feat(zed): recover corrupted frame gaps in MCAP export

Make ZED MCAP export skip corrupted frame runs until recovery and

treat unreadable tail frames as end-of-stream instead of hard

failing conversion.

Update bundled nearest-mode export to emit partial bundles during

corruption gaps, extend BundleManifest with explicit member status

and skipped-frame counts, and only write payload messages for

present cameras.

Tighten batch probing so bundled MCAP validation checks /bundle

coverage and per-camera message counts, and improve failure

excerpts to include stderr tail output.

Also add a local cppzmq CMake fallback, refresh the multi-record

tester for the new bundle schema, and document the mixed NVENC

limitations in the README.

CMakeLists.txt

@@ -28,6 +28,14 @@ if (NOT TARGET OpenSSL::Crypto AND DEFINED OPENSSL_CRYPTO_LIBRARY)
 		INTERFACE_INCLUDE_DIRECTORIES "${OPENSSL_INCLUDE_DIR}")
 endif()
 find_package(cppzmq QUIET)
+set(CPPZMQ_LOCAL_ROOT "${CMAKE_CURRENT_LIST_DIR}/../cppzmq" CACHE PATH "Path to a local cppzmq checkout")
+if (NOT TARGET cppzmq::cppzmq AND NOT TARGET cppzmq)
+	if (EXISTS "${CPPZMQ_LOCAL_ROOT}/zmq.hpp")
+		add_library(cppzmq::cppzmq INTERFACE IMPORTED GLOBAL)
+		set_target_properties(cppzmq::cppzmq PROPERTIES
+			INTERFACE_INCLUDE_DIRECTORIES "${CPPZMQ_LOCAL_ROOT}")
+	endif()
+endif()
 if (DEFINED CVMMAP_STREAMER_USE_SYSTEM_CNATS)
 	message(FATAL_ERROR
 		"CVMMAP_STREAMER_USE_SYSTEM_CNATS was removed; use CVMMAP_CNATS_PROVIDER=system")
@@ -161,8 +169,10 @@ protobuf_generate(
 	TARGET cvmmap_streamer_depth_proto
 	LANGUAGE cpp
 	PROTOC_OUT_DIR "${CMAKE_CURRENT_BINARY_DIR}"
-	PROTOS "${CMAKE_CURRENT_LIST_DIR}/proto/cvmmap_streamer/DepthMap.proto"
-	IMPORT_DIRS "${CMAKE_CURRENT_LIST_DIR}")
+	PROTOS
+		"${CMAKE_CURRENT_LIST_DIR}/proto/cvmmap_streamer/DepthMap.proto"
+		"${CMAKE_CURRENT_LIST_DIR}/proto/cvmmap_streamer/BundleManifest.proto"
+	IMPORT_DIRS "${CMAKE_CURRENT_LIST_DIR}/proto")
 add_library(cvmmap_streamer_protobuf INTERFACE)
 target_include_directories(cvmmap_streamer_foxglove_proto
 	PUBLIC

README.md

@@ -265,9 +265,11 @@ uv run python scripts/zed_batch_svo_to_mcap.py \
 
 The batch MCAP wrapper writes `<segment>/<segment>.mcap` by default, skips existing outputs unless told otherwise, and returns a nonzero exit code if any segment fails.
 The repo includes a minimal pose config at `config/zed_pose_config.toml` so MCAP conversion does not depend on a separate `cv-mmap` checkout.
-In bundled multi-camera mode, `--start-frame` and `--end-frame` mean the first and last emitted synced frame-group indices from the common start timestamp, inclusive.
+In bundled multi-camera mode, `--start-frame` and `--end-frame` mean the first and last emitted bundle indices from the common start timestamp, inclusive.
 When stderr is attached to a TTY, `zed_batch_svo_to_mcap.py` uses a `progress-table` view by default; otherwise it emits line-oriented start/completion/failure logs plus periodic heartbeat summaries. Use `--progress-ui table` or `--progress-ui text` to override the automatic mode selection.
 
+Bundled MCAP export now defaults to `--bundle-policy nearest`. That mode emits one `/bundle` manifest message per bundle timestamp on the common timeline and keeps the original per-camera timestamps on `/zedN/video`, `/zedN/depth`, and optional `/zedN/pose`. Consumers that care about grouping should follow `/bundle` instead of inferring bundle membership from identical message timestamps. Use `--bundle-policy strict` when you want the older thresholded sync behavior, and `--sync-tolerance-ms` only applies in that strict mode.
+
 ### Why Mixed Hardware/Software Mode Exists
 
 Bundled MCAP export opens one video encoder per camera stream. A four-camera segment therefore consumes four H.264/H.265 encoder sessions at once.

include/cvmmap_streamer/record/mcap_record_sink.hpp

@@ -8,6 +8,7 @@
 #include <cstddef>
 #include <cstdint>
 #include <expected>
+#include <optional>
 #include <span>
 #include <string>
 #include <string_view>
@@ -19,6 +20,16 @@ enum class DepthEncoding {
 	RvlF32,
 };
 
+enum class BundlePolicy {
+	Nearest,
+	Strict,
+};
+
+enum class BundleMemberStatus {
+	Present,
+	CorruptedGap,
+};
+
 struct RawDepthMapView {
 	std::uint64_t timestamp_ns{0};
 	std::uint32_t width{0};
@@ -57,6 +68,21 @@ struct RawBodyTrackingMessageView {
 	std::span<const std::uint8_t> bytes{};
 };
 
+struct RawBundleMemberView {
+	std::string_view camera_label{};
+	BundleMemberStatus status{BundleMemberStatus::Present};
+	std::optional<std::uint64_t> timestamp_ns{};
+	std::int64_t delta_ns{0};
+	std::uint32_t corrupted_frames_skipped{0};
+};
+
+struct RawBundleManifestView {
+	std::uint64_t timestamp_ns{0};
+	std::uint64_t bundle_index{0};
+	BundlePolicy policy{BundlePolicy::Nearest};
+	std::span<const RawBundleMemberView> members{};
+};
+
 struct McapRecordStreamConfig {
 	std::string topic{"/camera/video"};
 	std::string depth_topic{"/camera/depth"};
@@ -137,7 +163,8 @@ public:
 	[[nodiscard]]
 	static std::expected<MultiMcapRecordSink, std::string> create(
 		std::string path,
-		McapCompression compression);
+		McapCompression compression,
+		std::string bundle_topic = "/bundle");
 
 	[[nodiscard]]
 	std::expected<StreamId, std::string> add_stream(
@@ -179,6 +206,10 @@ public:
 		StreamId stream_id,
 		const RawPoseView &pose);
 
+	[[nodiscard]]
+	std::expected<void, std::string> write_bundle_manifest(
+		const RawBundleManifestView &bundle);
+
 	[[nodiscard]]
 	std::expected<void, std::string> write_body_tracking_message(
 		StreamId stream_id,

proto/cvmmap_streamer/BundleManifest.proto

@@ -0,0 +1,32 @@
+syntax = "proto3";
+
+package cvmmap_streamer;
+
+import "google/protobuf/timestamp.proto";
+
+message BundleManifest {
+  enum BundlePolicy {
+    BUNDLE_POLICY_UNKNOWN = 0;
+    BUNDLE_POLICY_NEAREST = 1;
+    BUNDLE_POLICY_STRICT = 2;
+  }
+
+  enum BundleMemberStatus {
+    BUNDLE_MEMBER_STATUS_UNKNOWN = 0;
+    BUNDLE_MEMBER_STATUS_PRESENT = 1;
+    BUNDLE_MEMBER_STATUS_CORRUPTED_GAP = 2;
+  }
+
+  message BundleMember {
+    string camera_label = 1;
+    google.protobuf.Timestamp timestamp = 2;
+    sint64 delta_ns = 3;
+    BundleMemberStatus status = 4;
+    uint32 corrupted_frames_skipped = 5;
+  }
+
+  google.protobuf.Timestamp timestamp = 1;
+  uint64 bundle_index = 2;
+  BundlePolicy policy = 3;
+  repeated BundleMember members = 4;
+}

scripts/zed_batch_svo_to_mcap.py

@@ -10,6 +10,7 @@ import re
 import subprocess
 import sys
 import time
+from collections import Counter
 from dataclasses import dataclass
 from pathlib import Path
 
@@ -34,6 +35,8 @@ class BatchConfig:
     mcap_compression: str
     depth_mode: str
     depth_size: str
+    bundle_policy: str
+    bundle_topic: str | None
     with_pose: bool
     pose_config: Path | None
     world_frame_id: str | None
@@ -102,6 +105,7 @@ class ActiveJobState:
 
 
 _MCAP_READER_MODULE = None
+_BUNDLE_MANIFEST_CLASS_CACHE: dict[bytes, tuple[object, int | None]] = {}
 TABLE_REFRESH_SECONDS = 1.0
 TEXT_HEARTBEAT_SECONDS = 30.0
 
@@ -500,7 +504,11 @@ def command_for_job(job: ConversionJob, config: BatchConfig, encoder_device: str
         config.depth_mode,
         "--depth-size",
         config.depth_size,
+        "--bundle-policy",
+        config.bundle_policy,
     ]
+    if config.bundle_topic:
+        command.extend(["--bundle-topic", config.bundle_topic])
     if config.with_pose:
         command.append("--with-pose")
     if config.pose_config is not None:
@@ -569,22 +577,131 @@ def required_topics_for(camera_labels: tuple[str, ...]) -> set[str]:
     return topics
 
