UnifiedFloppyTool v4.1.5

UnifiedFloppyTool v4.1.5 Release Notes

Release Date: 2026-06-05
Tag: v4.1.5 (squash-merge of tests/v4.1.5-hardening onto main,
39 commits MF-240..MF-276 plus .claude hygiene v2/v3 plus Phase A/B/D
demo-ready bundle).
Version-string convention: VERSION.txt carries the numeric
4.1.5 per the existing release convention.

Hardware verification status (read this first)

The Greaseweazle production path (src/hal/uft_greaseweazle_full.c) is
byte-identical to the v4.1.4-rc1 commit that was hardware-verified
2026-05-15 — zero lines changed across the 39 squashed commits, the
file is in .claude/CLAUDE.md's protected-paths list. HIL.GW
formal-bench session (tests/hil/run_hil.py) is deferred to the
post-tag window; substantively the HIL output for v4.1.5 would be
identical to v4.1.4-rc1's because the GW code is unchanged.

UFT-008 SCP-Direct Tier-3 verification still requires real SCP
hardware and is explicitly v4.1.6 scope — see docs/CAPABILITIES.md
for the per-controller status matrix.

What changed (TL;DR)

v4.1.5 hardens the v4.1.4 release across 9 dimensions without
breaking any v4.1.4 user-facing behavior. The Greaseweazle workflow is
unchanged. Everything new is either honesty (more accurate status
labels), structural correctness (ABI gate, single-chokepoint
preflight), or scaffolding (Tier-2.5 simulators, sidecar reports).

Code-side gates closed in this release (V415_GOAL_PLAN.md):

Gate	Status	Impact
All 9 V2 providers have a live code path	✅ MF-249..MF-258	No silent no-op when user clicks "Connect" on a non-Greaseweazle controller
22/22 SCP USB opcodes byte-verified	✅ MF-254 + MF-261	Cross-referenced against samdisk/SuperCardPro.h. Pre-MF-254 placeholder bytes (0x02-0x40) are replaced with the documented 0x80-0xD2 opcode space
SCP-Direct M3.1 read-flux payload parsing	✅ MF-276	Full samdisk-reference algorithm: CMD_READ_FLUX → CMD_GET_FLUX_INFO → CMD_SENDRAM_USB; 16-bit big-endian samples decoded with 0x0000-overflow accumulator, reset per revolution. 9/9 mock-tests pass. impl_complete stays false until real-HW bench
uft_format_plugin_t ABI gate	✅ MF-260	New api_version field + _Static_assert(sizeof == 216) + registrar reject for future-ABI plugins
Plugin Prinzip-7 compliance 84/84	✅ MF-262 + MF-263	spec_status and features populated for every plugin (was 15/84 and 5/84 respectively)
LOSS.preflight Phase 1 + 2	✅ MF-263 + MF-268	All 44 conversion paths gated through uft_preflight_check() at the uft_convert_file() chokepoint. LOSSY_DOCUMENTED paths emit a .loss.json sidecar after successful convert with category-level loss entries
Tier-2.5 hardware simulator system	✅ MF-267 + MF-270	KryoFlux + FluxEngine + FC5025 via Python "fake binary" sims, SCP-Direct + XUM1541 via libusb-mock framework, Greaseweazle via serial-protocol sim. 10/10 SIMULATED in CI
ARCH-7-C Teensy-probe wrapper	✅ MF-213 + MF-263	Probe-on-Connect with mismatch warning for ADFCopy ↔ Applesauce (both Teensy 0x16C0:0x0483) — prevents disk-corrupting cross-wire
Test pass-rate 47/180 → 153/153	✅ MF-245..MF-260 + MF-270 + MF-276	Stale exclusions reviewed; new test_plugin_abi.c, test_scp_direct_usb_mock.c (9 tests), test_xum1541_usb_mock.c
Build-system parity baseline shrunk	✅ MF-262 + MF-272	verify_build_sources.py baseline 224 → 219 → 216 (LOSS.preflight sources wired)
qmake .pro wiring fixed	✅ MF-259 + MF-266	preflight/loss_report/roundtrip sources, libusb IMPORTED target, audit-vector pins
Phase A demo-ready docs bundle	✅ MF-273 + MF-274	docs/SHOWCASE.md (179 LOC), docs/CAPABILITIES.md (189 LOC), docs/demo/QUICK_DEMO.md (253 LOC), docs/KNOWN_ISSUES.md Demo-Impact annotation, docs/release/v4.1.5_github_release_body.md

