Python

deancochran-ftms is a pure, synchronous Python protocol package. This 0.1.0a2 is an alpha with an evolving API: it implements FTMS Features, all five supported ranges (raw and normalized), structural range inspection, raw Control Point requests/responses, all six raw measurement families, bidirectional Machine/Training Status, and static capability evaluation. The earlier 0.1.0a1 artifact has no capability API. It has no BLE, lifecycle, or logging APIs and is not a complete port of every API in the TypeScript/C packages.

Its wire support profile is FullWire raw codecs, with RangeInspection, selected NormalizedViews, and CapabilityEvidence. FullWire does not mean cross-language convenience parity.

Integrate through the consumer adapter seam: transport conversion, BLE/session lifecycle, retries, subscriptions, UI/application policy and control safety remain outside this protocol package.

Install and compatibility

The distribution name is deancochran-ftms; import deancochran_ftms. It has no runtime dependencies and declares Python >=3.11. Python 3.11 and 3.14 are tested by the package verification commands in this milestone.

In a virtual environment, explicitly opt into current prereleases from PyPI:

python -m pip install --pre deancochran-ftms

The interface remains alpha. Inspect the resolved version and lock dependencies with your project’s tooling; pip install alone does not create a lockfile. For a future stable release, omit --pre. To reproduce the verified alpha exactly:

python -m pip install 'deancochran-ftms==0.1.0a2'

from deancochran_ftms import decode_features, encode_features_raw, FeaturesRaw

wire = encode_features_raw(FeaturesRaw(machine=0, target=1 << 3))
result = decode_features(wire)
assert result.ok and result.value is not None
assert result.value.power_target_setting_supported

decode_features_raw() and encode_features_raw() are strict wire codecs and raise RawCodecError for wrong byte inputs, non-8-byte payloads, or invalid raw words. decode_features() instead returns an immutable FeatureDecodeResult: wrong lengths produce a FeatureDiagnostic with code="length". Feature payloads are exactly eight bytes; trailing bytes are rejected, matching the canonical TypeScript Feature behavior. bytes, bytearray, and contiguous one-dimensional byte memoryview inputs are accepted and copied as immutable evidence. Raw words retain all unknown/reserved bits. Integer raw words must be plain int values from 0 through 2^32-1; bool is rejected.

The normalized Features model exposes the canonical v1 feature names in snake_case. Its three convenience properties (supports_erg, supports_sim, and supports_resistance) correspond to the v1 compatibility names.

Static capability evidence

evaluate_capabilities(snapshot, options=None) is a pure interpretation of one immutable, caller-owned CapabilitySnapshot. It accepts ordered immutable CharacteristicEvidence observations and returns an immutable CapabilityReport. The report preserves discovery/scope uncertainty, duplicates, raw feature words, unknown bits, range evidence, all 21 operation prerequisites, and diagnostics. report.to_wire() produces fresh JSON-compatible containers in the exact shared capability-corpus representation. It deliberately has no canExecute field: BLE I/O, encryption, indication setup, control ownership, retries, and safety approval remain outside this package. The evaluator supports explicit C.7 C7Evidence; unknown is not false.

As with standalone range decoding, pass RangeFormatOptions(resistance_format="signed16Tenths") to explicitly select the six-byte resistance range alternative. The default is three unsigned whole-level bytes; this selection affects only the resistance range and is never inferred from observed bytes or Feature bits.

This API requires 0.1.0a2 or newer, available on PyPI. You can also install from the source checkout (python -m pip install ./packages/python from the repository root). Publication evidence is tracked separately in the release matrix.

from deancochran_ftms import (
    CapabilitySnapshot,
    CharacteristicEvidence,
    DiscoveryState,
    ReadState,
    ServiceScope,
    evaluate_capabilities,
)

# The application supplies discovery/read evidence; this does not use Bluetooth.
feature = CharacteristicEvidence(
    uuid="00002acc00001000800000805f9b34fb",
    properties=0x02,  # Read
    read_state=ReadState.SUCCESS,
    read_bytes=bytes(8),  # Valid all-zero feature words, not a missing read.
)
report = evaluate_capabilities(
    CapabilitySnapshot(
        discovery=DiscoveryState.COMPLETE,
        scope=ServiceScope.PRESENT,
        generation=1,
        characteristics=(feature,),
    )
)
assert report.observation_count == 1
assert len(report.operations) == 21
assert "canExecute" not in report.to_wire()

UUIDs use 32 lowercase hexadecimal characters without hyphens. Read payloads must be immutable bytes, observations must be a tuple, and state fields use the exported enums. Positional report rows follow the shared capability report layout. Missing Control Point/Machine Status evidence and unknown C.7 facts in this example are retained as diagnostics, not silently treated as satisfied.