 
-def probe_output(output_path: Path, camera_labels: tuple[str, ...]) -> OutputProbeResult:
+def load_bundle_manifest_type(schema_data: bytes) -> tuple[object, int | None]:
+    cached = _BUNDLE_MANIFEST_CLASS_CACHE.get(schema_data)
+    if cached is not None:
+        return cached
+
+    from google.protobuf import descriptor_pb2, descriptor_pool, message_factory, timestamp_pb2
+
+    descriptor_set = descriptor_pb2.FileDescriptorSet()
+    descriptor_set.ParseFromString(schema_data)
+    pool = descriptor_pool.DescriptorPool()
+    has_embedded_timestamp = any(
+        file_descriptor.name == "google/protobuf/timestamp.proto"
+        for file_descriptor in descriptor_set.file
+    )
+    if has_embedded_timestamp:
+        for file_descriptor in descriptor_set.file:
+            if file_descriptor.name == "google/protobuf/timestamp.proto":
+                pool.Add(file_descriptor)
+                break
+    else:
+        pool.AddSerializedFile(timestamp_pb2.DESCRIPTOR.serialized_pb)
+    for file_descriptor in descriptor_set.file:
+        if file_descriptor.name == "google/protobuf/timestamp.proto":
+            continue
+        pool.Add(file_descriptor)
+    message_descriptor = pool.FindMessageTypeByName("cvmmap_streamer.BundleManifest")
+    message_class = message_factory.GetMessageClass(message_descriptor)
+    present_value = None
+    if "BundleMemberStatus" in message_descriptor.enum_types_by_name:
+        status_enum = message_descriptor.enum_types_by_name["BundleMemberStatus"]
+        present_value = status_enum.values_by_name["BUNDLE_MEMBER_STATUS_PRESENT"].number
+    _BUNDLE_MANIFEST_CLASS_CACHE[schema_data] = (message_class, present_value)
+    return message_class, present_value
+
+
+def probe_output(
+    output_path: Path,
+    camera_labels: tuple[str, ...],
+    *,
+    bundle_topic: str | None,
+) -> OutputProbeResult:
     if not output_path.is_file():
         return OutputProbeResult(output_path=output_path, status="missing")
 
     reader_module = load_mcap_reader()
     expected_topics = required_topics_for(camera_labels)
+    require_bundle = len(camera_labels) > 1 and bool(bundle_topic)
+    if require_bundle:
+        expected_topics.add(bundle_topic or "/bundle")
     found_topics: set[str] = set()
+    video_counts: Counter[str] = Counter()
+    depth_counts: Counter[str] = Counter()
+    bundle_present_counts: Counter[str] = Counter()
+    expected_camera_labels = set(camera_labels)
 
     try:
         with output_path.open("rb") as stream:
             reader = reader_module.make_reader(stream)
-            for _schema, channel, _message in reader.iter_messages():
+            for schema, channel, message in reader.iter_messages():
                 if channel.topic in expected_topics:
                     found_topics.add(channel.topic)
-                if found_topics == expected_topics:
-                    return OutputProbeResult(output_path=output_path, status="valid")
+                if channel.topic.endswith("/video"):
+                    video_counts[channel.topic.removeprefix("/").removesuffix("/video")] += 1
+                    continue
+                if channel.topic.endswith("/depth"):
+                    depth_counts[channel.topic.removeprefix("/").removesuffix("/depth")] += 1
+                    continue
+                if require_bundle and channel.topic == bundle_topic:
+                    if schema is None or schema.name != "cvmmap_streamer.BundleManifest":
+                        return OutputProbeResult(
+                            output_path=output_path,
+                            status="invalid",
+                            reason=f"bundle topic '{bundle_topic}' is missing the BundleManifest schema",
+                        )
+                    try:
+                        bundle_class, present_value = load_bundle_manifest_type(schema.data)
+                        bundle = bundle_class()
+                        bundle.ParseFromString(message.data)
+                    except Exception as error:  # noqa: BLE001
+                        return OutputProbeResult(
+                            output_path=output_path,
+                            status="invalid",
+                            reason=f"failed to parse bundle manifest: {error}",
+                        )
+
+                    bundle_labels: set[str] = set()
+                    for member in bundle.members:
+                        label = str(member.camera_label)
+                        if label not in expected_camera_labels:
+                            return OutputProbeResult(
+                                output_path=output_path,
+                                status="invalid",
+                                reason=f"bundle manifest referenced unknown camera label '{label}'",
+                            )
+                        if label in bundle_labels:
+                            return OutputProbeResult(
+                                output_path=output_path,
+                                status="invalid",
+                                reason=f"bundle manifest duplicated camera label '{label}'",
+                            )
+                        bundle_labels.add(label)
+                        is_present = member.HasField("timestamp")
+                        if present_value is not None:
+                            is_present = member.status == present_value
+                            if is_present and not member.HasField("timestamp"):
+                                return OutputProbeResult(
+                                    output_path=output_path,
+                                    status="invalid",
+                                    reason=f"bundle member '{label}' is present but missing a timestamp",
+                                )
+                        if is_present:
+                            bundle_present_counts[label] += 1
+                    if bundle_labels != expected_camera_labels:
+                        missing_labels = sorted(expected_camera_labels - bundle_labels)
+                        extra_labels = sorted(bundle_labels - expected_camera_labels)
+                        details = []
+                        if missing_labels:
+                            details.append("missing=" + ",".join(missing_labels))
+                        if extra_labels:
+                            details.append("extra=" + ",".join(extra_labels))
+                        return OutputProbeResult(
+                            output_path=output_path,
+                            status="invalid",
+                            reason="bundle manifest camera coverage mismatch: " + " ".join(details),
+                        )
     except Exception as error:  # noqa: BLE001
         return OutputProbeResult(output_path=output_path, status="invalid", reason=str(error))
 
@@ -595,6 +712,27 @@ def probe_output(output_path: Path, camera_labels: tuple[str, ...]) -> OutputPro
             status="invalid",
             reason="missing expected topics: " + ", ".join(missing_topics),
         )
+    if require_bundle:
+        for label in camera_labels:
+            present_count = bundle_present_counts[label]
+            if video_counts[label] != present_count:
+                return OutputProbeResult(
+                    output_path=output_path,
+                    status="invalid",
+                    reason=(
+                        f"video count mismatch for {label}: "
+                        f"bundle_present={present_count} video_messages={video_counts[label]}"
+                    ),
+                )
+            if depth_counts[label] != present_count:
+                return OutputProbeResult(
+                    output_path=output_path,
+                    status="invalid",
+                    reason=(
+                        f"depth count mismatch for {label}: "
+                        f"bundle_present={present_count} depth_messages={depth_counts[label]}"
+                    ),
+                )
     return OutputProbeResult(output_path=output_path, status="valid")
 
 
@@ -635,8 +773,13 @@ def split_lines_for_excerpt(text: str, max_lines: int = 8) -> list[str]:
     lines = [line.rstrip() for line in text.splitlines() if line.strip()]
     if len(lines) <= max_lines:
         return lines
-    excerpt = lines[:max_lines]
-    excerpt.append(f"... ({len(lines) - max_lines} more lines)")
+    head_count = max(1, max_lines // 2)
+    tail_count = max_lines - head_count
+    excerpt = lines[:head_count]
+    omitted = len(lines) - head_count - tail_count
+    if omitted > 0:
+        excerpt.append(f"... ({omitted} omitted line(s))")
+    excerpt.extend(lines[-tail_count:])
     return excerpt
 
 
@@ -1011,6 +1154,19 @@ def build_worker_slots(
     default="optimal",
     show_default=True,
 )
+@click.option(
+    "--bundle-policy",
+    type=click.Choice(("nearest", "strict")),
+    default="nearest",
+    show_default=True,
+    help="Bundling policy for multi-camera MCAP export.",
+)
+@click.option(
+    "--bundle-topic",
+    default="/bundle",
+    show_default=True,
+    help="Topic used for bundled multi-camera manifest messages.",
+)
 @click.option("--with-pose", is_flag=True, help="Enable per-camera positional tracking export when available.")
 @click.option(
     "--pose-config",
@@ -1026,19 +1182,19 @@ def build_worker_slots(
     "--start-frame",
     type=click.IntRange(min=0),
     default=None,
-    help="First synced frame group to export (inclusive) in bundled multi-camera mode.",
+    help="First bundle index to export (inclusive) in bundled multi-camera mode.",
 )
 @click.option(
     "--end-frame",
     type=click.IntRange(min=0),
     default=None,
-    help="Last synced frame group to export (inclusive) in bundled multi-camera mode.",
+    help="Last bundle index to export (inclusive) in bundled multi-camera mode.",
 )
 @click.option(
     "--sync-tolerance-ms",
     type=click.FloatRange(min=0.0, min_open=True),
     default=None,
-    help="Override the maximum timestamp delta used for bundled multi-camera sync.",
+    help="Override the maximum timestamp delta used by strict bundled sync.",
 )
 @click.option(
     "--progress-ui",
@@ -1070,6 +1226,8 @@ def main(
     mcap_compression: str,
     depth_mode: str,
     depth_size: str,
+    bundle_policy: str,
+    bundle_topic: str,
     with_pose: bool,
     pose_config: Path | None,
     world_frame_id: str | None,
@@ -1102,6 +1260,8 @@ def main(
         mcap_compression=mcap_compression,
         depth_mode=depth_mode,
         depth_size=depth_size,
+        bundle_policy=bundle_policy,
+        bundle_topic=bundle_topic,
         with_pose=with_pose,
         pose_config=pose_config.expanduser().resolve() if pose_config is not None else None,
         world_frame_id=world_frame_id,
@@ -1156,7 +1316,7 @@ def main(
             continue
 
