Eve library

Phase 148 implementation evidence

Tracker

Revision 1

On this page

Historical task source. The Eve task board holds current task state.

This ledger records direct implementation evidence for checked Phase 148 requirements. Unchecked requirements remain open. A related data field, protocol seam, or Phase 133 implementation is not treated as completion unless executable behavior and focused tests cover the Phase 148 requirement.

148.1 Deck engine and transport#

Implementation roots: libs/neith/dj/crates/neith-dj/src/deck_engine.rs, decks.rs, time_stretch.rs, key_detect.rs, beatgrid.rs, and beat_phase.rs.

Requirements Direct evidence
148.7.1 A release-mode acceptance test constructs and indexes exactly 1,000,000 distinct library tracks, times an exact lookup only after indexing, requires the correct singleton result in under 100 ms, and passes when run alone under resource monitoring.
148.12.1 decode_video_frames invokes FFmpeg directly, bounds output by requested RGBA dimensions and frame count, rejects malformed output, and converts decoded pixels into the native compositor frame. End-to-end fixtures encode then decode ProRes, H.264, HEVC, and AV1.
148.10.1 encode_master_stereo sends validated finite interleaved stereo to FFmpeg without a shell, selects explicit WAV/FLAC/AIFF/Ogg Vorbis codecs and containers, fails on encoder errors, and verifies each emitted file's container magic in an end-to-end test.
148.2.1 Four independently stateful channel strips apply pre-gain, a real three-way 250 Hz/2.5 kHz crossover EQ with individual band gain/kill, resonant filter, fader curve, and crossfader assignment.
148.1.1.1–.5 DeckLayout admits exactly 2/4/6 decks; DeckRig::six_deck materializes A–F. Every TransportClock independently advances a fractional native-source frame while DeckEngine::render_to_mix linearly resamples mono, stereo, or M-S native audio to a 48 kHz interleaved f32 stereo bus.
148.1.1.6–.10 Bit-perfect bypass fails closed unless the complete path is 48 kHz stereo at unity gain; equal-power block crossfades and media-state replacement implement hot swap; generation leases invalidate stale controllers; ReplayGain-style target-minus-measured LUFS applies bounded per-deck pre-gain.
148.1.2.2–.4, .8, .10 The phase-vocoder supports independent pitch shift and key-locked tempo. TempoControl bounds every declared pitch range, ±12-semitone shift and momentary jog bend, and continuously interpolates validated variable-tempo maps.
148.1.2.6–.7 CamelotPosition renders Camelot and Open Key labels, enumerates ±1/±7 suggestions, and now classifies both signed dominant jumps rather than relying on an impossible seven-step minimum distance.
148.1.3.1–.2 detect_grid computes an onset-energy/transient envelope and autocorrelation grid. FlexibleBeatGrid supplies an explicit bar-one anchor and moves all dependent anchors coherently.
148.1.3.4–.9 Ordered tempo segments preserve musical-beat continuity; anchors nudge by exactly one sample; grid phase normalizes to a signed half-beat; arbitrary signed bar jumps cover the 1/2/4/8/16/32 set; cue quantization and monotonic source-to-target warp markers round-trip samples and beats.
148.1.2.5 estimate_tuning sweeps a 10-cent grid across one semitone, correlating resonant-filter energy against tuning-scaled equal-tempered note frequencies with circular parabolic refinement; tests recover +30 cents and a 432-Hz master (≈ −31.8 cents). detect_key_tuned folds the offset into the chromagram, and detect_edo_tonic finds the root inside an arbitrary EDO (24-EDO test). TrackAnalyser records the tuning at import.
148.1.2.9 process_formant_preserving estimates the per-frame spectral envelope by cepstral liftering, whitens to a source residual, transposes the residual, and re-imposes the original envelope so formants stay fixed; a band-energy test shows the formant held near 1.4 kHz while pitch rises an octave, versus a decisive shift when formant_shift = 2. TempoControl::pitch_processing now drives the formerly-inert preserve_formants flag into a real Smart Pitch config.
148.1.3.3 downbeat_phase measures per-beat low-band (kick) energy and spectral-flux novelty, normalizes and combines them, and picks the bar phase with the greatest summed evidence; align_downbeats re-numbers the grid to it and runs at import. Tests prove the downbeat follows a kick placed on phase 2/3 rather than defaulting to index 0.
148.1.3.10 FlexibleBeatGrid::export_serato_beatgrid emits the documented Serato Serato BeatGrid tag (version 01 00, marker count, non-terminal position + beats-till-next markers, terminal position + bpm, footer); parse_serato_beatgrid round-trips constant- and variable-tempo grids and rejects version/length/count corruption. Rekordbox XML and Engine JSON exports remain.

