+//   - Nullable annotation ({string?} when {null} + concrete type at same path)
+//   - Cleaner type names (no number when unique, per-base-name numbering for conflicts)
+//   - Duplicate type definitions elided (referenced by name only)
+func Coalesce(lines []string) []string {
+	parsed := parseRawLines(lines)
+	types := buildCoalesceTypes(parsed)
+	groups := groupTypesByKeySet(types)
+	nameMap := assignCanonicalNames(groups)
+	nullables := detectNullablePaths(parsed)
+	return emitCoalesced(parsed, nameMap, nullables)
+}
+
+// rawLine is a parsed streamer output line.
+type rawLine struct {
+	path    string // everything before the last {
+	typeVal string // everything inside the last {}
+}
+
+func parseRawLines(lines []string) []rawLine {
+	result := make([]rawLine, 0, len(lines))
+	for _, line := range lines {
+		line = strings.TrimSpace(line)
+		if line == "" {
+			continue
+		}
+		result = append(result, parseRawLine(line))
+	}
+	return result
+}
+
+func parseRawLine(line string) rawLine {
+	idx := strings.LastIndex(line, "{")
+	if idx < 0 || !strings.HasSuffix(line, "}") {
+		return rawLine{path: line}
+	}
+	return rawLine{
+		path:    line[:idx],
+		typeVal: line[idx+1 : len(line)-1],
+	}
+}
+
+var namedTypePattern = regexp.MustCompile(`^[A-Z][a-zA-Z0-9-]*\d+$`)
+
+func isNamedType(typeVal string) bool {
+	return namedTypePattern.MatchString(typeVal)
+}
+
+// coalesceType is a named type from the streamer output.
+type coalesceType struct {
+	name    string   // original name, e.g., "Ability2"
+	path    string   // path where introduced
+	keys    []string // sorted direct field names
+	keysSig string   // comma-joined keys for comparison
+}
+
+func buildCoalesceTypes(lines []rawLine) []*coalesceType {
+	var types []*coalesceType
+
+	for i, line := range lines {
+		if !isNamedType(line.typeVal) {
+			continue
+		}
+
+		// Find the extent: until next type intro at the same path.
+		end := len(lines)
+		for j := i + 1; j < len(lines); j++ {
+			if isNamedType(lines[j].typeVal) && lines[j].path == line.path {
+				end = j
+				break
+			}
+		}
+
+		// Extract direct field names from lines[i+1:end].
+		prefix := coalesceFieldPrefix(line.path)
+		fieldNames := make(map[string]bool)
+		for j := i + 1; j < end; j++ {
+			p := lines[j].path
+			if !strings.HasPrefix(p, prefix) {
+				continue
+			}
+			remainder := p[len(prefix):]
+			// Direct field: no "." in the remainder.
+			if strings.Contains(remainder, ".") {
+				continue
+			}
+			// Strip array brackets to get field name.
+			bracketIdx := strings.Index(remainder, "[")
+			var fieldName string
+			if bracketIdx < 0 {
+				fieldName = remainder
+			} else {
+				fieldName = remainder[:bracketIdx]
+			}
+			if fieldName != "" {
+				fieldNames[fieldName] = true
+			}
+		}
+
+		keys := make([]string, 0, len(fieldNames))
+		for k := range fieldNames {
+			keys = append(keys, k)
+		}
+		sort.Strings(keys)
+
+		types = append(types, &coalesceType{
+			name:    line.typeVal,
+			path:    line.path,
+			keys:    keys,
+			keysSig: strings.Join(keys, ","),
+		})
+	}
+
+	return types
+}
+
+// keySetGroup holds all types that share the same key set.
+type keySetGroup struct {
+	keysSig string
+	keys    []string
+	members []*coalesceType
+}
+
+func groupTypesByKeySet(types []*coalesceType) []*keySetGroup {
+	groupMap := make(map[string]*keySetGroup)
+	var groups []*keySetGroup
+
+	for _, t := range types {
+		if g, ok := groupMap[t.keysSig]; ok {
+			g.members = append(g.members, t)
+		} else {
+			g = &keySetGroup{
+				keysSig: t.keysSig,
+				keys:    t.keys,
+				members: []*coalesceType{t},
+			}
+			groupMap[t.keysSig] = g
+			groups = append(groups, g)
+		}
+	}
+
+	return groups
+}
+
+func assignCanonicalNames(groups []*keySetGroup) map[string]string {
+	nameMap := make(map[string]string) // old name → new name
+
+	// Pick a base name for each group.