         if report_existing:
-            probe_result = probe_output(output_path, job.camera_labels)
+            probe_result = probe_output(output_path, job.camera_labels, bundle_topic=config.bundle_topic)
             if probe_result.status == "valid":
                 valid_existing.append(probe_result)
             elif probe_result.status == "invalid":
@@ -1171,7 +1331,7 @@ def main(
             continue
 
         if config.probe_existing:
-            probe_result = probe_output(output_path, job.camera_labels)
+            probe_result = probe_output(output_path, job.camera_labels, bundle_topic=config.bundle_topic)
             if probe_result.status == "valid":
                 valid_existing.append(probe_result)
                 skipped_results.append(

src/record/mcap_record_sink.cpp

@@ -3,6 +3,7 @@
 #include "cvmmap_streamer/record/mcap_record_sink.hpp"
 
 #include "protobuf_descriptor.hpp"
+#include "proto/cvmmap_streamer/BundleManifest.pb.h"
 #include "proto/cvmmap_streamer/DepthMap.pb.h"
 #include "proto/foxglove/CameraCalibration.pb.h"
 #include "proto/foxglove/CompressedVideo.pb.h"
@@ -117,6 +118,28 @@ cvmmap_streamer::DepthMap::Encoding to_proto_depth_encoding(DepthEncoding encodi
 	}
 }
 
+[[nodiscard]]
+cvmmap_streamer::BundleManifest::BundlePolicy to_proto_bundle_policy(BundlePolicy policy) {
+	switch (policy) {
+	case BundlePolicy::Strict:
+		return cvmmap_streamer::BundleManifest::BUNDLE_POLICY_STRICT;
+	case BundlePolicy::Nearest:
+	default:
+		return cvmmap_streamer::BundleManifest::BUNDLE_POLICY_NEAREST;
+	}
+}
+
+[[nodiscard]]
+cvmmap_streamer::BundleManifest::BundleMemberStatus to_proto_bundle_member_status(BundleMemberStatus status) {
+	switch (status) {
+	case BundleMemberStatus::CorruptedGap:
+		return cvmmap_streamer::BundleManifest::BUNDLE_MEMBER_STATUS_CORRUPTED_GAP;
+	case BundleMemberStatus::Present:
+	default:
+		return cvmmap_streamer::BundleManifest::BUNDLE_MEMBER_STATUS_PRESENT;
+	}
+}
+
 void append_start_code(std::vector<std::uint8_t> &output) {
 	output.push_back(0x00);
 	output.push_back(0x00);
@@ -464,6 +487,60 @@ std::expected<void, std::string> write_calibration_message(
 	return {};
 }
 
+[[nodiscard]]
+std::expected<void, std::string> write_bundle_manifest_message(
+	mcap::McapWriter &writer,
+	mcap::ChannelId channel_id,
+	std::uint32_t &sequence,
+	const RawBundleManifestView &bundle) {
+	if (channel_id == 0) {
+		return std::unexpected("bundle topic is disabled");
+	}
+	if (bundle.members.empty()) {
+		return std::unexpected("bundle manifest must contain at least one member");
+	}
+
+	cvmmap_streamer::BundleManifest message{};
+	*message.mutable_timestamp() = to_proto_timestamp(bundle.timestamp_ns);
+	message.set_bundle_index(bundle.bundle_index);
+	message.set_policy(to_proto_bundle_policy(bundle.policy));
+	for (const auto &member_view : bundle.members) {
+		if (member_view.camera_label.empty()) {
+			return std::unexpected("bundle member camera label is empty");
+		}
+		if (member_view.status == BundleMemberStatus::Present && !member_view.timestamp_ns.has_value()) {
+			return std::unexpected("present bundle member is missing a timestamp");
+		}
+		auto *member = message.add_members();
+		member->set_camera_label(std::string(member_view.camera_label));
+		member->set_status(to_proto_bundle_member_status(member_view.status));
+		if (member_view.timestamp_ns.has_value()) {
+			*member->mutable_timestamp() = to_proto_timestamp(*member_view.timestamp_ns);
+		}
+		member->set_delta_ns(member_view.delta_ns);
+		member->set_corrupted_frames_skipped(member_view.corrupted_frames_skipped);
+	}
+
+	std::string serialized{};
+	if (!message.SerializeToString(&serialized)) {
+		return std::unexpected("failed to serialize cvmmap_streamer.BundleManifest");
+	}
+
+	mcap::Message record{};
+	record.channelId   = channel_id;
+	record.sequence    = sequence++;
+	record.logTime     = bundle.timestamp_ns;
+	record.publishTime = bundle.timestamp_ns;
+	record.data        = reinterpret_cast<const std::byte *>(serialized.data());
+	record.dataSize    = serialized.size();
+
+	const auto write_status = writer.write(record);
+	if (!write_status.ok()) {
+		return std::unexpected("failed to write MCAP bundle manifest: " + write_status.message);
+	}
+	return {};
+}
+
 }
 
 struct McapRecordSink::State {
@@ -835,8 +912,12 @@ struct MultiMcapRecordSink::State {
 	std::string path{};
 	mcap::SchemaId video_schema_id{0};
 	mcap::SchemaId depth_schema_id{0};
+	mcap::SchemaId bundle_schema_id{0};
 	mcap::SchemaId calibration_schema_id{0};
 	mcap::SchemaId pose_schema_id{0};
+	std::string bundle_topic{};
+	mcap::ChannelId bundle_channel_id{0};
+	std::uint32_t bundle_sequence{0};
 	std::vector<StreamState> streams{};
 };
 
@@ -959,10 +1040,12 @@ MultiMcapRecordSink &MultiMcapRecordSink::operator=(MultiMcapRecordSink &&other)
 
 std::expected<MultiMcapRecordSink, std::string> MultiMcapRecordSink::create(
 	std::string path,
-	McapCompression compression) {
+	McapCompression compression,
+	std::string bundle_topic) {
 	MultiMcapRecordSink sink{};
 	auto state = std::make_unique<State>();
 	state->path = std::move(path);
+	state->bundle_topic = std::move(bundle_topic);
 
 	mcap::McapWriterOptions options("");
 	options.compression = to_mcap_compression(compression);
@@ -989,6 +1072,21 @@ std::expected<MultiMcapRecordSink, std::string> MultiMcapRecordSink::create(
 	state->writer.addSchema(depth_schema);
 	state->depth_schema_id = depth_schema.id;
 
+	if (!state->bundle_topic.empty()) {
+		const auto bundle_descriptor_set = build_file_descriptor_set(cvmmap_streamer::BundleManifest::descriptor());
+		std::string bundle_schema_bytes{};
+		if (!bundle_descriptor_set.SerializeToString(&bundle_schema_bytes)) {
+			return std::unexpected("failed to serialize cvmmap_streamer.BundleManifest descriptor set");
+		}
+		mcap::Schema bundle_schema("cvmmap_streamer.BundleManifest", "protobuf", bundle_schema_bytes);
+		state->writer.addSchema(bundle_schema);
+		state->bundle_schema_id = bundle_schema.id;
+
+		mcap::Channel bundle_channel(state->bundle_topic, "protobuf", state->bundle_schema_id);
+		state->writer.addChannel(bundle_channel);
+		state->bundle_channel_id = bundle_channel.id;
+	}
+
 	const auto calibration_descriptor_set = build_file_descriptor_set(foxglove::CameraCalibration::descriptor());
 	std::string calibration_schema_bytes{};
 	if (!calibration_descriptor_set.SerializeToString(&calibration_schema_bytes)) {
@@ -1237,6 +1335,18 @@ std::expected<void, std::string> MultiMcapRecordSink::write_pose(
 	return {};
 }
 