Explicitly still open in 148.1#

  • 148.1.2.1 needs an objective elastique-class quality comparison; the current phase vocoder is functional but no such reference evidence exists.

148.2 Mixer and routing#

Implementation roots: mixer.rs, effects.rs, stem_mix.rs, and stem_fx.rs.

Requirements Direct evidence
148.2.2–.5 Three-band isolator kills resolve to exact zero; Classic, Innofader, and Pro-X profiles produce distinct crossfader gains; every channel independently assigns A/B/thru; custom fader curves require monotonic endpoints and interpolate between editable control points.
148.2.6–.8 PFL can blend with master or split mono PFL/master across the headphone channels. One processing call produces independent master, booth, record, and zone frames, each through its own threshold/ceiling limiter.
148.2.10–.13 Microphone threshold auto-ducks program before adding speech; each channel has a send and the mixer accepts an FX return; four independent stem gains isolate channel content; auxiliary/USB returns join the program bus explicitly.
148.2.9, .14–.15 Each channel's peak/VU feeds a bounded rolling three-second energy window for short-term loudness. Stateful Chamberlin low/high-pass filtering exposes resonance, while color inserts execute feedback dub delay, deterministic generated noise, and nonlinear drive.

148.3 Sovereign DVS#

Implementation roots: timecode.rs and docs/sovereign-timecode-v1.md.

Requirements Direct evidence
148.3.4 The public v1 specification fixes the preamble, 32-bit absolute tick, CRC-8/ATM, MSB ordering, biphase-mark transitions, wrap semantics, position equation, modes, and fail-closed validation. The encoder and decoder round-trip a bit-exact fixture and reject corruption.
148.3.5–.9 DvsTracker holds fractional sample position, smooths measured velocity, phase-corrects absolute drift, implements immediate absolute needle drop plus relative/internal modes, reports signed forward/reverse/stationary scratch direction, and counts decoded/dropout frames. analyse_signal reports RMS, SNR and dropout ratio.
148.3.10 TimecodeSilenceGate distinguishes silence before carrier lock from sustained silence after it. Gated tracking ignores lead-in movement, permits a configurable grace interval, and then damps/stops transport during outro rather than accumulating false dropout motion.
148.3.13–.15 Sticker calibration solves the absolute offset from desired media sample and observed tick; monitor pass-through preserves both input channels sample-for-sample; the setup report measures quality, recommends bounded input gain, diagnoses channel polarity, and gates readiness on configurable SNR/dropout limits.

Explicitly still open in 148.3#

  • 148.3.1–.3 remain open because no licensed signal fixtures or validated decoders exist for Serato NOISEMAP, Traktor Scratch MK2, or RB-VS1-K.
  • 148.3.11–.12 remain open: existing CDJ support is tested Pro DJ Link network protocol, not a demonstrated USB HID implementation with jog feedback.

148.4 Cues, loops, and sampler#

Implementation roots: hot_cues.rs, cue_loops.rs, loops.rs, decks.rs, sampler.rs, sampler_performance.rs, and controller codecs under controllers/.

Requirements Direct evidence
148.4.1.1–.4 Hot-cue banks support 8/16/32 slots with persisted label/color and explicit cue, loop, loop-roll, slice, and beat-jump kinds. Quantization resolves the stored sample through a flexible beat grid only when enabled.
148.4.1.5–.10 Sorted CueSequence steps implement Flip/chains; preview returns audio without deck mutation; PerformanceMemory serializes per-track banks and sequences; cues move by exact samples; DDJ/Traktor controller codecs and the shared controller mapper already exercise colored pad recall.
148.4.2.1–.10 The loop state machine implements manual and auto loops, half/double, and sample-exact shifting. PERFORMANCE_LADDER spans 1/32 beat through 32 four-beat bars. LoopRoll separates audible wrap from underlying transport; saved-loop vectors are unbounded; jump-inside-loop wraps both directions; smart loops snap to the grid; phrase capture preserves the exited range; loop stutter alternates start/off-beat.
148.4.3.1–.3 Sampler banks materialize 8 or 16 slots and render one-shot, loop, hold, and stutter modes. SampleTone::playback_ratio locks a declared source BPM to master tempo while composing independent pitch.
148.4.3.5–.10 Ordered non-overlapping velocity layers select by MIDI velocity. SampleTone executes gain, pitch/tempo ratio, one-pole filtering and equal-power pan. Deck ranges and recorded master blocks become real PadSamples; named banks serialize into a library; tick-ordered press/release/stop capture replays MIDI performance.

