v1.0.0

Interface contract: carbon-interface v1.5.6 — unchanged from v0.17.1.
Firmware compatibility: install FPGA v1.5.6 / MCU v4.1.0 with the v7.1.X
bringup script / updater. Not compatible with FPGA v1.3.3 / MCU v2.6.0 or older.

The first stable release. Recordings move to a new native .vynt format, installation
no longer builds Cap'n Proto from source, the shipped command-line tools drop their
carbon qualifier, and the Python and C++ surfaces are tidied for 1.0. Recordings made
before v1.0.0 need a one-time conversion, and the whole toolchain has to be updated
together — pre-v1.0.0 tools cannot read the new files.

Breaking Changes

Recordings

• New .vynt recording format. It replaces the Cap'n Proto-based format used
  through v0.17.1. Recordings made before v1.0.0 are rejected when opened rather than
  misparsed — convert them once: the visualizer offers the conversion when you open an
  old recording, or script it with the new voyant_recording_migrate tool.
• Pre-v1.0.0 tools open v1.0.0 recordings as empty — no frames, and no error
  explaining why. Update the visualizer, CLI tools, and bindings together.
• Old raw-peak CSV captures are unsupported. There is no converter; re-capture, or
  reach out to Voyant support about updating existing captures.

Installation

• Cap'n Proto is no longer a build dependency. The voyant-api-dev package, the
  Python wheel, and the ROS driver all install without building it from source.

Command-line tools

• Shipped tools lose their carbon qualifier: voyant_carbon_client_check →
  voyant_client_check, voyant_carbon_simulator → voyant_simulator, and likewise
  for the rest. The old names are gone, so scripts that call them need updating.
• The recorder's buffer-size knob is removed — the buffer now sizes itself. This
  drops voyant_logger_binary --buffer-size-mb, the Python buffer_size_mb argument,
  and the C++ bufferSizeMb field. The dead --send-buffer-size flag is also gone
  from the client tools.
• The CSV converters (voyant_bin_to_csv / voyant_csv_to_bin) are now
  internal-only and no longer publicly documented. For analysis exports, use the
  Python bindings' points() matrix or the ROS driver.
• voyant_hello_world prints the interface-contract version rather than a proto
  version.

Python

• points() returns a different column set.
• The per-field accessor families and the voyant_api.utils compatibility shim are
  removed.
• describe() returns a string instead of printing.
• The fourth column of xyzv() is renamed from radial_vel to doppler_mps.
• valid_mask() returns a NumPy boolean array.
• Playback's filter_points becomes keep_invalid_points; the PCD helper's
  valid_only likewise.
• The client config's set_keep_invalid_points becomes set_diagnostic_mode
  (capture-side only — the playback-side keep_invalid_points flags keep their name).
• VoyantClient is removed.
• The public peak-dump API (the start_peak_dump family) is replaced by diagnostic
  capture.
• The elevation-interpolation options are removed from the client config
  (set_missing_elevations, set_interp_range_threshold,
  set_interp_doppler_threshold) along with the pipeline stage behind them. The
  feature was off by default.

C++

• Frames are the native VoyantFrame class rather than the Cap'n Proto
  VoyantFrameWrapper: tryReceiveFrame() returns std::optional<VoyantFrame>
  directly and the separate latestFrame() accessor is gone; point data is read
  through points() / xyz(), and state through sensorState() / hostState().
• VoyantClient and PointsClient are removed.
• Playback's filter_points becomes keep_invalid_points, and its default is no
  longer inverted.
• CarbonConfig::setKeepInvalidPoints becomes setDiagnosticMode (capture-side only —
  the playback-side keepInvalidPoints flag keeps its name).
• The peak-dump API is replaced by diagnostic capture.
• The elevation-interpolation setters are removed from CarbonConfig
  (setMissingElevations, setInterpRangeThreshold, setInterpDopplerThreshold),
  along with the pipeline stage behind them. The feature was off by default.
• voyant_playback_is_file_open is removed from the C FFI with no C++ replacement.
  Python keeps is_open.
• C FFI only, no C++ change: voyant_recorder_create gained error output parameters,
  matching voyant_playback_open_file.

New Features

• voyant_recording_migrate converts pre-v1.0.0 recordings to .vynt.
• Diagnostic capture records a support bundle for Voyant support: a recording that
  keeps invalid points, paired with a raw-peaks sidecar file. It replaces the old
  peak-dump tooling in the visualizer and both bindings.
• Edit a recording and save it back. Read frames, change their points, and write a
  new recording — including a per-point user_data column that native recordings
  preserve verbatim.
• Build frames from your own points. Both bindings can now create synthetic frames
  (identified by a simulator device id) and stateless frames, for points that arrive
  without a sensor heartbeat behind them — a simulator scene, a ROS cloud, a CSV.
• Sensor and host state are readable from a frame — sensor_state() /
  host_state() in Python, sensorState() / hostState() in C++.
• The API and interface-contract versions are readable from both bindings,
  mirroring what voyant_hello_world prints.
• The recorder reports the file it is writing — the current_file_path property
  in Python, getCurrentFilePath() in C++ — with timestamped and split naming
  resolved, so a file that has to pair with a recording can be named from it.
• Visualizer: new color-by-field modes with in-panel legends, a marker highlighting
  picked points, combine-method and user-data rows in the point inspector, and a
  recording can be opened straight from the command line with --input. The
  pre-release banner is gone.
• C++ additions: setPointsXyz(), the supported inverse of VoyantFrame::xyz()
  for points that arrive as x/y/z, and VoyantRecorder::getLastError(), so an invalid
  recorder can say why it is invalid rather than only that it is.

Behavior Changes

• An existing recording is no longer overwritten unless you opt in. Recording to a
  path that already exists fails instead of clobbering it.
• Unusable playback rates play unpaced. A playback rate of zero, or one that is
  negative, non-finite, or below 0.001×, plays as fast as possible and logs a warning.

Bug Fixes

• Returns between the mirror center and the lidar datum are no longer recorded as
  valid points at a negative range. The range sanity check ran before the datum
  shift, so a near-field return that was valid at the mirror could land behind the
  datum and still be reported as valid. It is now checked on both sides of the shift
  and dropped. Point counts fall very slightly as a result — measured at roughly
  0.008–0.016% of otherwise-valid points on a Carbon 30 — and reflections from inside
  the sensor housing no longer appear in point clouds.
• Fixed a use-after-free in the C++ recorder, where finalizing and then destroying
  a recorder freed the same resources twice.