Ranges and inspection

decode_supported_range_raw(data, kind, options=None) and encode_supported_range_raw(value, options=None) operate on SupportedRangeRaw. Kinds are speed, inclination, resistance, heartRate, and power. Values retain integer numerators, scale_divisor, and units; the normalized decode_supported_range() divides these values into physical units. All three raise RawCodecError for invalid arguments, lengths, or range values.

Resistance ranges default to three unsigned whole-level bytes. Select the six-byte signed-tenths alternative explicitly with RangeFormatOptions(resistance_format="signed16Tenths"); this option is invalid for other range kinds. Range units do not imply resistance percentages.

inspect_supported_range_raw() returns the shared inspection report shape: selected profile, actual/expected lengths, status, selected value, and ordered structural candidates. Malformed wire lengths/values appear as candidate statuses; invalid caller kinds/options still raise. Report values use the canonical numeric unit identifiers (0 speed, 1 inclination, 2 resistance, 3 heart rate, 4 power). The returned dictionary is caller-owned; selected and candidate values do not alias. A valid alternative candidate never selects a profile automatically or establishes physical units on a particular machine.

Control Point requests and responses

from deancochran_ftms import (
    ControlRequestRaw,
    decode_control_request_raw,
    encode_control_request_raw,
    decode_control_response_raw,
)

# Set Target Power: opcode 0x05, raw operand in watts. This does not send anything.
wire = encode_control_request_raw(ControlRequestRaw(0x05, (75,)))
assert wire == bytes.fromhex("05 4b 00")
assert decode_control_request_raw(wire).operands == (75,)
response = decode_control_response_raw(bytes.fromhex("80 05 01"))
assert response.request_opcode == 5 and response.result_code == 1

All 21 request opcodes support raw encode/decode. Operands must be tuples of plain integers in wire units, not arbitrary human-unit floats. For example simulation operands are wind speed in thousandths of m/s, grade in hundredths of percent, rolling coefficient in ten-thousandths, and wind coefficient in hundredths kg/m. There is no public normalized control encoder in this milestone; the legacy codec-v1 test adapter translates its normalized fixtures to raw operands.

ControlFormatOptions(resistance_format="uint8Tenths") explicitly selects the alternative resistance command width; the default is signed16Tenths. Control options are independent of range options and do not change other opcodes.

ControlResponseRaw preserves request/result codes plus integer diagnostic flags unknown_request, unknown_result, and unexpected_parameters. Its parameter is 0 for none or 1 for successful spin-down speeds (low/high in hundredths km/h). encode_control_response_raw() accepts only canonical response forms; decodable malformed/unknown evidence is not necessarily encodable. Invalid wire headers/lengths and invalid encoder arguments raise RawCodecError.

No codec acquires control, sends bytes, retries commands, or authorizes execution.

Development verification

From packages/python/ in a full repository checkout (with uv installed):

uv run --locked --group dev python scripts/verify.py

That aggregate command runs both supported test interpreters, quality checks, all codec corpus runners, the structural matrix, and isolated installation. Individual checks are also available:

uv run --locked --group dev pytest
uv run --locked --group dev ruff format --check .
uv run --locked --group dev ruff check .
uv run --locked --group dev mypy src tests scripts
uv run --locked --group dev python scripts/run_features_conformance.py
uv run --locked --group dev python scripts/run_measurement_status_conformance.py
uv run --locked --group dev python scripts/run_measurement_matrix.py
uv run --locked --group dev python scripts/run_capability_conformance.py
uv run --locked --group dev python scripts/verify_package.py