Highlight: SCP-Direct M3.1 read-flux implementation (MF-276)

uft_scp_direct_read_flux() was the last NOT_IMPLEMENTED in the SCP
M3.1 scaffold. v4.1.5 implements it as a byte-for-byte port of the
samdisk SuperCardPro::ReadFlux() reference (the de-facto open-source
SCP USB reference implementation).

Algorithm:

1. CMD_SIDE [side]
2. CMD_READ_FLUX [revs, ff_Index=0x01] — waits on index pulse
3. CMD_GET_FLUX_INFO → 40-byte rev_index table (5 pairs of
   [index_time_be32, flux_count_be32])
4. Per revolution: CMD_SENDRAM_USB [offset_be32, length_be32] →
   flux_count × uint16_t big-endian samples
5. Decode: 0x0000 accumulates 0x10000 ticks (cell overflow);
   non-zero emits (accum + sample) × 25 ns; accumulator resets
   per revolution (matches samdisk — each capture starts fresh
   after its index pulse)

uft_scp_direct_write_flux() remains intentionally NOT_IMPLEMENTED.
A malformed flux stream can physically damage forensic media; write is
blocked at the HAL boundary until the read path is bench-verified
against real SCP hardware. Forensic safety overrides feature-
completeness.

Mock-test coverage (9 tests in tests/test_scp_direct_usb_mock.c):
open/close, seek with packet format verification, error-status handling,
bulk-out timeout, read_flux two-revs happy-path with full 5-step wire
dance, overflow accumulation ({0000, 0010, 0000, 0000, 0020} → 2
emits), single-revolution rejection (revs=1 → UFT_ERR_INVALID_ARG,
no silent coercion), per-revolution accumulator-reset regression-guard.

Highlight: Tier-2.5 hardware simulator system

The biggest forensic-confidence win of v4.1.5: all V2 providers can
now be exercised end-to-end without owning the corresponding hardware.

Three different mocking strategies for three different transport layers:

Transport	Mocking strategy	Coverage
QProcess (CLI subprocess: KryoFlux DTC, FluxEngine, FC5025 fcimage)	Python fake-binary scripts on PATH	7/7 SIMULATED
libusb-direct (SCP-Direct, XUM1541)	Scripted libusb-mock framework (tests/usb_mock/), per-target wrapper TU, zero production-code changes	5+3 = 8 SIMULATED
USB-CDC virtual serial (Greaseweazle, Applesauce, ADFCopy)	Python protocol simulators paired with com0com/socat	3 byte-compile-verified

The libusb-mock framework is novel: a libusb_mock.h header that
pre-defines LIBUSB_H (the real libusb's include-guard) so any later
#include <libusb-1.0/libusb.h> becomes a no-op. The mock provides
minimal libusb types + a scripted-exchange queue that test code
populates. Production C-HAL code is bytewise unchanged.

Coverage runs in CI on every push via the Audit job, reporting
SIMULATED / FAIL / NOT_RUN per controller. SIMULATED is never
reported as PASS — real hardware (Tier 3) remains the only PASS
authority.

Highlight: LOSS.preflight — Prinzip 1 enforcement

uft_convert_file() is the single chokepoint for all 44 conversion
paths. v4.1.5 wires uft_preflight_check() into that entry point
(Phase 1, MF-263) plus .loss.json sidecar emission for successful
LOSSY_DOCUMENTED conversions (Phase 2, MF-268).

• LOSSLESS runs silently as before.
• LOSSY_DOCUMENTED requires accept_data_loss=true — the GUI/CLI
  must obtain explicit user consent. On success, a category-level
  <target>.loss.json is written: FLUX_TIMING, WEAK_BITS,
  INDEX_PULSES, MULTI_REVOLUTION, plus a path-quality warning.
• IMPOSSIBLE and UNTESTED ABORT with a diagnostic.

