diff --git a/script/process-json-numbers.c b/script/process-json-numbers.c
new file mode 100644
index 0000000..681ffe8
--- /dev/null
+++ b/script/process-json-numbers.c
@@ -0,0 +1,299 @@
+// Copyright 2020 The Wuffs Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// ----------------
+
+// process-json-numbers.c processes all the numbers in the JSON-formatted data
+// read from stdin. It succeeds (with exit code 0) if the input is valid JSON
+// and all of the numbers within were processed without error.
+//
+// Without further flags, processing is a no-op and the program only verifies
+// the JSON structure.
+//
+// Pass -e (--emit-number-str) to emit each number (as a string) on its own
+// line.
+//
+// Pass -p (--parse-number-f64) to call wuffs_base__parse_number_f64 on each
+// number. Timing this program with and without this flag gives a rough measure
+// of how much time is spent solely in wuffs_base__parse_number_f64.
+//
+// Pass -r (--render-number-f64) to call wuffs_base__render_number_f64 (with
+// WUFFS_BASE__RENDER_NUMBER_FXX__JUST_ENOUGH_PRECISION) on each number. Timing
+// this program with and without this flag gives a rough measure of how much
+// time is spent solely in wuffs_base__render_number_f64.
+//
+// The -r flag is ignored unless -p is also passed.
+//
+// This program's purpose is to benchmark the wuffs_base__etc_f64 functions.
+// It's not about JSON per se, but JSON files are a source of realistic
+// floating point numbers.
+
+#include <inttypes.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+
+// Wuffs ships as a "single file C library" or "header file library" as per
+// https://github.com/nothings/stb/blob/master/docs/stb_howto.txt
+//
+// To use that single file as a "foo.c"-like implementation, instead of a
+// "foo.h"-like header, #define WUFFS_IMPLEMENTATION before #include'ing or
+// compiling it.
+#define WUFFS_IMPLEMENTATION
+
+// Defining the WUFFS_CONFIG__MODULE* macros are optional, but it lets users of
+// release/c/etc.c whitelist which parts of Wuffs to build. That file contains
+// the entire Wuffs standard library, implementing a variety of codecs and file
+// formats. Without this macro definition, an optimizing compiler or linker may
+// very well discard Wuffs code for unused codecs, but listing the Wuffs
+// modules we use makes that process explicit. Preprocessing means that such
+// code simply isn't compiled.
+#define WUFFS_CONFIG__MODULES
+#define WUFFS_CONFIG__MODULE__BASE
+#define WUFFS_CONFIG__MODULE__JSON
+
+// If building this program in an environment that doesn't easily accommodate
+// relative includes, you can use the script/inline-c-relative-includes.go
+// program to generate a stand-alone C++ file.
+#include "../release/c/wuffs-unsupported-snapshot.c"
+
+// Wuffs allows either statically or dynamically allocated work buffers. This
+// program exercises static allocation.
+#define WORK_BUFFER_ARRAY_SIZE \
+  WUFFS_JSON__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE
+#if WORK_BUFFER_ARRAY_SIZE > 0
+uint8_t g_work_buffer_array[WORK_BUFFER_ARRAY_SIZE];
+#else
+// Not all C/C++ compilers support 0-length arrays.
+uint8_t g_work_buffer_array[1];
+#endif
+
+#ifndef SRC_BUFFER_ARRAY_SIZE
+#define SRC_BUFFER_ARRAY_SIZE (64 * 1024 * 1024)
+#endif
+#ifndef TOKEN_BUFFER_ARRAY_SIZE
+#define TOKEN_BUFFER_ARRAY_SIZE (128 * 1024)
+#endif
+
+uint8_t g_src_buffer_array[SRC_BUFFER_ARRAY_SIZE];
+wuffs_base__token g_tok_buffer_array[TOKEN_BUFFER_ARRAY_SIZE];
+
+wuffs_base__io_buffer g_src;
+wuffs_base__token_buffer g_tok;
+
+wuffs_json__decoder g_dec;
+
+#define TRY(error_msg)         \
+  do {                         \
+    const char* z = error_msg; \
+    if (z) {                   \
+      return z;                \
+    }                          \
+  } while (false)
+
+// ignore_return_value suppresses errors from -Wall -Werror.
+static void  //
+ignore_return_value(int ignored) {}
+
+const char*  //
+read_src() {
+  if (g_src.meta.closed) {
+    return "main: internal error: read requested on a closed source";
+  }
+  wuffs_base__io_buffer__compact(&g_src);
+  if (g_src.meta.wi >= g_src.data.len) {
+    return "main: g_src buffer is full";
+  }
+  size_t n = fread(g_src.data.ptr + g_src.meta.wi, sizeof(uint8_t),
+                   g_src.data.len - g_src.meta.wi, stdin);
+  g_src.meta.wi += n;
+  g_src.meta.closed = feof(stdin);
+  if ((n == 0) && !g_src.meta.closed) {
+    return "main: read error";
+  }
+  return NULL;
+}
+
+// ----
+
+struct {
+  int remaining_argc;
+  char** remaining_argv;
+
+  bool emit_number_str;
+  bool parse_number_f64;
+  bool render_number_f64;
+} g_flags = {0};
+
+const char*  //
+parse_flags(int argc, char** argv) {
+  int c = (argc > 0) ? 1 : 0;  // Skip argv[0], the program name.
+  for (; c < argc; c++) {
+    char* arg = argv[c];
+    if (*arg++ != '-') {
+      break;
+    }
+
+    // A double-dash "--foo" is equivalent to a single-dash "-foo". As special
+    // cases, a bare "-" is not a flag (some programs may interpret it as
+    // stdin) and a bare "--" means to stop parsing flags.