+std::expected<void, std::string> MultiMcapRecordSink::write_bundle_manifest(
+	const RawBundleManifestView &bundle) {
+	if (state_ == nullptr) {
+		return std::unexpected("MCAP sink is not open");
+	}
+	return write_bundle_manifest_message(
+		state_->writer,
+		state_->bundle_channel_id,
+		state_->bundle_sequence,
+		bundle);
+}
+
 std::expected<void, std::string> MultiMcapRecordSink::write_body_tracking_message(
 	const StreamId stream_id,
 	const RawBodyTrackingMessageView &body_message) {

src/testers/mcap_multi_record_tester.cpp

@@ -1,5 +1,6 @@
 #include <mcap/reader.hpp>
 
+#include "proto/cvmmap_streamer/BundleManifest.pb.h"
 #include "proto/cvmmap_streamer/DepthMap.pb.h"
 #include "cvmmap_streamer/common.h"
 #include "cvmmap_streamer/record/mcap_record_sink.hpp"
@@ -73,7 +74,8 @@ int main(int argc, char **argv) {
 
 	auto sink = cvmmap_streamer::record::MultiMcapRecordSink::create(
 		output_path.string(),
-		compression);
+		compression,
+		"/bundle");
 	if (!sink) {
 		spdlog::error("failed to create MCAP sink: {}", sink.error());
 		return exit_code(TesterExitCode::CreateError);
@@ -125,6 +127,31 @@ int main(int argc, char **argv) {
 		0.0, 500.0, 240.0, 0.0,
 		0.0, 0.0, 1.0, 0.0,
 	};
+	const std::vector<cvmmap_streamer::record::RawBundleMemberView> bundle_members{
+		{
+			.camera_label = "zed1",
+			.status = cvmmap_streamer::record::BundleMemberStatus::Present,
+			.timestamp_ns = 100,
+			.delta_ns = -5,
+		},
+		{
+			.camera_label = "zed2",
+			.status = cvmmap_streamer::record::BundleMemberStatus::CorruptedGap,
+			.timestamp_ns = std::nullopt,
+			.delta_ns = 0,
+			.corrupted_frames_skipped = 3,
+		},
+	};
+
+	if (auto write = sink->write_bundle_manifest(cvmmap_streamer::record::RawBundleManifestView{
+			.timestamp_ns = 105,
+			.bundle_index = 7,
+			.policy = cvmmap_streamer::record::BundlePolicy::Nearest,
+			.members = bundle_members,
+		}); !write) {
+		spdlog::error("failed to write bundle manifest: {}", write.error());
+		return exit_code(TesterExitCode::WriteError);
+	}
 
 	for (const auto [stream_id, label, pose_x] : {
 			 std::tuple{*zed1, std::string("zed1"), 1.0},
@@ -242,6 +269,30 @@ int main(int argc, char **argv) {
 			continue;
 		}
 
+		if (it->schema->name == "cvmmap_streamer.BundleManifest") {
+			cvmmap_streamer::BundleManifest bundle{};
+			if (!bundle.ParseFromArray(it->message.data, static_cast<int>(it->message.dataSize))) {
+				spdlog::error("failed to parse cvmmap_streamer.BundleManifest");
+				reader.close();
+				return exit_code(TesterExitCode::VerificationError);
+			}
+			if (bundle.bundle_index() != 7 || bundle.members_size() != 2) {
+				spdlog::error("bundle manifest contents mismatch");
+				reader.close();
+				return exit_code(TesterExitCode::VerificationError);
+			}
+			if (bundle.members(0).status() != cvmmap_streamer::BundleManifest::BUNDLE_MEMBER_STATUS_PRESENT ||
+				!bundle.members(0).has_timestamp() ||
+				bundle.members(1).status() != cvmmap_streamer::BundleManifest::BUNDLE_MEMBER_STATUS_CORRUPTED_GAP ||
+				bundle.members(1).has_timestamp() ||
+				bundle.members(1).corrupted_frames_skipped() != 3) {
+				spdlog::error("bundle manifest member status mismatch");
+				reader.close();
+				return exit_code(TesterExitCode::VerificationError);
+			}
+			continue;
+		}
+
 		if (it->schema->name == "foxglove.CameraCalibration") {
 			foxglove::CameraCalibration calibration{};
 			if (!calibration.ParseFromArray(it->message.data, static_cast<int>(it->message.dataSize))) {
@@ -266,6 +317,7 @@ int main(int argc, char **argv) {
 	reader.close();
 
 	for (const auto &topic : {
+			 "/bundle",
 			 "/zed1/video",
 			 "/zed1/depth",
 			 "/zed1/calibration",

src/tools/zed_svo_to_mcap.cpp

@@ -55,6 +55,8 @@ struct CliOptions {
 	std::string mcap_compression{"zstd"};
 	std::string depth_mode{"neural_plus"};
 	std::string depth_size{"optimal"};
+	std::string bundle_policy{"nearest"};
+	std::string bundle_topic{"/bundle"};
 	bool with_pose{false};
 	std::uint32_t start_frame{0};
 	std::uint32_t end_frame{0};
@@ -126,6 +128,7 @@ struct CameraStream {
 	TrackingSample next_tracking{};
 	std::uint64_t current_timestamp_ns{0};
 	std::uint64_t next_timestamp_ns{0};
+	std::uint32_t next_corrupted_frames_skipped{0};
 	std::uint64_t first_timestamp_ns{0};
 	std::uint64_t last_timestamp_ns{0};
 	std::uint64_t total_frames{0};
@@ -152,6 +155,24 @@ struct CameraStream {
 	CalibrationData depth_calibration{};
 };
 
+struct BundledFrameSelection {
+	std::size_t stream_index{0};
+	cvmmap_streamer::record::BundleMemberStatus status{
+		cvmmap_streamer::record::BundleMemberStatus::Present,
+	};
+	bool use_next{false};
+	std::optional<std::uint64_t> timestamp_ns{};
+	std::int64_t delta_ns{0};
+	std::uint32_t corrupted_frames_skipped{0};
+};
+
+struct GrabResult {
+	int svo_position{-1};
+	std::uint32_t corrupted_frames_skipped{0};
+	int first_corrupted_position{-1};
+	int last_corrupted_position{-1};
+};
+
 [[nodiscard]]
 constexpr int exit_code(const ToolExitCode code) {
 	return static_cast<int>(code);
@@ -302,6 +323,18 @@ std::expected<sl::Resolution, std::string> parse_depth_size(const std::string_vi
 	return sl::Resolution(std::stoi(match[1].str()), std::stoi(match[2].str()));
 }
 
+[[nodiscard]]
+std::expected<cvmmap_streamer::record::BundlePolicy, std::string> parse_bundle_policy(const std::string_view raw) {
+	const auto normalized = lowercase(std::string(raw));
+	if (normalized == "nearest") {
+		return cvmmap_streamer::record::BundlePolicy::Nearest;
+	}
+	if (normalized == "strict") {
+		return cvmmap_streamer::record::BundlePolicy::Strict;
+	}
+	return std::unexpected("invalid bundle policy: '" + std::string(raw) + "' (expected: nearest|strict)");
+}
+
 [[nodiscard]]
 std::uint64_t frame_period_ns(const float fps) {
 	if (!(fps > 0.0f)) {
@@ -659,36 +692,118 @@ void maybe_log_tracking_state(CameraStream &stream, const sl::POSITIONAL_TRACKIN
 }
 