Per-track exact loss counts (e.g. "23 weak bits in track 18 sector 4")
are v4.1.6 follow-up — Phase 2 emits category-level entries, sufficient
for the user to know "this converter lost flux timing".

Highlight: SCP USB protocol — VENDOR-DOCUMENTED status

MF-261 cross-referenced all 22 SCP USB command bytes + response code +
checksum init in include/uft/hal/uft_scp_direct.h against
samdisk's SuperCardPro.h.
samdisk has shipped SCP read/write against real hardware since 2017
and is the de-facto open reference implementation.

Pre-MF-254 the header carried six placeholder opcodes (0x02-0x40)
traceable to a8rawconv, not to the vendor SDK. The current opcodes
0x80-0xD2 (CMD_SELA..CMD_SCPINFO), PR_OK=0x4F, CHECKSUM_INIT=0x4A
are byte-exact with samdisk. MF-276 adds CMD_GET_FLUX_INFO=0xA1,
CMD_SENDRAM_USB=0xA9, and ff_Index=0x01 flag (also byte-exact).

Highlight: Plugin Prinzip-7 compliance

Before v4.1.5 only 5 plugins had a hand-curated .features array
(ADF, HFE, IPF, STX, WOZ) and 15 had a non-UNKNOWN .spec_status.
The audit_plugin_compliance.py script flagged the gap as a §7.1 +
§7.2 violation.

Two mass-populator scripts solved it:

• scripts/populate_spec_status.py — per-format-subdir mapping
  (OFFICIAL_FULL / OFFICIAL_PARTIAL / REVERSE_ENGINEERED / DERIVED)
  based on documentation provenance.
• scripts/populate_features.py — derives a per-plugin feature
  matrix from .capabilities bitfield (Read/Write/Create/Flux/
  Timing/Weak Bits/MultiRev → SUPPORTED or UNSUPPORTED).

Result: 84/84 (100%) Prinzip-7 compliance. Per-plugin hand-tuning
where the heuristic was too coarse (e.g. WOZ PARTIAL with explanatory
note) is ongoing in v4.1.6 as documentation hygiene.

Resolved structural issues

ID	Pre-v4.1.5 state	Resolution
UFT-002	CMakeLists.txt header comment "Version: 4.1.3" stale	Removed numeric; refers to VERSION.txt SSOT
UFT-003	HardwareTab honest-stub providers visually identical to production GW	Orange "Disconnect (Beta)" Button + tooltip distinct from green production GW
UFT-004	uft_format_plugin_t had no API-version field; ABI-bomb risk	api_version + _Static_assert(sizeof==216) + registrar gate
UFT-005	test_transitions_ns_contract covered only MockProviderV2	KryoFlux + FluxEngine contract-shield FFI added
UFT-006	.claude/CLAUDE.md listed 6 controllers, code had 9 V2-Providers	List corrected with per-provider status
UFT-007	Claimed VID/PID mismatch between header + GUI-hint	Verified consistent (MF-212): 0x16D0:0x0F8C via shared macro
UFT-T01	<threads.h> include broke MinGW build	__has_include guard added
UFT-T02	4 tests linked phantom symbols	Per-test target_sources wired correctly
UFT-T04	43 tests excluded from build	Stale exclusions reviewed; 4 re-enabled (test_scp_direct_hal, test_applesauce_hal, test_fnmatch_shim, test_whdload_resload). Remaining 38 are MF-011-deleted-impl orphans, documented per-line
UFT-T05	events CMakeLists Standalone-Pfad bug	Verified already fixed (uses CMAKE_CURRENT_SOURCE_DIR)
ARCH-2	KryoFlux/FluxEngine providers fabricated FluxCaptured from container bytes	Closed in MF-208 (KF) + MF-209 (FE); shield test in MF-260 prevents regression
SCOPE	src/switch/ (10MB Nintendo Switch + mbedtls) off-mission	Decision: Option C — Delete, post-tag (MF-271). Backup tag planned: archive/pre-mf271-switch-removal

What this release does NOT ship

These are intentionally deferred to v4.1.6 (with reasoning):