The conformance test reads ../../shared/conformance/v1 and ../../shared/conformance/values/v1 directly; it never copies fixtures. It executes all codec-v1 categories and the additive values, controls, inspection, compatibility, measurements and statuses contracts directly from shared fixtures. The capability runner validates the separate canonical schema and executes all 63 exact reports directly from ../../shared/conformance/capabilities/v1. scripts/verify_package.py builds both artifacts, rebuilds from an extracted sdist outside this checkout, installs the wheel non-editably into a fresh environment, and writes its distinct machine-readable artifact hashes, interpreter, and per-step install/rebuild evidence to ignored build/package-verification-report.json. Conformance identity and case evidence are separately recorded in build/verification-report.json and build/measurement-status-verification-report.json. The structural matrix records 181,760 layouts in both directions, 46 sentinel cases, 47 RFU cases, and 315 incomplete prefixes in build/measurement-matrix-verification-report.json. These generated cases are not separate pytest test registrations. The sdist intentionally contains no shared corpus dependency: installing or building the package needs no checkout, but running the conformance suite does. The package check asserts that rebuilding the extracted sdist produces an identical wheel; it does not assert byte-for-byte reproducibility of the sdist archive itself.

This is host protocol regression evidence only, not live control execution, device interoperability or Bluetooth qualification. Public availability is separate from these host verification results.

Coverage and limits

Version 0.1.0a2 is an alpha with language-specific API coverage. Features, ranges, controls, all six measurement families and both status characteristics have raw codecs. Normalized measurement/status projections and normalized range decoding are available. Static capability evaluation is included starting with 0.1.0a2. Range, control and measurement format choices are independent and explicit; no API infers them from packet length, feature declarations or BLE state. Host fixtures establish codec regression evidence, not live-device compatibility or qualification.

Measurements and statuses

MeasurementRaw is immutable raw evidence: flags, 30 wire-unit values, and present/unavailable bit masks retain field-level availability. Use decode_measurement_raw(data, kind, options=None) and encode_measurement_raw(value, options=None). Kinds are 0 Treadmill, 1 Cross Trainer, 2 Step Climber, 3 Stair Climber, 4 Rower, and 5 Indoor Bike. Decoders retain More Data, Cross Trainer backward direction, RFU, truncation, and trailing byte diagnostics. Encoders validate the wire flags, presence/unavailable masks and selected values; auxiliary decode diagnostics do not themselves authorize or prevent encoding, matching the shared raw contract. Compatibility formats are explicit only: MeasurementFormatOptions(resistance_format="signed16Tenths") and treadmill_pace_format="uint8Legacy"; they are independent selections and are never inferred from packet length, features, or machine identity.

from deancochran_ftms import MeasurementRaw, decode_measurement_raw, encode_measurement_raw

raw = MeasurementRaw(5, 0, 1, 0, (1234,) + (0,) * 29)  # speed: 12.34 km/h
assert encode_measurement_raw(raw) == bytes.fromhex("00 00 d2 04")
assert decode_measurement_raw(bytes.fromhex("00 00 d2 04"), 5).values[0] == 1234

Machine Status uses MachineStatusRaw with raw opcode/action and optional (control_opcode, operands) parameter tuple. Training Status uses TrainingStatusRaw; text is UTF-8 bytes, so malformed UTF-8 is retained and diagnosed by invalid_utf8 rather than silently replaced. The corresponding decode_*_status_raw APIs preserve partial/trailing/reserved evidence; encoders accept only canonical values and raise RawCodecError on invalid widths, flags, UTF-8, sentinels, or diagnostics. Raw units/scales remain FTMS wire units.

Host tests cover Python 3.11 and 3.14 plus shared literal measurement/status corpora. This finite evidence is not BLE, real-device, PTS, or Bluetooth qualification evidence.

normalize_measurement(raw, options=None) provides the immutable-raw-to-codec-v1 normalized metric projection (for example raw speed / 360 to m/s, cadence / 2, and unavailable or incomplete fields as None). normalize_machine_status() and normalize_training_status() provide the corresponding normalized status views; raw codecs remain the source of wire evidence.