+	type groupName struct {
+		group    *keySetGroup
+		baseName string
+	}
+	var gns []groupName
+	baseNameCount := make(map[string]int)
+	for _, g := range groups {
+		name := pickGroupName(g)
+		gns = append(gns, groupName{g, name})
+		baseNameCount[name]++
+	}
+
+	// Assign final names, numbering when base names conflict.
+	baseNameCounter := make(map[string]int)
+	for _, gn := range gns {
+		var finalName string
+		if baseNameCount[gn.baseName] == 1 {
+			finalName = gn.baseName
+		} else {
+			idx := baseNameCounter[gn.baseName]
+			baseNameCounter[gn.baseName]++
+			finalName = fmt.Sprintf("%s%d", gn.baseName, idx)
+		}
+		for _, t := range gn.group.members {
+			nameMap[t.name] = finalName
+		}
+	}
+
+	return nameMap
+}
+
+// pickGroupName selects the best canonical name for a key-set group.
+func pickGroupName(g *keySetGroup) string {
+	// Count distinct base names among members.
+	baseNames := make(map[string]int)
+	var firstBase string
+	for _, t := range g.members {
+		base := stripTrailingDigits(t.name)
+		baseNames[base]++
+		if firstBase == "" {
+			firstBase = base
+		}
+	}
+
+	// If many distinct names share this shape, derive from keys
+	// (e.g., {name, url} → "NameUrl").
+	if len(baseNames) > 2 {
+		return deriveNameFromKeys(g.keys)
+	}
+
+	// If one base name dominates, use it.
+	if len(baseNames) == 1 {
+		return firstBase
+	}
+
+	// Two distinct names — pick the most common, then shortest.
+	var best string
+	var bestCount int
+	for name, count := range baseNames {
+		if count > bestCount {
+			best = name
+			bestCount = count
+		} else if count == bestCount && len(name) < len(best) {
+			best = name
+		}
+	}
+	return best
+}
+
+func deriveNameFromKeys(keys []string) string {
+	if len(keys) == 0 {
+		return "Unknown"
+	}
+	// Use all keys (up to 3) to form the name.
+	n := len(keys)
+	if n > 3 {
+		n = 3
+	}
+	var parts []string
+	for _, k := range keys[:n] {
+		parts = append(parts, snakeToPascal(k))
+	}
+	return strings.Join(parts, "")
+}
+
+// stripTrailingDigits removes trailing digits from a type name.
+func stripTrailingDigits(name string) string {
+	i := len(name) - 1
+	for i >= 0 && name[i] >= '0' && name[i] <= '9' {
+		i--
+	}
+	if i < 0 {
+		return name
+	}
+	return name[:i+1]
+}
+
+// detectNullablePaths returns the set of paths where {null} appears
+// alongside at least one concrete type.
+func detectNullablePaths(lines []rawLine) map[string]bool {
+	pathTypes := make(map[string][]string)
+	for _, line := range lines {
+		pathTypes[line.path] = append(pathTypes[line.path], line.typeVal)
+	}
+
+	nullables := make(map[string]bool)
+	for path, typeVals := range pathTypes {
+		hasNull := false
+		hasOther := false
+		for _, tv := range typeVals {
+			if tv == "null" {
+				hasNull = true
+			} else if tv != "undefined" && tv != "empty" {
+				hasOther = true
+			}
+		}
+		if hasNull && hasOther {
+			nullables[path] = true
+		}
+	}
+
+	return nullables
+}
+
+func emitCoalesced(lines []rawLine, nameMap map[string]string, nullables map[string]bool) []string {
+	emittedTypes := make(map[string]bool) // canonical name → already expanded
+	var result []string
+
+	i := 0
+	for i < len(lines) {
+		line := lines[i]
+
+		// Skip standalone {null} lines at nullable paths (merged into ?).
+		if line.typeVal == "null" && nullables[line.path] {
+			i++
+			continue
+		}
+
+		if isNamedType(line.typeVal) {
+			canonicalName := nameMap[line.typeVal]
+			nullable := ""
+			if nullables[line.path] {
+				nullable = "?"
+			}
+
+			if emittedTypes[canonicalName] {
+				// Emit reference only, skip field lines.
+				result = append(result, line.path+"{"+canonicalName+nullable+"}")
+				i++
+				prefix := coalesceFieldPrefix(line.path)
+				for i < len(lines) && strings.HasPrefix(lines[i].path, prefix) {
+					i++
+				}
+				continue
+			}
+
+			// First occurrence — emit type intro and let fields be emitted normally.