• HIL.GW formal bench session — production GW code unchanged so
  substantive risk is null, but the formal report
  (releases/v4.1.4/hil_report.md) is still pending.
• UFT-008 SCP-Direct Tier-3 verify — needs real SCP hardware
• M3.2 XUM1541 libusb — wired but Tier-3 hardware verification needs opencbm + real device
• M3.3 Applesauce serial-protocol full handshake — ?vers is in; the ?disk-readx state machine still has gaps
• UFI Windows + macOS backends (ufi_win.c via DeviceIoControl, ufi_mac.c via IOKit). Linux SG_IO is in (MF-207)
• ARCH-9 XUM1541 macOS .dylib loader — needs OpenCbmLibrary::load() path
• 5-and-3 Apple GCR (DOS 3.2 13-Sektoren)
• FM-Decoder completion — flux_decode_fm flagged as incomplete in the decoder source
• Per-converter sidecar precision — current Phase 2 emits category-level entries; per-track exact counts are v4.1.6
• uft-decode CLI binary build wiring — scaffold at cli/uft-decode/ ships; CMake/qmake enable + emulator.yml round-trip integration is v4.1.6
• MF-271 Switch/cart7 delete — decided 2026-05-25, execution post-tag
• USBFloppy SG_IO mock — Tier-2.5 framework supports it but SG_IO-mock not yet written; v4.1.6

Upgrade notes

• No breaking changes. v4.1.4 → v4.1.5 is purely additive.
• The uft_format_plugin_t ABI gate (MF-260) is opt-in: legacy plugins
  with api_version=0 are accepted with a one-shot stderr warning.
  Plugin authors should set .api_version = UFT_PLUGIN_API_VERSION
  before v5.0.
• LOSS.preflight requires explicit consent for LOSSY_DOCUMENTED
  conversions. GUI workflows that previously ran lossy converts
  silently now need to set accept_data_loss=true in
  uft_convert_options_t (preserve_errors field maps to this).
• uft_scp_direct_read_flux() rejects revolutions=1 with
  UFT_ERR_INVALID_ARG (SCP firmware requires ≥2; honest, no silent
  coercion). Provider layer
  (src/hardware_providers/scp_provider_v2.cpp) still accepts
  revolutions=1 in its parameter range — open audit finding to be
  reconciled in v4.1.6 (see docs/KNOWN_ISSUES.md).

Hardware Compatibility

Detailed per-controller capability matrix (Tier-3 / Tier-2.5 / mock-only /
scaffold) lives in docs/CAPABILITIES.md.

Status quick-glance:

• 1 production: Greaseweazle (Tier-3 PASS, read+write)
• 3 read-real via subprocess wrapper: KryoFlux (DTC), FluxEngine (CLI), FC5025 (fcimage)
• 3 libusb-wired, mock-only: SCP-Direct, XUM1541, Applesauce (byte-protocol validated)
• 1 USB-CDC-wired with sim: ADF-Copy
• 1 Linux-only: USB-Floppy (SG_IO; Win/Mac backends are v4.1.6)

Any provider not in Tier-3 PASS reports not_implemented for unwired calls
— never a silent no-op. GUI shows orange "Preview" badge instead of
the green Production badge.

Verified on (CI green)

• Local: MinGW-w64 g++ 13.1.0, Qt 6.10.1
• CI: Linux GCC (Qt 6.7.3 + 6.10.1 with c++17 + c++20), macOS Clang, Windows MinGW
• Plugin compliance: audit_plugin_compliance.py 84/84 ✓
• Build parity: verify_build_sources.py 216/0
• Consistency: check_consistency.py 0/0/0/0 ✓
• Test suite: ctest 153/153 (100%)
• Tier-2.5 simulator gate: 10/10 SIMULATED (Audit CI job)
• All 5 CI workflows green on tests/v4.1.5-hardening HEAD before squash (Audit · Emulator-CI · Sanitizer · Coverage · CI)

Commit ledger

39 commits squashed from tests/v4.1.5-hardening (MF-240..MF-276 plus
.claude hygiene v2/v3 plus Phase A/B/D bundles plus hermes-PoC park):