Explicitly still open in 148.4#

  • 148.4.3.4 remains open because there are no validated Ableton, MPC, and Maschine sampler-bank import fixtures/parsers.

Verification#

Focused tests were added for every row above. The final package-wide gate passed 768 unit tests with no failures (the count rose from 738 as tasks 148.1.2.5, 148.1.2.9, 148.1.3.3, and 148.1.3.10 landed with focused coverage); the million-track acceptance benchmark and one doc example are intentionally ignored in ordinary runs. The benchmark passes separately in release mode. cargo clippy -p neith-dj --all-targets -- -D warnings, cargo fmt --all -- --check, the optimized release build, and the task-scoped git diff --check pass. The focused Themis Shield API server contract suite passes 9 tests.

148.5 Real-time stems#

Implementation roots: stem_separator.rs, stem_separator_realtime.rs, stem_mix.rs, stem_fx.rs, decks.rs, and stem_workflow.rs.

Requirements Direct evidence
148.5.2–.6 Realtime configuration reports a 5.33 ms default analysis frame and tests both default and sub-3 ms modes. Every deck owns a four-stem mix with gain/mute/solo and one independent EQ/filter/echo/reverb-capable effects rack per stem.
148.5.7–.11 Equal-length validation precedes single-stem cross-deck swaps. The higher-window offline separator bakes into a cache pinned by track, model version, and canonical source fingerprint; import precache executes the same path. Scratch policy switches between an isolated stem and constructive mix; key detection sums only the selected mask.
148.5.13–.15 Drum repeat validates and tiles an exact isolated range. Isolation rating reports target/residual dB and a bounded five-point separation score. Stable FNV-1a over canonical sample bits provides deterministic, content-sensitive logging identity.

Explicitly still open in 148.5#

  • 148.5.1 remains open: the built-in sovereign separators are constructive STFT/HPSS and pluggable-mask implementations, not a demonstrated on-device Demucs-class neural model.
  • 148.5.12 remains open because six generic output stems do not prove the named guitar/keys/strings 5-stem and 8-stem sub-separation.

148.6 Performance FX#

Implementation roots: performance_fx.rs, effects.rs, and time_stretch.rs.

Requirements Direct evidence
148.6.1–.2 DriveFilter is a resonant state-variable low/high-pass filter with nonlinear drive; BeatEcho supports free or beat-fraction delay and a saturating tape-feedback path.
148.6.3 Plate and hall tune room/damping profiles independently. Shimmer runs the reverb return through a dual-head, crossfaded one-octave-up delay shifter before feedback; freeze recirculates that return without new input.
148.6.4–.6 The existing flanger is complemented by a stateful phaser, slice-reordering Beat Mash, and step-pattern gate.
148.6.7–.14 Noise sweep, ring modulation, granular stutter, brake/backspin envelopes, reverse with cue-back, pitch shifting, transient shaping, and sidechain ducking all process sample buffers rather than merely storing UI state.
148.6.15–.16 FxChain3 enforces a three-effect maximum and supports serial and parallel routing; FxMacro maps one normalized control to multiple bounded parameters.
148.6.18–.20 Timestamped automation is ordered and interpolated, effect triggers can be beat-quantized, and BeatFraction spans 1/64 beat through 64 beats (16 four-beat bars).

Explicitly still open in 148.6#

  • 148.6.17 remains open because there is no VST3/CLAP ABI host, discovery, sandbox, or plug-in lifecycle support.

148.7 Library and crates#

Implementation roots: library.rs, library_workflow.rs, import.rs, and cue_loops.rs.