 [[nodiscard]]
-std::expected<void, std::string> read_frame_data(
+std::expected<GrabResult, std::string> grab_next_readable_frame(
+	sl::Camera &camera,
+	const sl::RuntimeParameters &runtime,
+	const std::string_view source_label,
+	const std::string_view source_path,
+	const std::uint64_t total_frames,
+	const std::optional<std::uint64_t> current_timestamp_ns,
+	const std::optional<std::uint64_t> last_timestamp_ns) {
+	GrabResult result{};
+	while (true) {
+		const auto grab_status = camera.grab(runtime);
+		const auto svo_position = camera.getSVOPosition();
+		if (grab_status == sl::ERROR_CODE::END_OF_SVOFILE_REACHED) {
+			return std::unexpected("end-of-svo");
+		}
+		if (grab_status == sl::ERROR_CODE::CORRUPTED_FRAME) {
+			if (result.first_corrupted_position < 0) {
+				result.first_corrupted_position = svo_position;
+			}
+			result.last_corrupted_position = svo_position;
+			result.corrupted_frames_skipped += 1;
+
+			const auto last_frame_position = total_frames > 0
+				? static_cast<int>(total_frames - 1)
+				: 0;
+			const bool at_or_past_tail = svo_position >= last_frame_position;
+			const bool beyond_last_readable =
+				last_timestamp_ns.has_value() &&
+				current_timestamp_ns.has_value() &&
+				*current_timestamp_ns >= *last_timestamp_ns;
+			if (at_or_past_tail || beyond_last_readable) {
+				spdlog::warn(
+					"treating {} unreadable tail frame(s) as end-of-svo for {} last_corrupted_position={} current_timestamp_ns={} last_timestamp_ns={}",
+					result.corrupted_frames_skipped,
+					source_path,
+					svo_position,
+					current_timestamp_ns.value_or(0),
+					last_timestamp_ns.value_or(0));
+				return std::unexpected("end-of-svo");
+			}
+
+			if (svo_position < 0) {
+				return std::unexpected(
+					"failed to advance past corrupted frame for " + std::string(source_label) +
+					": invalid svo_position=" + std::to_string(svo_position));
+			}
+
+			const auto next_position = svo_position + 1;
+			if (next_position >= static_cast<int>(total_frames)) {
+				spdlog::warn(
+					"treating {} unreadable tail frame(s) as end-of-svo for {} last_corrupted_position={} current_timestamp_ns={} last_timestamp_ns={}",
+					result.corrupted_frames_skipped,
+					source_path,
+					svo_position,
+					current_timestamp_ns.value_or(0),
+					last_timestamp_ns.value_or(0));
+				return std::unexpected("end-of-svo");
+			}
+			camera.setSVOPosition(next_position);
+			continue;
+		}
+		if (grab_status != sl::ERROR_CODE::SUCCESS) {
+			return std::unexpected(
+				"failed to grab frame for " + std::string(source_label) +
+				": " + zed_status_string(grab_status) +
+				" svo_position=" + std::to_string(svo_position) +
+				" total_frames=" + std::to_string(total_frames) +
+				" current_timestamp_ns=" + std::to_string(current_timestamp_ns.value_or(0)) +
+				" last_timestamp_ns=" + std::to_string(last_timestamp_ns.value_or(0)) +
+				" corrupted_frames_skipped=" + std::to_string(result.corrupted_frames_skipped));
+		}
+
+		result.svo_position = svo_position;
+		return result;
+	}
+}
+
+void maybe_log_recovered_corruption_gap(
+	const std::string_view source_label,
+	const GrabResult &grab_result,
+	const std::uint64_t recovered_timestamp_ns) {
+	if (grab_result.corrupted_frames_skipped == 0) {
+		return;
+	}
+	spdlog::warn(
+		"recovered {} after skipping {} corrupted frame(s) positions={}..{} recovered_position={} recovered_timestamp_ns={}",
+		source_label,
+		grab_result.corrupted_frames_skipped,
+		grab_result.first_corrupted_position,
+		grab_result.last_corrupted_position,
+		grab_result.svo_position,
+		recovered_timestamp_ns);
+}
+
+[[nodiscard]]
+std::expected<GrabResult, std::string> read_frame_data(
 	CameraStream &stream,
 	sl::Mat &left_frame,
 	sl::Mat &depth_frame,
 	TrackingSample &tracking_sample,
 	std::optional<std::uint64_t> fallback_timestamp_ns,
 	std::uint64_t &timestamp_ns_out) {
-	const auto grab_status = stream.camera->grab(stream.runtime);
-	if (grab_status == sl::ERROR_CODE::END_OF_SVOFILE_REACHED) {
-		return std::unexpected("end-of-svo");
-	}
-	if (grab_status == sl::ERROR_CODE::CORRUPTED_FRAME &&
-		stream.last_timestamp_ns != 0 &&
-		stream.current_timestamp_ns >= stream.last_timestamp_ns) {
-		spdlog::warn(
-			"treating unreadable tail frame as end-of-svo for {} current_timestamp_ns={} last_timestamp_ns={}",
+	auto grab = grab_next_readable_frame(
+		*stream.camera,
+		stream.runtime,
+		stream.source.label,
 		stream.source.path.string(),
-			stream.current_timestamp_ns,
-			stream.last_timestamp_ns);
-		return std::unexpected("end-of-svo");
-	}
-	if (grab_status != sl::ERROR_CODE::SUCCESS) {
-		const auto svo_position = stream.camera->getSVOPosition();
-		return std::unexpected(
-			"failed to grab frame for " + stream.source.label +
-			": " + zed_status_string(grab_status) +
-			" svo_position=" + std::to_string(svo_position) +
-			" total_frames=" + std::to_string(stream.total_frames) +
-			" current_timestamp_ns=" + std::to_string(stream.current_timestamp_ns) +
-			" last_timestamp_ns=" + std::to_string(stream.last_timestamp_ns));
+		stream.total_frames,
+		stream.current_timestamp_ns == 0 ? std::nullopt : std::optional<std::uint64_t>{stream.current_timestamp_ns},
+		stream.last_timestamp_ns == 0 ? std::nullopt : std::optional<std::uint64_t>{stream.last_timestamp_ns});
+	if (!grab) {
+		return std::unexpected(grab.error());
 	}
 
 	const auto image_status = stream.camera->retrieveImage(left_frame, sl::VIEW::LEFT_BGR, sl::MEM::CPU);
@@ -718,6 +833,7 @@ std::expected<void, std::string> read_frame_data(
 		return std::unexpected(timestamp_ns.error());
 	}
 	timestamp_ns_out = *timestamp_ns;
+	maybe_log_recovered_corruption_gap(stream.source.label, *grab, timestamp_ns_out);
 
 	tracking_sample = {};
 	if (stream.tracking_enabled) {
@@ -741,7 +857,7 @@ std::expected<void, std::string> read_frame_data(
 			};
 		}
 	}
-	return {};
+	return *grab;
 }
 
 [[nodiscard]]
@@ -756,11 +872,13 @@ std::expected<void, std::string> fill_next_frame(CameraStream &stream) {
 	if (!next) {
 		if (next.error() == "end-of-svo") {
 			stream.has_next = false;
+			stream.next_corrupted_frames_skipped = 0;
 			return {};
 		}
 		return std::unexpected(next.error());
 	}
 	stream.has_next = true;
+	stream.next_corrupted_frames_skipped = next->corrupted_frames_skipped;
 	return {};
 }
 
@@ -775,6 +893,7 @@ std::expected<void, std::string> promote_next_frame(CameraStream &stream) {
 	std::swap(stream.current_tracking, stream.next_tracking);
 	std::swap(stream.current_timestamp_ns, stream.next_timestamp_ns);
 	stream.has_next = false;
+	stream.next_corrupted_frames_skipped = 0;
 	return fill_next_frame(stream);
 }
 
@@ -1049,14 +1168,19 @@ std::expected<void, std::string> sync_streams_to_timestamp(
 		if (log_shutdown_request(shutdown_logged, "multi-camera sync")) {
 			return std::unexpected("interrupted");
 		}
-		stream.sync_position = stream.camera->getSVOPositionAtTimestamp(sl::Timestamp{effective_start_ts});
-		if (stream.sync_position < 0) {
+		const auto sdk_position = stream.camera->getSVOPositionAtTimestamp(sl::Timestamp{effective_start_ts});
+		if (sdk_position < 0) {
 			return std::unexpected(
 				"failed to compute synced start frame for " + stream.source.path.string() + " at timestamp " +
 				std::to_string(effective_start_ts));
 		}
+		stream.sync_position = std::clamp(
+			sdk_position,
+			0,
+			static_cast<int>(stream.total_frames > 0 ? stream.total_frames - 1 : 0));
 
-		stream.camera->setSVOPosition(stream.sync_position);
+		const auto read_at_position = [&](const int position) -> std::expected<void, std::string> {
+			stream.camera->setSVOPosition(position);
 			auto current = read_frame_data(
 				stream,
 				stream.current_left_frame,
@@ -1071,8 +1195,24 @@ std::expected<void, std::string> sync_streams_to_timestamp(
 			if (!next) {
 				return std::unexpected(next.error());
 			}
+			return {};
+		};
 
-		while (stream.current_timestamp_ns < effective_start_ts && stream.has_next) {
+		if (auto loaded = read_at_position(stream.sync_position); !loaded) {
+			return std::unexpected(loaded.error());
+		}
+
+		while (stream.sync_position > 0 && stream.current_timestamp_ns > effective_start_ts) {
+			if (log_shutdown_request(shutdown_logged, "multi-camera sync")) {
+				return std::unexpected("interrupted");
+			}
+			stream.sync_position -= 1;
+			if (auto loaded = read_at_position(stream.sync_position); !loaded) {
+				return std::unexpected(loaded.error());
+			}
+		}
+
+		while (stream.has_next && stream.next_timestamp_ns <= effective_start_ts) {
 			if (log_shutdown_request(shutdown_logged, "multi-camera sync")) {
 				return std::unexpected("interrupted");
 			}
@@ -1107,6 +1247,119 @@ bool have_sync_window(const std::vector<CameraStream> &streams, const std::uint6
 	return true;
 }
 