MF-240    docs(release): pre-tag cleanup
MF-241/2  docs/release alignment
MF-243    build(release): uft_p24_verify.sh — 4-BEFUND verification gate
MF-244    build(hil): goal v4.1.4-final Phase-3 infrastructure
MF-245/6  v4.1.5-hardening: full audit findings — 26%→100% pass rate
MF-247    UFT-003 HardwareTab honest-stub visual + UFT-007 verified
MF-248    M3.3 concrete Applesauce production-transport plan
MF-249    M3.3 QSerialPort-backed Applesauce transport
MF-250    M3.3 Applesauce live end-to-end + HardwareTab wiring
MF-251/2/3 M3.3 Applesauce sim + ADFCopy transport — UFT-001 3/9
MF-254    M3.1 SCP-Direct libusb wiring — UFT-001 4/9
MF-255    M3.2 XUM1541 libusb wiring — UFT-001 5/9
MF-256    M3 FluxEngine + KryoFlux QProcess runners — UFT-001 7/9
MF-257    M3 FC5025 fcimage QProcess runners — UFT-001 8/9
MF-258    M3.4 USB Floppy UFI runners — UFT-001 9/9 COMPLETE
MF-259    CI qmake .pro wiring + libusb IMPORTED + audit-vector pins
MF-260    UFT-004 plugin ABI gate + UFT-005 KF/FE shields + UFT-T04 reductions
MF-261/2  V415-PLAN — SCP.D1 verify + spec_status 100% + BUILD.rebaseline
MF-263    V415-PLAN deep — features 100% + LOSS Phase 1 + EMUCI/release scaffolds
MF-264    fix stray-comma after trailing comment in 79 plugin literals
MF-265    UFT_FORMAT_ID_T_DEFINED guard in roundtrip.h + format_validate.h
MF-266    add preflight + loss_report + roundtrip sources to .pro
MF-267    Tier-2.5 hardware simulator system (KryoFlux/FluxEngine/FC5025)
MF-268    LOSS.preflight Phase 2 + ADFCopy sim + CI gate + RELEASE_NOTES draft
MF-269    hw simulators self-contained on CI fresh checkout
MF-270    Tier-2.5 USB-mock framework (SCP/XUM/GW without hardware)
MF-272    rebaseline build_system_baseline (3 resolved entries)
MF-273    Phase A demo-ready bundle (SHOWCASE+CAPABILITIES+KNOWN_ISSUES Demo-Impact+RELEASE_NOTES section)
MF-274    Phase C.1 QUICK_DEMO.md (10-min demo script pull-forward)
MF-275    Phase B/D pre-work bundle + rescue ignored release files
MF-276    M3.1 SCP-Direct read-flux payload parsing — full samdisk impl + per-rev-reset fix
+ chore(.claude): hygiene-pass v2 (a0f8187) — agent count sync + EOL + htb-mentor flatten
+ chore(.claude): hygiene-pass v3 (db14e70) — Eigenständigkeit, Eigenverantwortung, Vollständigkeit
+ docs(.claude): park hermes-agent PoC goal for post-v4.1.5-tag
+ spike(proto/hermes-eval): Session #1+#2 (skeleton + prompts + rubric + hermes CLI research)

---

UnifiedFloppyTool v4.1.4 Release Notes

Release Date: 2026-05 (v4.1.4-rc1 pre-release tag first;
14-day window; promote to v4.1.4 final on success).
Branch: main already carries the type-driven HAL refactor
(PR #24 squash-merged) and the 4 Phase-2 hardening commits
(PR #26 merged 2026-05-15). The v4.1.4 tag goes on main HEAD —
no further merge step required.
Version-string convention: VERSION.txt carries the numeric
4.1.4. The -rc1 / -rc2 suffix lives only in the git tag
(v4.1.4-rc1, v4.1.4), per the project's existing release-notes
header convention.

What changed (TL;DR)

• Working Greaseweazle workflow: unchanged behavior, structurally
  hardened internals. Read / write / detect / motor / seek / RPM /
  recalibrate all preserve every byte they used to. Auto-detect on
  Connect still happens (now via the V2 surface).
• 5 X1541-family controller-combo entries gone — XA1541 / XAP1541
  / XM1541 / XE1541 / X1541. They never had a working backend in any
  release; selecting them only produced an error dialog. Removed for
  honesty.
• 8 non-Greaseweazle controllers (SCP, KryoFlux, FluxEngine,
  FC5025, XUM1541, Applesauce, ADFCopy, USBFloppy) now display a
  clear "Controller routing pending" message on Connect instead of
  silently no-op'ing. Their V2 wrappers exist, are conformance-
  tested, and will be wired in v4.1.5 (task P1.23).