Requirements Direct evidence
148.7.2–.4 The base library owns hierarchical crates, ordered playlists, and persisted AND/OR smart lists. The workflow adds exact BPM/key/energy/date/genre filtering and tokenizes title, artist, album, genre, custom tags, and free text into an intersection search index.
148.7.5–.7 Each play records session, track, start/end, and deck while updating count/last-played. Preview slices decoded audio without deck transport. Peak-binned signed-16 thumbnails have deterministic bin boundaries and bounds.
148.7.8–.9 Duplicate grouping prefers exact audio fingerprint and falls back to normalized artist/title/duration metadata. Relink accepts only a candidate with the missing track's fingerprint and performs the URI replacement through the audit path.
148.7.10 Fixture-tested parsers cover Music.app/iTunes XML property lists, Rekordbox XML, Traktor NML, Serato's tagged UTF-16BE crate format, and a validated portable Engine JSON export, including playlist references.
148.7.12–.13 RoamingLibrary serializes a schema-versioned snapshot and emits content-hash-bound AWS SigV4 PUT/GET requests compatible with path-style sovereign S3/MinIO. Per-entity Lamport stamps plus device-ID tie-breaking make concurrent machine merges commutative and idempotent; track tombstones propagate offline deletions.
148.7.14–.15 Smart lists serialize rules, combination, sort, direction, and limit. Every supported track-field mutation records monotonic sequence, timestamp, actor, track, field, before, and after values; relink is unable to bypass it.

Explicitly still open in 148.7#

  • 148.7.11 remains open until a filesystem-level, validated CDJ USB export is materialized.

148.8 Track analysis#

Implementation roots: analysis.rs, beatgrid.rs, key_detect.rs, and library_workflow.rs.

Requirements Direct evidence
148.8.1–.3 TrackAnalyser runs grid/downbeat, key, normalized energy, integrated LUFS, four-times interpolated true peak, and target-loudness ReplayGain analysis in one versioned result.
148.8.5–.8 One-second energy windows classify intro/outro, breakdown, and drop; isolated-vocal energy classifies vocal/instrumental windows; beat/bar data emits configurable phrase starts; first/last above-threshold samples produce a bounded trim suggestion.
148.8.9–.10 A portable peak pyramid halves deterministic signed-16 bin counts down to one. Statistics include RMS, sample peak, true peak, crest factor, integrated LUFS, and gain correction.
148.8.11 composite_fingerprint feeds validated mono PCM to FFmpeg's libchromaprint muxer without a shell and pairs its non-empty acoustic fingerprint with the deterministic gain-normalized sovereign fingerprint; an end-to-end test proves repeatability.
148.8.13, .15 Each result pins an algorithm version; cache lookup explicitly demands reanalysis on mismatch. The complete cache serializes to JSON and round-trips without process-local state.

Explicitly still open in 148.8#

  • 148.8.4 remains open without a Nous-trained mood/genre classifier.
  • 148.8.12 remains open without an exercised GPU batch-analysis path.
  • 148.8.14 remains open because no consented training-feedback pipeline exists.

148.9 Hardware controllers#

Implementation roots: controllers/ and controller_workflow.rs.

Requirements Direct evidence
148.9.8–.9 One-shot MIDI learn captures channel/control address, logical action, optional deck, editable response curve, and output range. Schema-versioned JSON definitions validate IDs, ranges, channel/control bounds, action names, and unique addresses and round-trip without vendor code.
148.9.11–.15 Jog torque interpolates explicit light/heavy endpoints through an editable exponent. Existing output events and vendor encoders preserve each hot cue's RGB. Registry connect/disconnect increments generation without assuming stable device pointers; combined-deck ownership rejects collisions; headless mode accepts audio actions while rejecting display-only commands.

Explicitly still open in 148.9#

  • 148.9.1 stays open because DDJ-FLX10/DDJ-family USB-MIDI codecs do not prove the checklist's explicitly requested HID mapping.
  • 148.9.2 stays open because CDJ-2000/3000 Pro DJ Link is not USB HID.
  • 148.9.3 stays open because existing DJM-S MIDI/SysEx does not prove the combined DJM HID plus MIDI family requirement.
  • 148.9.4 stays open because S2/S3/S4/S5/S8 HID coverage lacks the named Z2.
  • 148.9.5–.7 stay open without validated Denon, Rane, and Reloop wire fixtures.
  • 148.9.10 stays open: text, tempo, LEDs, and jog feedback do not prove waveform transport to every controller screen.