+[[nodiscard]]
+std::expected<void, std::string> advance_streams_to_timestamp(
+	std::vector<CameraStream> &streams,
+	const std::uint64_t target_timestamp_ns) {
+	for (auto &stream : streams) {
+		while (stream.has_next && stream.next_timestamp_ns <= target_timestamp_ns) {
+			stream.dropped_frames += 1;
+			auto promote = promote_next_frame(stream);
+			if (!promote) {
+				return std::unexpected(promote.error());
+			}
+		}
+	}
+	return {};
+}
+
+[[nodiscard]]
+std::uint64_t timestamp_delta_abs(const std::uint64_t left, const std::uint64_t right) {
+	return left >= right ? left - right : right - left;
+}
+
+[[nodiscard]]
+std::expected<std::vector<BundledFrameSelection>, std::string> select_nearest_bundle(
+	const std::vector<CameraStream> &streams,
+	const std::uint64_t bundle_timestamp_ns,
+	const std::uint64_t common_end_ts) {
+	std::vector<BundledFrameSelection> selections{};
+	selections.reserve(streams.size());
+
+	for (std::size_t stream_index = 0; stream_index < streams.size(); ++stream_index) {
+		const auto &stream = streams[stream_index];
+		if (stream.current_timestamp_ns > common_end_ts) {
+			return std::unexpected("no bundle frames remain within the common overlap window");
+		}
+
+		const bool blocked_by_corruption_gap =
+			stream.has_next &&
+			stream.next_corrupted_frames_skipped > 0 &&
+			bundle_timestamp_ns > stream.current_timestamp_ns &&
+			bundle_timestamp_ns < stream.next_timestamp_ns;
+		if (blocked_by_corruption_gap) {
+			selections.push_back(BundledFrameSelection{
+				.stream_index = stream_index,
+				.status = cvmmap_streamer::record::BundleMemberStatus::CorruptedGap,
+				.use_next = false,
+				.timestamp_ns = std::nullopt,
+				.delta_ns = 0,
+				.corrupted_frames_skipped = stream.next_corrupted_frames_skipped,
+			});
+			continue;
+		}
+
+		bool use_next = false;
+		std::uint64_t selected_timestamp_ns = stream.current_timestamp_ns;
+		if (stream.has_next && stream.next_timestamp_ns <= common_end_ts) {
+			const auto current_delta = timestamp_delta_abs(stream.current_timestamp_ns, bundle_timestamp_ns);
+			const auto next_delta = timestamp_delta_abs(stream.next_timestamp_ns, bundle_timestamp_ns);
+			if (next_delta <= current_delta) {
+				use_next = true;
+				selected_timestamp_ns = stream.next_timestamp_ns;
+			}
+		}
+
+		selections.push_back(BundledFrameSelection{
+			.stream_index = stream_index,
+			.status = cvmmap_streamer::record::BundleMemberStatus::Present,
+			.use_next = use_next,
+			.timestamp_ns = selected_timestamp_ns,
+			.delta_ns = static_cast<std::int64_t>(selected_timestamp_ns) -
+				static_cast<std::int64_t>(bundle_timestamp_ns),
+			.corrupted_frames_skipped = 0,
+		});
+	}
+
+	return selections;
+}
+
+[[nodiscard]]
+std::vector<BundledFrameSelection> make_strict_bundle(
+	const std::vector<CameraStream> &streams,
+	const std::uint64_t bundle_timestamp_ns) {
+	std::vector<BundledFrameSelection> selections{};
+	selections.reserve(streams.size());
+	for (std::size_t stream_index = 0; stream_index < streams.size(); ++stream_index) {
+		const auto &stream = streams[stream_index];
+		selections.push_back(BundledFrameSelection{
+			.stream_index = stream_index,
+			.status = cvmmap_streamer::record::BundleMemberStatus::Present,
+			.use_next = false,
+			.timestamp_ns = stream.current_timestamp_ns,
+			.delta_ns = static_cast<std::int64_t>(stream.current_timestamp_ns) -
+				static_cast<std::int64_t>(bundle_timestamp_ns),
+			.corrupted_frames_skipped = 0,
+		});
+	}
+	return selections;
+}
+
+[[nodiscard]]
+const sl::Mat &selected_left_frame(const CameraStream &stream, const BundledFrameSelection &selection) {
+	return selection.use_next ? stream.next_left_frame : stream.current_left_frame;
+}
+
+[[nodiscard]]
+const sl::Mat &selected_depth_frame(const CameraStream &stream, const BundledFrameSelection &selection) {
+	return selection.use_next ? stream.next_depth_frame : stream.current_depth_frame;
+}
+
+[[nodiscard]]
+const TrackingSample &selected_tracking(const CameraStream &stream, const BundledFrameSelection &selection) {
+	return selection.use_next ? stream.next_tracking : stream.current_tracking;
+}
+
 [[nodiscard]]
 std::expected<std::optional<std::uint64_t>, std::string> next_synced_group_timestamp(
 	std::vector<CameraStream> &streams,
@@ -1167,15 +1420,66 @@ std::expected<void, std::string> encode_and_write_group(
 	cvmmap_streamer::record::MultiMcapRecordSink &sink,
 	std::vector<CameraStream> &streams,
 	const CliOptions &options,
-	const std::uint64_t group_timestamp_ns) {
-	for (auto &stream : streams) {
-		const auto video_step_bytes = stream.current_left_frame.getStepBytes(sl::MEM::CPU);
+	const cvmmap_streamer::record::BundlePolicy bundle_policy,
+	const std::uint64_t bundle_index,
+	const std::uint64_t bundle_timestamp_ns,
+	const std::span<const BundledFrameSelection> selections) {
+	if (selections.size() != streams.size()) {
+		return std::unexpected("bundle selection size does not match stream count");
+	}
+
+	std::vector<cvmmap_streamer::record::RawBundleMemberView> bundle_members{};
+	bundle_members.reserve(selections.size());
+	for (const auto &selection : selections) {
+		const auto &stream = streams[selection.stream_index];
+		bundle_members.push_back(cvmmap_streamer::record::RawBundleMemberView{
+			.camera_label = stream.source.label,
+			.status = selection.status,
+			.timestamp_ns = selection.timestamp_ns,
+			.delta_ns = selection.delta_ns,
+			.corrupted_frames_skipped = selection.corrupted_frames_skipped,
+		});
+	}
+	if (auto write = sink.write_bundle_manifest(cvmmap_streamer::record::RawBundleManifestView{
+			.timestamp_ns = bundle_timestamp_ns,
+			.bundle_index = bundle_index,
+			.policy = bundle_policy,
+			.members = bundle_members,
+		}); !write) {
+		return std::unexpected("failed to write bundle manifest: " + write.error());
+	}
+
+	std::vector<BundledFrameSelection> ordered_selections{};
+	ordered_selections.reserve(selections.size());
+	for (const auto &selection : selections) {
+		if (selection.status == cvmmap_streamer::record::BundleMemberStatus::Present &&
+			selection.timestamp_ns.has_value()) {
+			ordered_selections.push_back(selection);
+		}
+	}
+	std::sort(
+		ordered_selections.begin(),
+		ordered_selections.end(),
+		[](const auto &left, const auto &right) {
+			if (left.timestamp_ns != right.timestamp_ns) {
+				return left.timestamp_ns < right.timestamp_ns;
+			}
+			return left.stream_index < right.stream_index;
+		});
+
+	for (const auto &selection : ordered_selections) {
+		auto &stream = streams[selection.stream_index];
+		const auto &left_frame = selected_left_frame(stream, selection);
+		const auto &depth_frame = selected_depth_frame(stream, selection);
+		const auto &tracking = selected_tracking(stream, selection);
+
+		const auto video_step_bytes = left_frame.getStepBytes(sl::MEM::CPU);
 		const auto video_bytes = std::span<const std::uint8_t>(
-			stream.current_left_frame.getPtr<sl::uchar1>(sl::MEM::CPU),
-			video_step_bytes * stream.current_left_frame.getHeight());
+			left_frame.getPtr<sl::uchar1>(sl::MEM::CPU),
+			video_step_bytes * left_frame.getHeight());
 		cvmmap_streamer::encode::RawVideoFrame raw_video{
 			.info = stream.frame_info,
-			.source_timestamp_ns = group_timestamp_ns,
+			.source_timestamp_ns = *selection.timestamp_ns,
 			.row_stride_bytes = video_step_bytes,
 			.bytes = video_bytes,
 		};
@@ -1196,7 +1500,7 @@ std::expected<void, std::string> encode_and_write_group(
 
 		if (!stream.calibration_written) {
 			cvmmap_streamer::record::RawCameraCalibrationView calibration{
-				.timestamp_ns = group_timestamp_ns,
+				.timestamp_ns = *selection.timestamp_ns,
 				.width = stream.video_calibration.width,
 				.height = stream.video_calibration.height,
 				.distortion_model = "plumb_bob",
@@ -1214,7 +1518,7 @@ std::expected<void, std::string> encode_and_write_group(
 		if (!stream.depth_calibration_written &&
 			(stream.depth_width != stream.width || stream.depth_height != stream.height)) {
 			cvmmap_streamer::record::RawCameraCalibrationView depth_calibration{
-				.timestamp_ns = group_timestamp_ns,
+				.timestamp_ns = *selection.timestamp_ns,
 				.width = stream.depth_calibration.width,
 				.height = stream.depth_calibration.height,
 				.distortion_model = "plumb_bob",
@@ -1229,15 +1533,15 @@ std::expected<void, std::string> encode_and_write_group(
 			stream.depth_calibration_written = true;
 		}
 
-		const auto depth_width = static_cast<std::uint32_t>(stream.current_depth_frame.getWidth());
-		const auto depth_height = static_cast<std::uint32_t>(stream.current_depth_frame.getHeight());
+		const auto depth_width = static_cast<std::uint32_t>(depth_frame.getWidth());
+		const auto depth_height = static_cast<std::uint32_t>(depth_frame.getHeight());
 		if (depth_width != stream.depth_width || depth_height != stream.depth_height) {
 			return std::unexpected(
 				"depth resolution changed unexpectedly for " + stream.source.label + ": " +
 				std::to_string(depth_width) + "x" + std::to_string(depth_height) + " vs " +
 				std::to_string(stream.depth_width) + "x" + std::to_string(stream.depth_height));
 		}
-		const auto depth_step_bytes = stream.current_depth_frame.getStepBytes(sl::MEM::CPU);
+		const auto depth_step_bytes = depth_frame.getStepBytes(sl::MEM::CPU);
 		const auto packed_depth_bytes = static_cast<std::size_t>(depth_width) * sizeof(std::uint16_t);
 		if (depth_step_bytes < packed_depth_bytes) {
 			return std::unexpected(
@@ -1249,15 +1553,15 @@ std::expected<void, std::string> encode_and_write_group(
 		std::span<const std::uint16_t> depth_pixels{};
 		if (depth_step_bytes == packed_depth_bytes) {
 			depth_pixels = std::span<const std::uint16_t>(
-				stream.current_depth_frame.getPtr<sl::ushort1>(sl::MEM::CPU),
+				depth_frame.getPtr<sl::ushort1>(sl::MEM::CPU),
 				static_cast<std::size_t>(depth_width) * static_cast<std::size_t>(depth_height));
 		} else {
-			compact_depth = copy_compact_u16_plane(stream.current_depth_frame);
+			compact_depth = copy_compact_u16_plane(depth_frame);
 			depth_pixels = *compact_depth;
 		}
 
 		cvmmap_streamer::record::RawDepthMapU16View depth_map{
-			.timestamp_ns = group_timestamp_ns,
+			.timestamp_ns = *selection.timestamp_ns,
 			.width = depth_width,
 			.height = depth_height,
 			.pixels = depth_pixels,
@@ -1266,12 +1570,12 @@ std::expected<void, std::string> encode_and_write_group(
 			return std::unexpected("failed to write depth map for " + stream.source.label + ": " + write.error());
 		}
 