Why this is v4.1.4 and not v5.0.0

The Type-Driven HAL refactor underneath this RC is substantial — 10
mixin-composed V2 providers, codegen-driven GUI wiring, 65-section
conformance harness, ~12 700 LOC of V1 hierarchy deleted. From a USER
perspective the change set is:

1. The Greaseweazle pipeline is preserved + hardened.
2. Phantom-feature combo entries are cleaned up (UI more honest).
3. Non-GW controllers display a "pending" message instead of silent
   no-op (UI more honest).

None of those break a working user workflow. UFT is an end-user
application — no third-party code links the deleted C++ classes —
so there is no external SemVer contract to honor. The PATCH bump
v4.1.3 → v4.1.4 reflects "hardening of the v4.1 line" honestly;
the refactor magnitude is documented in CHANGELOG.md's v4.1.4-rc1
section, in docs/REFACTOR_BRIEF.md (the architecture spec), and in
this release-notes file. Full rationale in REFACTOR_BRIEF.md §9.

Hot-fixes included (pre-existing 4.1.4 plan)

• Windows COM-port \\.\ prefix bug across all serial-port providers
  (MF-145).
• Applesauce ↔ ADFCopy USB VID/PID disambiguation (MF-146).
• Greaseweazle bootloader-firmware-mode detection (MF-129).
• MFM IDAM/DAM standalone sector parser for SCP→ADF/IMG/D64 paths
  (MF-141 / AUD-002).

Quality gates that ship with this RC

• 14 test executables, all green:
  test_hal_foundation, test_hal_conformance (65 sections),
  test_greaseweazle_v2, test_scp_provider_v2,
  test_kryoflux_provider_v2, test_fluxengine_provider_v2,
  test_fc5025_provider_v2, test_xum1541_provider_v2,
  test_applesauce_provider_v2, test_adfcopy_provider_v2,
  test_usbfloppy_provider_v2, test_mock_provider_v2,
  test_mock_hardware, test_wiring_runtime.
• scripts/check_consistency.py 0/0/0/0.
• scripts/verify_build_sources.py clean (no new divergence).
• tools/wiring_codegen_tests/run_tests.py 6/6 (incl. drift
  detector for generated/tab_hardware_wiring.gen.cpp).
• CI: green on Linux GCC (Qt 6.7.3 + 6.10.1), macOS Clang,
  Windows MinGW — all 3 platform jobs.

Pre-release window

This is v4.1.4-rc1. 14-day window starts on tag. During the window:

• Hand-test on real Greaseweazle hardware
  (tests/HARDWARE_TRUTH_TESTS.md checklist).
• File any regression in working v4.1.3 workflows as a release-blocker.
• Watch CI runs on the refactor/type-driven-hal branch.

If nothing breaks in 14 days: squash-merge → v4.1.4 final tag.

Known limitations of this release

What v4.1.4 honestly does not ship — flagged here so users and
reviewers know what is intentionally absent, not silently broken
(MF-235 / MF-240 follow-up to the original "queued for v4.1.5" block
that pretended these were just architectural polish):