+			emittedTypes[canonicalName] = true
+			result = append(result, line.path+"{"+canonicalName+nullable+"}")
+			i++
+			continue
+		}
+
+		// Regular line (primitive, null, undefined, empty, etc.).
+		tv := line.typeVal
+		nullable := ""
+		if nullables[line.path] && tv != "undefined" && tv != "empty" {
+			nullable = "?"
+		}
+
+		result = append(result, line.path+"{"+tv+nullable+"}")
+		i++
+	}
+
+	return result
+}
+
+func coalesceFieldPrefix(path string) string {
+	if path == "" {
+		return "."
+	}
+	return path + "."
+}
diff --git a/tools/jsontypes/coalesce_test.go b/tools/jsontypes/coalesce_test.go
new file mode 100644
index 0000000..765ffb3
--- /dev/null
+++ b/tools/jsontypes/coalesce_test.go
@@ -0,0 +1,248 @@
+package jsontypes
+
+import (
+	"encoding/json"
+	"os"
+	"strings"
+	"testing"
+)
+
+func TestCoalesceBasic(t *testing.T) {
+	input := []string{
+		"{Root0}",
+		".name{string}",
+		".email{string}",
+		".email{null}",
+		".friends[]{FriendsItem1}",
+		".friends[].name{string}",
+		".friends[].age{int}",
+	}
+
+	got := Coalesce(input)
+
+	// Root0 → Root (unique)
+	assertLineContains(t, got, "{Root}")
+	// email should be {string?} (nullable)
+	assertLineContains(t, got, ".email{string?}")
+	// standalone {null} should be removed
+	assertLineNotContains(t, got, ".email{null}")
+	// FriendsItem1 → FriendsItem (unique)
+	assertLineContains(t, got, ".friends[]{FriendsItem}")
+
+	t.Logf("output (%d lines):\n%s", len(got), strings.Join(got, "\n"))
+}
+
+func TestCoalesceDedup(t *testing.T) {
+	// Two types with the same key set at different paths.
+	input := []string{
+		"{Root0}",
+		".author{Author1}",
+		".author.name{string}",
+		".author.url{string}",
+		".reviewer{Reviewer2}",
+		".reviewer.name{string}",
+		".reviewer.url{string}",
+	}
+
+	got := Coalesce(input)
+
+	// Both Author and Reviewer have {name, url} → merged into one type.
+	// Only one should have fields expanded.
+	nameUrlCount := 0
+	fieldCount := 0
+	for _, line := range got {
+		if strings.Contains(line, ".name{string}") {
+			fieldCount++
+		}
+		if strings.Contains(line, "{Author}") || strings.Contains(line, "{Reviewer}") {
+			nameUrlCount++
+		}
+	}
+
+	// The canonical name should appear at both paths.
+	if nameUrlCount != 2 {
+		t.Errorf("expected 2 references to canonical type, got %d", nameUrlCount)
+	}
+
+	// But fields should only appear once (for the first occurrence).
+	if fieldCount != 1 {
+		t.Errorf("expected 1 field expansion, got %d", fieldCount)
+	}
+
+	t.Logf("output (%d lines):\n%s", len(got), strings.Join(got, "\n"))
+}
+
+func TestCoalesceDedupManyNames(t *testing.T) {
+	// >2 distinct base names with same key set → derive name from keys.
+	input := []string{
+		"{Root0}",
+		".ability{Ability1}",
+		".ability.name{string}",
+		".ability.url{string}",
+		".version{Version2}",
+		".version.name{string}",
+		".version.url{string}",
+		".species{Species3}",
+		".species.name{string}",
+		".species.url{string}",
+	}
+
+	got := Coalesce(input)
+
+	// 3 distinct base names → derived name "NameUrl".
+	assertLineContains(t, got, "{NameUrl}")
+	// Fields expanded only once.
+	nameCount := 0
+	for _, line := range got {
+		if strings.HasSuffix(line, ".name{string}") {
+			nameCount++
+		}
+	}
+	if nameCount != 1 {
+		t.Errorf("expected 1 .name{string} (deduped), got %d", nameCount)
+	}
+
+	t.Logf("output (%d lines):\n%s", len(got), strings.Join(got, "\n"))
+}
+
+func TestCoalesceUndefinedPreserved(t *testing.T) {
+	input := []string{
+		"[]{RootItem0}",
+		"[].name{string}",
+		"[].email{undefined}",
+		"[].email{string}",
+	}
+
+	got := Coalesce(input)
+
+	// {undefined} should be preserved (not collapsed into ?).