+    if (*arg == '\x00') {
+      break;
+    } else if (*arg == '-') {
+      arg++;
+      if (*arg == '\x00') {
+        c++;
+        break;
+      }
+    }
+
+    if (!strcmp(arg, "e") || !strcmp(arg, "emit-number-str")) {
+      g_flags.emit_number_str = true;
+      continue;
+    }
+    if (!strcmp(arg, "p") || !strcmp(arg, "parse-number-f64")) {
+      g_flags.parse_number_f64 = true;
+      continue;
+    }
+    if (!strcmp(arg, "r") || !strcmp(arg, "render-number-f64")) {
+      g_flags.render_number_f64 = true;
+      continue;
+    }
+
+    return "main: unrecognized flag argument";
+  }
+
+  g_flags.remaining_argc = argc - c;
+  g_flags.remaining_argv = argv + c;
+  return NULL;
+}
+
+const char*  //
+main1(int argc, char** argv) {
+  TRY(parse_flags(argc, argv));
+  if (g_flags.remaining_argc > 0) {
+    return "main: bad argument: use \"program < input\", not \"program input\"";
+  }
+
+  uint8_t new_line[1];
+  new_line[0] = '\n';
+
+  g_src = wuffs_base__make_io_buffer(
+      wuffs_base__make_slice_u8(g_src_buffer_array, SRC_BUFFER_ARRAY_SIZE),
+      wuffs_base__empty_io_buffer_meta());
+
+  g_tok = wuffs_base__make_token_buffer(
+      wuffs_base__make_slice_token(g_tok_buffer_array, TOKEN_BUFFER_ARRAY_SIZE),
+      wuffs_base__empty_token_buffer_meta());
+
+  wuffs_base__status init_status = wuffs_json__decoder__initialize(
+      &g_dec, sizeof__wuffs_json__decoder(), WUFFS_VERSION, 0);
+  if (!wuffs_base__status__is_ok(&init_status)) {
+    return wuffs_base__status__message(&init_status);
+  }
+
+  uint64_t pos = 0;
+  while (true) {
+    wuffs_base__status status = wuffs_json__decoder__decode_tokens(
+        &g_dec, &g_tok, &g_src,
+        wuffs_base__make_slice_u8(g_work_buffer_array, WORK_BUFFER_ARRAY_SIZE));
+
+    while (g_tok.meta.ri < g_tok.meta.wi) {
+      wuffs_base__token* t = &g_tok.data.ptr[g_tok.meta.ri++];
+      uint64_t len = wuffs_base__token__length(t);
+
+      if (wuffs_base__token__value_base_category(t) ==
+          WUFFS_BASE__TOKEN__VBC__NUMBER) {
+        uint64_t buf_pos = pos - g_src.meta.pos;
+        uint64_t buf_len = g_src.data.len;
+        if ((buf_len < buf_pos) || ((buf_len - buf_pos) < len)) {
+          return "main: internal error: inconsistent token position/length";
+        }
+
+        if (g_flags.emit_number_str) {
+          const int stdout_fd = 1;
+          ignore_return_value(write(stdout_fd, &g_src.data.ptr[buf_pos], len));
+          ignore_return_value(write(stdout_fd, &new_line[0], 1));
+        }
+
+        if (g_flags.parse_number_f64) {
+          wuffs_base__result_f64 r = wuffs_base__parse_number_f64(
+              wuffs_base__make_slice_u8(&g_src.data.ptr[buf_pos], len));
+          if (!wuffs_base__status__is_ok(&r.status)) {
+            return wuffs_base__status__message(&r.status);
+          }
+
+          if (g_flags.render_number_f64) {
+            uint8_t render_buffer[2048];
+            size_t n = wuffs_base__render_number_f64(
+                wuffs_base__make_slice_u8(&render_buffer[0], 2048), r.value, 0,
+                WUFFS_BASE__RENDER_NUMBER_FXX__JUST_ENOUGH_PRECISION);
+            if (n == 0) {
+              return "main: internal error: couldn't render_number_f64";
+            }
+          }
+        }
+      }
+
+      pos += len;
+      if (0 > ((int64_t)pos)) {
+        return "main: input is too long";
+      }
+    }
+
+    if (status.repr == NULL) {
+      return NULL;
+    } else if (status.repr == wuffs_base__suspension__short_read) {
+      TRY(read_src());
+    } else if (status.repr == wuffs_base__suspension__short_write) {
+      wuffs_base__token_buffer__compact(&g_tok);
+    } else {
+      return wuffs_base__status__message(&status);
+    }
+  }
+}
+
+// ----
+
+int  //
+compute_exit_code(const char* status_msg) {
+  if (!status_msg) {
+    return 0;
+  }
+  size_t n = strnlen(status_msg, 2047);
+  if (n >= 2047) {
+    status_msg = "main: internal error: error message is too long";
+    n = strnlen(status_msg, 2047);
+  }
+  fprintf(stderr, "%s\n", status_msg);
+  // Return an exit code of 1 for regular (forseen) errors, e.g. badly
+  // formatted or unsupported input.
+  //
+  // Return an exit code of 2 for internal (exceptional) errors, e.g. defensive
+  // run-time checks found that an internal invariant did not hold.
+  //
+  // Automated testing, including badly formatted inputs, can therefore
+  // discriminate between expected failure (exit code 1) and unexpected failure
+  // (other non-zero exit codes). Specifically, exit code 2 for internal
+  // invariant violation, exit code 139 (which is 128 + SIGSEGV on x86_64
+  // linux) for a segmentation fault (e.g. null pointer dereference).
+  return strstr(status_msg, "internal error:") ? 2 : 1;
+}
+
+int  //
+main(int argc, char** argv) {
+  const char* z = main1(argc, argv);
+  int exit_code = compute_exit_code(z);
+  return exit_code;
+}