• M3.x production transports — not in this release. Only
  Greaseweazle has a real USB/serial transport wired through. The
  other 8 controllers (SCP-Direct, KryoFlux, FluxEngine, FC5025,
  XUM1541, Applesauce, ADFCopy, USBFloppy) have V2 wrappers that are
  conformance-tested against the Mock surface but display
  "Controller routing pending" on Connect. M3.1 (SCP USB), M3.2
  (XUM1541 USB), M3.3 (Applesauce serial) and the remaining
  transports are v4.1.5 scope — see docs/MASTER_PLAN.md §M3.
• ARCH-7 sub-B / sub-C field-validation follow-ups. Sub-B
  (SCP 0x16D0:0x0F8C) and sub-C (ADF-Copy / Applesauce stock-Teensy
  0x16C0:0x0483 two-tier probe) are implemented in code
  (MF-212 / MF-213) and unit-tested; they have not yet been
  re-verified end-to-end on real hardware after the type-driven HAL
  refactor. The single-device readout that produced the original
  ID values still stands, but a post-refactor HIL pass on SCP +
  ADF-Copy + Applesauce is queued for v4.1.5. See
  audit/ARCH-7_VID_PID.md.
• §6.1 Emulator-CI — not in this release. The HIL runner
  (tests/hil/run_hil.py) has a two-tier Golden-Reference catalog
  (CI-runnable software tier + hardware tier), but no emulator
  loop (Greaseweazle firmware simulator, virtual SCP/KryoFlux
  device, WinUAE / VICE harnesses) runs in GitHub Actions yet. The
  hardware tier therefore stays NOT_RUN in CI by design; only the
  software tier is exercised. Emulator-CI is planned for v4.3 —
  see docs/M3_HAL_PLAN.md §M3.5.
• ARCH-8 — official protocol sources not yet vendored. GW
  cmd.h, SCP SDK, OpenCBM and fluxengine / DTC are currently
  declared recalled (working from memory + observed behaviour)
  in the audit-compliance matrix. Promoting them to vendored is
  what turns audit.yml into a real CI gate — until then the gate
  is advisory. Vendoring is v4.1.5+ scope.
• §1.2 Loss-Report wiring incomplete. The 44 convert_*
  format-conversion paths still need ~10 lines of glue each to
  propagate "data that did not survive" into the audit trail.
  Lossy conversion (Flux → Sector drops timing + weak bits) is
  labelled as lossy at the format-graph level, but per-conversion
  loss reports are not yet emitted end-to-end. v4.1.5+ scope.
• Plugin-Compliance matrix sparsely populated. Of 80 plugin
  registrations: spec_status is populated for 15, features for
  5, compat_entries for 1, is_stub flag for 0/287 stubs. The
  data model is in place (MF-187 onward), filling it iteratively
  is 4.2.x scope.
• ARCH-9 — XUM1541 macOS .dylib load path. The libxum1541
  shared-object lookup on macOS does not yet match the system's
  Homebrew / /usr/local/lib conventions. Linux + Windows paths
  work; macOS users currently need a manual DYLD_LIBRARY_PATH.
  v4.1.5 scope.

These are real gaps, not paperwork. v4.1.4 is shippable because the
Greaseweazle workflow — the one path used in practice — is
preserved + hardened, and the 8 non-GW controllers show an honest
"pending" message instead of silently no-op'ing. Anyone relying on
a non-GW transport should wait for v4.1.5.

---

UnifiedFloppyTool v4.1.3 Release Notes

Release Date: 2026-04-16

Highlights

• CRC / status flag propagation wired across the PLL → sector pipeline
• IMD + FDI read_track (real ImageDisk + Formatted Disk Image extraction)
• Plugin-B parsers added for DO, PO, ADL, V9T9, VDK, STX, PRO, ATX, ADF, SCL
• write_track support for TRD, D64, ATR, SAD, SSD, HFE, D80, D82, D71, D81
• Format registry: 138 IDs / 80 plugin registrations after dead-code cleanup
• 6 hardware controllers: Greaseweazle production; SCP-Direct, XUM1541,
  Applesauce as M3 partial scaffolds (real lifecycle + utility code,
  USB/serial wiring pending — see docs/MASTER_PLAN.md §M3)

Fixed

• 3 silent I/O errors in SAD, SSD, HFE parsers
• NULL-checks and silent-error fixes in 8 registry plugins
• Memory allocation limits, bounds checks, forensic fill (security pass)
• CMake Sanitizer / Coverage builds — dead file references removed
• Test include paths repaired

Changed

• Cleanup: 674 orphaned source files deleted (~386 000 LOC dead code)
• 109 dead format files removed from disk
• 165 non-floppy stubs removed (32 active plugins after pruning)

Known Issues

• 6 test modules excluded (missing dependencies)
• Some C headers use protected as field name (C++ incompatible)
• DeepRead forensic modules (write-splice, aging, cross-track, fingerprint,
  soft-decode) and ML analysis are planned for v4.2.0 — not in this release

Full Changelog

See CHANGELOG.md for the complete entry list.