148.10 Session recording#

Implementation root: recording.rs.

Requirements Direct evidence
148.10.2–.4 Recording ingests a master and named equal-length channel tracks, records exact track-change split samples, closes the previous play, and emits ID3v2.4 title/custom text frames containing millisecond offsets plus artist/title.
148.10.7, .9–.11 Every play carries its rights fingerprint. Bookmarks and normalized performance controls store absolute samples. Replay at any sample reconstructs active track, latest control values, and elapsed bookmarks for time-travel inspection.
148.10.12–.13 Append-only NDJSON journal records monotonic sequence, typed events, and payload checksum; each flush fsyncs only new events and recovery rejects gaps, malformed data, or checksum mismatch. Before record, a bounded five-minute VecDeque continuously evicts oldest samples; start drains it into the master.

Explicitly still open in 148.10#

  • 148.10.5–.6 remain open without validated Mixcloud/SoundCloud/sovereign, Icecast, and RTMP network destinations.
  • 148.10.8 remains open without a composed waveform/camera video capture path.
  • 148.10.14 remains open without verified Atmos/binaural session print output.
  • 148.10.15 remains open without an exercised sovereign radio/podcast publish service.

148.11 Show synchronization#

Implementation root: sync.rs.

Requirements Direct evidence
148.11.2 MidiClockPll accepts timestamped 24-PPQN 0xF8 ticks, smooths observed interval, reports phase error and corrected BPM, and generates timestamped output ticks from the corrected interval.
148.11.3 MtcQuarterFrameDecoder reassembles and range-validates all eight F1 pieces including the SMPTE rate; decode_ltc_bits validates the 80-bit sync word and decodes the LTC BCD transport fields into the same typed position.
148.11.4–.8 Sans-I/O encoders emit Art-Net trigger and sACN application payloads. Validated padded OSC messages carry play, BPM, bar, beat, and tick to generic transport, Composer, VJ, and video-deck paths; four independent video decks seek musical beats to media frames.
148.11.10 FreewheelClock retains last BPM and integrates elapsed wall time after its sync timeout while explicitly changing synchronized state to false.

Explicitly still open in 148.11#

  • 148.11.1 remains open without actual Ableton Link peer discovery/protocol interoperability.
  • 148.11.9 remains open until network-load testing, not a pure jitter utility, proves the 0.5 ms bound.

148.12 Video and visuals#

Implementation root: visuals.rs.

Requirements Direct evidence
148.12.2–.3 A beat-ordered visual timeline resolves the latest trigger at any musical beat. CPU reference compositing executes pixel-level crossfade, directional wipe, cube-like reveal, and configurable tile reveal over dimension-validated frames.
148.12.4–.5 Chroma key uses RGB distance with tolerance/softness; luma key uses Rec.709 weights, softness, and inversion. Posterize quantizes channels, glitch rotates alternating rows, and pixel sort orders each row by luma.
148.12.7–.8 DualScreenOutput holds independent booth and front-of-house frames. Four-corner projection maps normalized source coordinates through a bilinear keystone quad and tests exact corner preservation.

Explicitly still open in 148.12#

  • 148.12.6 remains open without platform-native NDI/Syphon/Spout output.
  • 148.12.9 remains open without camera/capture-card device ingestion.
  • 148.12.10 remains open without an exercised Phase 47 generative-visual model driven by live stem features.

148.13 Tour resilience#

Implementation roots: resilience.rs, deck_engine.rs, cloud/cache.rs, and controller_workflow.rs.

Requirements Direct evidence
148.13.3–.4 Each deck owns an explicit measured/budget CPU percentage. Over budget or 85% system load disables advanced FX first, then stems; sustained sub-70% load restores both.
148.13.5–.8 Heartbeat timeout emits bounded restart actions then promotes a generation-checked mirrored warm spare. Audio router accepts only enumerated devices and increments generation for external hot swap or internal-output failover; deck media crossfade remains gapless across the route change boundary.
148.13.10–.11 Controller registry disconnect removes the instance and increments generation without dereferencing stale state. Thermal classification emits normal/notice/throttled states and deliberately has no shutdown action.
148.13.13–.15 Tour lock permits only transport, fader, and emergency actions while locked and requires a 1.5-second continuous hold to unlock. Local media paths and CloudDirectPlay cache need no network hit on cached playback. Bounded local black-box telemetry evicts oldest events deterministically for post-set review.