-		if (options.with_pose && stream.current_tracking.has_pose) {
+		if (options.with_pose && tracking.has_pose) {
 			cvmmap_streamer::record::RawPoseView pose_view{
-				.timestamp_ns = group_timestamp_ns,
+				.timestamp_ns = *selection.timestamp_ns,
 				.reference_frame_id = pose_reference_frame_id(stream.pose_tracking, stream.source.label),
-				.position = stream.current_tracking.position,
-				.orientation = stream.current_tracking.orientation,
+				.position = tracking.position,
+				.orientation = tracking.orientation,
 			};
 			if (auto write = sink.write_pose(stream.mcap_stream_id, pose_view); !write) {
 				return std::unexpected("failed to write pose for " + stream.source.label + ": " + write.error());
@@ -1281,6 +1585,23 @@ std::expected<void, std::string> encode_and_write_group(
 	return {};
 }
 
+[[nodiscard]]
+std::expected<void, std::string> advance_after_nearest_emit(
+	std::vector<CameraStream> &streams,
+	const std::span<const BundledFrameSelection> selections) {
+	for (const auto &selection : selections) {
+		if (selection.status != cvmmap_streamer::record::BundleMemberStatus::Present || !selection.use_next) {
+			continue;
+		}
+		auto &stream = streams[selection.stream_index];
+		auto promote = promote_next_frame(stream);
+		if (!promote) {
+			return std::unexpected(promote.error());
+		}
+	}
+	return {};
+}
+
 [[nodiscard]]
 std::expected<void, std::string> advance_after_emit(std::vector<CameraStream> &streams) {
 	for (auto &stream : streams) {
@@ -1552,31 +1873,33 @@ int run_single_source(
 	const auto total_frames_to_emit = static_cast<std::uint64_t>(last_frame - options.start_frame + 1);
 	ProgressBar progress{total_frames_to_emit};
 
-	while (options.start_frame + emitted_frames <= last_frame) {
+	while (true) {
 		if (log_shutdown_request(shutdown_logged, "single-camera export")) {
 			interrupted = true;
 			break;
 		}
-		const auto grab_status = camera.grab(runtime_parameters);
-		if (grab_status == sl::ERROR_CODE::END_OF_SVOFILE_REACHED) {
-			break;
-		}
-		if (grab_status == sl::ERROR_CODE::CORRUPTED_FRAME &&
-			camera.getSVOPosition() >= static_cast<int>(last_frame)) {
-			spdlog::warn(
-				"treating unreadable tail frame as end-of-svo for {} at frame {}",
+		auto grab = grab_next_readable_frame(
+			camera,
+			runtime_parameters,
+			input_path.filename().string(),
 			input_path.string(),
-				camera.getSVOPosition());
+			static_cast<std::uint64_t>(total_frames),
+			last_timestamp_ns,
+			std::nullopt);
+		if (!grab && grab.error() == "end-of-svo") {
 			break;
 		}
-		if (grab_status != sl::ERROR_CODE::SUCCESS) {
+		if (!grab) {
 			progress.finish(emitted_frames, false);
 			sink->close();
 			backend->shutdown();
 			close_camera();
-			spdlog::error("failed to grab SVO frame: {}", zed_status_string(grab_status));
+			spdlog::error("{}", grab.error());
 			return exit_code(ToolExitCode::RuntimeError);
 		}
+		if (grab->svo_position > static_cast<int>(last_frame)) {
+			break;
+		}
 
 		const auto image_status = camera.retrieveImage(left_frame, sl::VIEW::LEFT_BGR, sl::MEM::CPU);
 		if (image_status != sl::ERROR_CODE::SUCCESS) {
@@ -1621,6 +1944,7 @@ int run_single_source(
 		if (last_timestamp_ns && timestamp_ns <= *last_timestamp_ns) {
 			timestamp_ns = *last_timestamp_ns + 1;
 		}
+		maybe_log_recovered_corruption_gap(input_path.filename().string(), *grab, timestamp_ns);
 		last_timestamp_ns = timestamp_ns;
 
 		const auto video_step_bytes = left_frame.getStepBytes(sl::MEM::CPU);
@@ -1783,7 +2107,7 @@ int run_single_source(
 					spdlog::warn(
 						"pose tracking state changed to {} at frame {}",
 						zed_tracking_state_string(tracking_state),
-						options.start_frame + emitted_frames);
+						grab->svo_position);
 				}
 			}
 			if (tracking_state == sl::POSITIONAL_TRACKING_STATE::OK) {
@@ -1861,10 +2185,11 @@ int run_multi_source(
 	const cvmmap_streamer::McapCompression compression,
 	const sl::DEPTH_MODE depth_mode,
 	const sl::Resolution depth_size,
-	const PoseTrackingOptions &pose_tracking) {
+	const PoseTrackingOptions &pose_tracking,
+	const cvmmap_streamer::record::BundlePolicy bundle_policy) {
 	if (options.has_end_frame && options.end_frame < options.start_frame) {
 		spdlog::error(
-			"invalid bundled group range: start-frame={} end-frame={}",
+			"invalid bundled range: start-frame={} end-frame={}",
 			options.start_frame,
 			options.end_frame);
 		return exit_code(ToolExitCode::UsageError);
@@ -1915,21 +2240,68 @@ int run_multi_source(
 		[](const auto &left, const auto &right) {
 			return left.nominal_frame_period_ns < right.nominal_frame_period_ns;
 		})->nominal_frame_period_ns;
+	const auto bundle_policy_name = bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest
+		? "nearest"
+		: "strict";
 	const auto tolerance_ns = options.has_sync_tolerance
 		? static_cast<std::uint64_t>(std::llround(options.sync_tolerance_ms * 1'000'000.0))
-		: std::max<std::uint64_t>(1, slowest_period_ns / 4);
+		: std::max<std::uint64_t>(1, slowest_period_ns);
 	spdlog::info(
-		"multi-camera sync window start_ts={} end_ts={} tolerance_ns={}",
+		"multi-camera bundle window start_ts={} end_ts={} policy={} bundle_period_ns={} tolerance_ns={}",
 		common_start_ts,
 		common_end_ts,
+		bundle_policy_name,
+		slowest_period_ns,
 		tolerance_ns);
 	const auto render_progress = stderr_supports_progress_bar();
-	const std::uint64_t selected_total_groups = options.has_end_frame
-		? static_cast<std::uint64_t>(options.end_frame - options.start_frame) + 1
+	const auto total_timeline_bundles = common_end_ts >= common_start_ts
+		? ((common_end_ts - common_start_ts) / slowest_period_ns) + 1
 		: 0;
-	const bool exact_group_progress = render_progress && options.has_end_frame;
-	const bool approximate_time_progress = render_progress && !options.has_end_frame;
-	if (auto synced = sync_streams_to_timestamp(streams, common_start_ts); !synced) {
+	if (bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest &&
+		options.start_frame >= total_timeline_bundles) {
+		close_camera_streams(streams);
+		spdlog::error(
+			"start-frame {} is out of range for bundled multi-camera mode (available bundles: {})",
+			options.start_frame,
+			total_timeline_bundles);
+		return exit_code(ToolExitCode::UsageError);
+	}
+	if (bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest &&
+		options.has_end_frame &&
+		options.end_frame >= total_timeline_bundles) {
+		close_camera_streams(streams);
+		spdlog::error(
+			"end-frame {} is out of range for bundled multi-camera mode (available bundles: {})",
+			options.end_frame,
+			total_timeline_bundles);
+		return exit_code(ToolExitCode::UsageError);
+	}
+
+	const auto selected_end_bundle = options.has_end_frame
+		? static_cast<std::uint64_t>(options.end_frame)
+		: (bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest
+			? total_timeline_bundles - 1
+			: 0);
+	const auto selected_total_groups = options.has_end_frame
+		? static_cast<std::uint64_t>(options.end_frame - options.start_frame) + 1
+		: (bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest
+			? total_timeline_bundles - static_cast<std::uint64_t>(options.start_frame)
+			: 0);
+	const bool exact_group_progress =
+		render_progress &&
+		(bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest || options.has_end_frame);
+	const bool approximate_time_progress =
+		render_progress &&
+		bundle_policy == cvmmap_streamer::record::BundlePolicy::Strict &&
+		!options.has_end_frame;
+	ProgressBar progress{exact_group_progress ? selected_total_groups : 0};
+	double last_progress_fraction = 0.0;
+	std::string last_progress_detail{};
+
+	const auto initial_target_ts = bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest
+		? common_start_ts + static_cast<std::uint64_t>(options.start_frame) * slowest_period_ns
+		: common_start_ts;
+	if (auto synced = sync_streams_to_timestamp(streams, initial_target_ts); !synced) {
 		close_camera_streams(streams);
 		if (synced.error() == "interrupted") {
 			return interrupted_exit_code();
@@ -1937,8 +2309,8 @@ int run_multi_source(
 		spdlog::error("{}", synced.error());
 		return exit_code(ToolExitCode::RuntimeError);
 	}
-	auto effective_progress_start_ts = common_start_ts;
-	if (options.start_frame > 0) {
+	auto effective_progress_start_ts = initial_target_ts;
+	if (bundle_policy == cvmmap_streamer::record::BundlePolicy::Strict && options.start_frame > 0) {
 		if (auto skipped_to = skip_bundled_start_groups(streams, options.start_frame, tolerance_ns, common_end_ts); !skipped_to) {
 			close_camera_streams(streams);
 			if (skipped_to.error() == "interrupted") {
@@ -1950,11 +2322,11 @@ int run_multi_source(
 			effective_progress_start_ts = *skipped_to;
 		}
 	}
-	ProgressBar progress{exact_group_progress ? selected_total_groups : 0};
-	double last_progress_fraction = 0.0;
-	std::string last_progress_detail{};
 