+	assertLineContains(t, got, "[].email{undefined}")
+	// {string} should NOT be nullable (undefined ≠ null).
+	assertLineContains(t, got, "[].email{string}")
+	assertLineNotContains(t, got, "[].email{string?}")
+
+	t.Logf("output (%d lines):\n%s", len(got), strings.Join(got, "\n"))
+}
+
+func TestCoalesceNullAndUndefined(t *testing.T) {
+	input := []string{
+		"[]{RootItem0}",
+		"[].email{undefined}",
+		"[].email{null}",
+		"[].email{string}",
+	}
+
+	got := Coalesce(input)
+
+	// {undefined} preserved
+	assertLineContains(t, got, "[].email{undefined}")
+	// {null} collapsed into {string?}
+	assertLineContains(t, got, "[].email{string?}")
+	assertLineNotContains(t, got, "[].email{null}")
+
+	t.Logf("output (%d lines):\n%s", len(got), strings.Join(got, "\n"))
+}
+
+func TestCoalescePokemonFixture(t *testing.T) {
+	lines := rawPathsFromFixture(t, "testdata/pokemon-pikachu.json")
+	got := Coalesce(lines)
+
+	// Should be significantly fewer lines than the raw output.
+	if len(got) >= len(lines) {
+		t.Errorf("coalesced output (%d lines) should be fewer than raw (%d lines)",
+			len(got), len(lines))
+	}
+
+	// {name, url} types should be merged (many types share this shape).
+	assertLineContains(t, got, "{NameUrl}")
+
+	// Root should lose its number.
+	assertLineContains(t, got, "{Root}")
+
+	// Core structure preserved.
+	assertLineContains(t, got, ".abilities[]{AbilitiesItem}")
+	assertLineContains(t, got, ".stats[]{StatsItem}")
+	assertLineContains(t, got, ".types[]{TypesItem}")
+
+	// Nullable fields: dream_world.front_female is null in some contexts.
+	// (In single-document mode, null-only fields stay as {null}.)
+
+	t.Logf("raw: %d lines → coalesced: %d lines", len(lines), len(got))
+	t.Logf("output:\n%s", strings.Join(got, "\n"))
+}
+
+func TestCoalesceSWAPIPlanets(t *testing.T) {
+	lines := rawPathsFromFixture(t, "testdata/swapi-planets.json")
+	got := Coalesce(lines)
+
+	// Root should be clean.
+	assertLineContains(t, got, "{Root}")
+	assertLineContains(t, got, ".results[]{ResultsItem}")
+
+	// No nullable fields in this fixture (previous is null but standalone).
+	assertLineContains(t, got, ".previous{null}")
+
+	t.Logf("raw: %d lines → coalesced: %d lines", len(lines), len(got))
+	t.Logf("output:\n%s", strings.Join(got, "\n"))
+}
+
+func TestCoalescePerBaseNameNumbering(t *testing.T) {
+	// Two types with the same base name but different key sets.
+	input := []string{
+		"{Root0}",
+		".config{Settings1}",
+		".config.host{string}",
+		".config.port{int}",
+		".prefs{Settings2}",
+		".prefs.theme{string}",
+		".prefs.lang{string}",
+		".prefs.debug{bool}",
+	}
+
+	got := Coalesce(input)
+
+	// Both are "Settings" base but different keys → numbered.
+	assertLineContains(t, got, ".config{Settings0}")
+	assertLineContains(t, got, ".prefs{Settings1}")
+
+	t.Logf("output (%d lines):\n%s", len(got), strings.Join(got, "\n"))
+}
+
+// Helper: run RawPaths on a fixture file and return the lines.
+func rawPathsFromFixture(t *testing.T, path string) []string {
+	t.Helper()
+	f, err := os.Open(path)
+	if err != nil {
+		t.Fatalf("open %s: %v", path, err)
+	}
+	defer f.Close()
+
+	var data any
+	dec := json.NewDecoder(f)
+	dec.UseNumber()
+	if err := dec.Decode(&data); err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+
+	return RawPaths(data)
+}
+
+func assertLineContains(t *testing.T, lines []string, want string) {
+	t.Helper()
+	for _, line := range lines {
+		if strings.Contains(line, want) {
+			return
+		}
+	}
+	t.Errorf("missing expected line containing %q", want)
+}
+
+func assertLineNotContains(t *testing.T, lines []string, want string) {
+	t.Helper()
+	for _, line := range lines {
+		if line == want {
+			t.Errorf("unexpected line %q", want)
+			return
+		}
+	}
+}