The tour-lock design follows the mandatory frontend skill: operational status and emergency transport remain dominant, unlock is deliberate, and no decorative or distracting state was introduced. No DJ frontend exists in the repository, so there is no applicable rendered Playwright surface yet.

Explicitly still open in 148.13#

  • 148.13.1–.2 remain open until ASIO/CoreAudio hardware tests prove <=4 ms round-trip and no dropout at 95% CPU.
  • 148.13.9 remains open because software brown-out state cannot provide the required battery-backed power handling.
  • 148.13.12 remains open until a supervised six-hour uninterrupted performance run is captured as evidence.

148.14 Rights and reporting#

Implementation roots: rights.rs and recording.rs.

Requirements Direct evidence
148.14.1–.4 Rights metadata validates normalized 12-character ISRC, optional T-prefixed ISWC, finite positive BPM, key, composers, publishers, and license. Timestamped set lists and frequency counts derive directly from recorded plays. CSV exports cover BMI, ASCAP (the checklist's “ASCRAP” typo), SACEM, GEMA, and JASRAC.
148.14.5 The Rust DJ boundary now emits the Phase 74 Shield API's authenticated POST /scan JSON contract with base64 PCM and duration/sample geometry, parses its wrapped or direct risk verdict, and fails closed on HTTP, schema, or unknown-risk responses.
148.14.6–.8 Every played range becomes an explicit Allowed/Mute/Block stream segment and missing policy fails closed to Block. The same escaped export supports venue, festival, online-platform, and event-organizer profiles with start/end, identity, musical metadata, and license.

Explicitly still open in 148.14#

  • 148.14.9 remains open without signatures issued and verified by the Phase 146 sovereign identity service.
  • 148.14.10 remains open without an actual optional Aje ledger/NFT write and read-back proof.

148.15 Testing and validation#

Requirements Direct evidence
148.15.1 The sovereign decoder regression fixes preamble, tick, CRC, bit ordering, biphase cells, and exact encoded length, round-trips a fixed 0x12345678 fixture, and proves a single-bit payload corruption fails CRC.
148.15.8 run_failover_drill drives a watchdog through every bounded restart, promotes only the expected mirrored generation, verifies restored deck/mixer state, and routes audio to the enumerated internal fallback.

Explicitly still open in 148.15#

  • 148.15.2–.7 and .9–.10 require unavailable proprietary references, large licensed datasets, hardware/network peers, 24-hour or multi-machine soaks, club access, or accumulated real-show evidence. Unit simulations are not presented as substitutes for those acceptance gates.

Locally actionable completion audit (2026-07-15)#

Every unchecked requirement was re-audited against the current repository. None has an implementation-and-verification path that can be completed solely in this checkout without substituting a mock, invented fixture, protocol seam, or short simulation for the requirement's named acceptance evidence:

  • Proprietary media, controller, and device interoperability remains gated by licensed timecode/sample/library fixtures or the named physical hardware: 148.3.1-.3, 148.3.11-.12, 148.4.3.4, 148.7.11, 148.9.1-.7, 148.9.10, 148.12.6, 148.12.9, 148.13.1, 148.13.9, 148.15.2, 148.15.6, and 148.15.9.
  • Model/runtime quality requirements remain gated by the named trained model, GPU/model runtime, consented feedback service, reference corpus, or licensed evaluation dataset: 148.1.2.1, 148.5.1, 148.5.12, 148.8.4, 148.8.12, 148.8.14, 148.12.10, 148.15.3, and 148.15.4.
  • Plug-in, publishing, synchronization, signing, and ledger interoperability remains gated by a real external peer/service and read-back or round-trip evidence: 148.6.17, 148.10.5-.6, 148.10.15, 148.11.1, 148.11.9, 148.14.9-.10, and 148.15.7.
  • Capture/spatial output and endurance/reliability gates require the relevant camera, Atmos renderer/listener validation, sustained load or elapsed real-world operation: 148.10.8, 148.10.14, 148.13.2, 148.13.12, 148.15.5, and 148.15.10.

The audit deliberately leaves those boxes unchecked. The executable local scope is complete; marking any item above complete from an adapter type, mocked response, synthetic signal, or abbreviated soak would weaken the checklist's stated evidence standard.