-	auto sink = cvmmap_streamer::record::MultiMcapRecordSink::create(output_path.string(), compression);
+	auto sink = cvmmap_streamer::record::MultiMcapRecordSink::create(
+		output_path.string(),
+		compression,
+		options.bundle_topic);
 	if (!sink) {
 		if (approximate_time_progress) {
 			progress.finish_fraction(last_progress_fraction, false, last_progress_detail);
@@ -1978,6 +2350,50 @@ int run_multi_source(
 	}
 
 	std::uint64_t emitted_groups{0};
+	if (bundle_policy == cvmmap_streamer::record::BundlePolicy::Nearest) {
+		for (std::uint64_t bundle_index = options.start_frame; bundle_index <= selected_end_bundle; ++bundle_index) {
+			if (log_shutdown_request(shutdown_logged, "multi-camera export")) {
+				interrupted = true;
+				break;
+			}
+			const auto bundle_timestamp_ns = common_start_ts + bundle_index * slowest_period_ns;
+			if (auto advanced = advance_streams_to_timestamp(streams, bundle_timestamp_ns); !advanced) {
+				progress.finish(emitted_groups, false);
+				sink->close();
+				close_camera_streams(streams);
+				spdlog::error("{}", advanced.error());
+				return exit_code(ToolExitCode::RuntimeError);
+			}
+			auto selections = select_nearest_bundle(streams, bundle_timestamp_ns, common_end_ts);
+			if (!selections) {
+				break;
+			}
+
+			if (auto write = encode_and_write_group(
+					*sink,
+					streams,
+					options,
+					bundle_policy,
+					bundle_index,
+					bundle_timestamp_ns,
+					*selections); !write) {
+				progress.finish(emitted_groups, false);
+				sink->close();
+				close_camera_streams(streams);
+				spdlog::error("{}", write.error());
+				return exit_code(ToolExitCode::RuntimeError);
+			}
+			if (auto advance = advance_after_nearest_emit(streams, *selections); !advance) {
+				progress.finish(emitted_groups, false);
+				sink->close();
+				close_camera_streams(streams);
+				spdlog::error("{}", advance.error());
+				return exit_code(ToolExitCode::RuntimeError);
+			}
+			emitted_groups += 1;
+			progress.update(emitted_groups);
+		}
+	} else {
 		while (true) {
 			if (log_shutdown_request(shutdown_logged, "multi-camera export")) {
 				interrupted = true;
@@ -2002,7 +2418,16 @@ int run_multi_source(
 				break;
 			}
 
-		if (auto write = encode_and_write_group(*sink, streams, options, **group_timestamp); !write) {
+			const auto bundle_index = static_cast<std::uint64_t>(options.start_frame) + emitted_groups;
+			const auto selections = make_strict_bundle(streams, **group_timestamp);
+			if (auto write = encode_and_write_group(
+					*sink,
+					streams,
+					options,
+					bundle_policy,
+					bundle_index,
+					**group_timestamp,
+					selections); !write) {
 				if (approximate_time_progress) {
 					progress.finish_fraction(last_progress_fraction, false, last_progress_detail);
 				} else {
@@ -2041,6 +2466,7 @@ int run_multi_source(
 				return exit_code(ToolExitCode::RuntimeError);
 			}
 		}
+	}
 
 	for (auto &stream : streams) {
 		if (auto flushed = flush_and_write(*sink, stream.mcap_stream_id, *stream.backend); !flushed) {
@@ -2083,15 +2509,16 @@ int run_multi_source(
 				remove_error.message());
 		}
 		spdlog::error(
-			"no synced frame groups were found across {} camera(s) for '{}'",
+			"no bundled frame groups were found across {} camera(s) for '{}' using policy={}",
 			sources.size(),
-			output_path.string());
+			output_path.string(),
+			bundle_policy_name);
 		return exit_code(ToolExitCode::RuntimeError);
 	}
 
 	if (interrupted) {
 		spdlog::warn(
-			"gracefully stopped after writing {} synced frame group(s) across {} camera(s) to '{}'",
+			"gracefully stopped after writing {} bundled frame group(s) across {} camera(s) to '{}'",
 			emitted_groups,
 			sources.size(),
 			output_path.string());
@@ -2099,10 +2526,11 @@ int run_multi_source(
 	}
 
 	spdlog::info(
-		"wrote {} synced frame group(s) across {} camera(s) to '{}'",
+		"wrote {} bundled frame group(s) across {} camera(s) to '{}' using policy={}",
 		emitted_groups,
 		sources.size(),
-		output_path.string());
+		output_path.string(),
+		bundle_policy_name);
 	return exit_code(ToolExitCode::Success);
 }
 
@@ -2127,20 +2555,23 @@ int main(int argc, char **argv) {
 		->check(CLI::IsMember({"neural_light", "neural", "neural_plus"}));
 	app.add_option("--depth-size", options.depth_size, "Depth output size (optimal|native|<width>x<height>)")
 		->default_val("optimal");
+	app.add_option("--bundle-policy", options.bundle_policy, "Bundling policy for multi-camera mode (nearest|strict)")
+		->check(CLI::IsMember({"nearest", "strict"}));
 	app.add_option(
 		"--start-frame",
 		options.start_frame,
-		"First frame/group to export (inclusive): raw SVO frame in single-camera mode, synced group index in bundled multi-camera mode")
+		"First frame/group to export (inclusive): raw SVO frame in single-camera mode, bundle index in multi-camera mode")
 		->check(CLI::NonNegativeNumber);
 	auto *end_frame_option = app.add_option(
 		"--end-frame",
 		options.end_frame,
-		"Last frame/group to export (inclusive): raw SVO frame in single-camera mode, synced group index in bundled multi-camera mode")
+		"Last frame/group to export (inclusive): raw SVO frame in single-camera mode, bundle index in multi-camera mode")
 		->check(CLI::NonNegativeNumber);
 	app.add_option("--frame-id", options.frame_id, "Frame id for image and depth topics");
 	app.add_option("--video-topic", options.video_topic, "MCAP topic for foxglove.CompressedVideo");
 	app.add_option("--depth-topic", options.depth_topic, "MCAP topic for cvmmap_streamer.DepthMap");
 	app.add_option("--calibration-topic", options.calibration_topic, "MCAP topic for foxglove.CameraCalibration");
+	app.add_option("--bundle-topic", options.bundle_topic, "MCAP topic for bundled multi-camera manifests");
 	app.add_option(
 		"--depth-calibration-topic",
 		options.depth_calibration_topic,
@@ -2155,7 +2586,7 @@ int main(int argc, char **argv) {
 	auto *sync_tolerance_option = app.add_option(
 		"--sync-tolerance-ms",
 		options.sync_tolerance_ms,
-		"Maximum allowed timestamp delta between cameras in multi-camera mode");
+		"Maximum allowed timestamp delta between cameras in strict multi-camera mode");
 	sync_tolerance_option->check(CLI::PositiveNumber);
 
 	try {
@@ -2191,6 +2622,11 @@ int main(int argc, char **argv) {
 		spdlog::error("{}", depth_size.error());
 		return exit_code(ToolExitCode::UsageError);
 	}
+	auto bundle_policy = parse_bundle_policy(options.bundle_policy);
+	if (!bundle_policy) {
+		spdlog::error("{}", bundle_policy.error());
+		return exit_code(ToolExitCode::UsageError);
+	}
 	auto pose_tracking = load_pose_tracking_options(options);
 	if (!pose_tracking) {
 		spdlog::error("{}", pose_tracking.error());
@@ -2226,5 +2662,6 @@ int main(int argc, char **argv) {
 		*compression,
 		*depth_mode,
 		*depth_size,
-		*pose_tracking);
+		*pose_tracking,
+		*bundle_policy);
 }