Producer-usefulness integration: notable-findings merge + Reconstruction Brief + deterministic Live 12 fallback

CLAUDE.md

@@ -445,7 +445,7 @@ A quick map from intent to the right place to start:
 
 ## Backport Candidates
 
-The full original `sonic-architect-app` backlog is **shipped** — genre profiles, Ableton device mappings (`abletonDevices.ts` — ported but **research-only** since the 2026-06-11 demotion decision: a scoring baseline for the recommendation-proof harness, imported by no product code; see the NEEDS-WIRING decision in `apps/backend/NEEDS.md`), mix doctor, eight detectors, Phase 3 audition-sample generation, and `patchSmith.ts` (Phase 3 Vital preset generation, [`apps/ui/src/services/patchSmith.ts`](apps/ui/src/services/patchSmith.ts)). See [`BACKLOG.md`](BACKLOG.md) for the landed inventory. Consult it before re-implementing genre detection, mix analysis, acid/reverb/vocal/supersaw/bass/kick detection — they're already in [`apps/backend/analyze_detection.py`](apps/backend/analyze_detection.py) and emit fields visible in `EXPECTED_TOP_LEVEL_KEYS`.
+The full original `sonic-architect-app` backlog is **shipped** — genre profiles, Ableton device mappings (`abletonDevices.ts` — **product-wired 2026-07-02** as the Phase-2-off deterministic fallback via `src/services/deterministicRecommendations.ts` (cited + hedged, renders only when interpretation is unavailable), reversing the 2026-06-11 research-only demotion; the eval bridge that scores it as the harness baseline is unchanged — see `apps/backend/NEEDS.md`), mix doctor, eight detectors, Phase 3 audition-sample generation, and `patchSmith.ts` (Phase 3 Vital preset generation, [`apps/ui/src/services/patchSmith.ts`](apps/ui/src/services/patchSmith.ts)). See [`BACKLOG.md`](BACKLOG.md) for the landed inventory. Consult it before re-implementing genre detection, mix analysis, acid/reverb/vocal/supersaw/bass/kick detection — they're already in [`apps/backend/analyze_detection.py`](apps/backend/analyze_detection.py) and emit fields visible in `EXPECTED_TOP_LEVEL_KEYS`.
 
 ## Companion Agent Docs
 

apps/backend/JSON_SCHEMA.md

@@ -15,7 +15,7 @@ Conventions:
 
 Top-level keys:
 
-`phase1Version`, `bpm`, `bpmConfidence`, `bpmPercival`, `bpmAgreement`, `bpmDoubletime`, `bpmSource`, `bpmRawOriginal`, `key`, `keyConfidence`, `keyProfile`, `tuningFrequency`, `tuningCents`, `timeSignature`, `timeSignatureSource`, `timeSignatureConfidence`, `durationSeconds`, `sampleRate`, `lufsIntegrated`, `lufsRange`, `lufsMomentaryMax`, `lufsShortTermMax`, `lufsCurve`, `truePeak`, `crestFactor`, `dynamicSpread`, `monoCompatible`, `plr`, `dynamicCharacter`, `textureCharacter`, `stereoDetail`, `spectralBalance`, `spectralBalanceTimeSeries`, `spectralDetail`, `stemAnalysis`, `transientDensityDetail`, `saturationDetail`, `snareDetail`, `hihatDetail`, `rhythmDetail`, `melodyDetail`, `transcriptionDetail`, `pitchDetail`, `grooveDetail`, `beatsLoudness`, `rhythmTimeline`, `sidechainDetail`, `reverbDetail`, `vocalDetail`, `acidDetail`, `supersawDetail`, `bassDetail`, `kickDetail`, `genreDetail`, `effectsDetail`, `synthesisCharacter`, `danceability`, `structure`, `arrangementDetail`, `segmentLoudness`, `segmentSpectral`, `segmentStereo`, `segmentKey`, `chordDetail`, `perceptual`, `essentiaFeatures`.
+`phase1Version`, `fundamentalsQuality`, `bpm`, `bpmConfidence`, `bpmPercival`, `bpmAgreement`, `bpmDoubletime`, `bpmSource`, `bpmRawOriginal`, `key`, `keyConfidence`, `keyProfile`, `tuningFrequency`, `tuningCents`, `timeSignature`, `timeSignatureSource`, `timeSignatureConfidence`, `durationSeconds`, `sampleRate`, `lufsIntegrated`, `lufsRange`, `lufsMomentaryMax`, `lufsShortTermMax`, `lufsCurve`, `truePeak`, `crestFactor`, `dynamicSpread`, `monoCompatible`, `plr`, `dynamicCharacter`, `textureCharacter`, `stereoDetail`, `spectralBalance`, `spectralBalanceTimeSeries`, `spectralDetail`, `stemAnalysis`, `transientDensityDetail`, `saturationDetail`, `snareDetail`, `hihatDetail`, `rhythmDetail`, `melodyDetail`, `transcriptionDetail`, `pitchDetail`, `grooveDetail`, `beatsLoudness`, `rhythmTimeline`, `sidechainDetail`, `reverbDetail`, `vocalDetail`, `acidDetail`, `supersawDetail`, `bassDetail`, `kickDetail`, `genreDetail`, `effectsDetail`, `synthesisCharacter`, `danceability`, `structure`, `arrangementDetail`, `segmentLoudness`, `segmentSpectral`, `segmentStereo`, `segmentKey`, `chordDetail`, `perceptual`, `essentiaFeatures`.
 
 **Shared (fast + full) vs full-only.** Fast-mode output is asserted byte-for-byte against the `EXPECTED_TOP_LEVEL_KEYS` set in [`tests/test_analyze.py`](tests/test_analyze.py). Full mode emits those keys *plus* a handful of detail-only fields that are deliberately absent from the shared snapshot: `keyProfile`, `tuningFrequency`, `tuningCents`, `lufsMomentaryMax`, `lufsShortTermMax`, and `pitchDetail`. When changing the schema, update both this list and `EXPECTED_TOP_LEVEL_KEYS`; full-only fields stay out of that set on purpose. See CLAUDE.md tripwire #4. Schema changes are also cross-checked executably against the frontend fixture and parser by `apps/ui/tests/services/phase1ContractParity.test.ts`, driven by the golden snapshot's `topLevelKeys`/`keyTree` — regenerating the golden (`UPDATE_PHASE1_GOLDEN=1`) is what arms the nested half of that gate.
 
@@ -150,6 +150,7 @@ Current server behavior that affects schema expectations:
 | Field | Type | Description | Units / Scale | LLM interpretation note |
 |---|---|---|---|---|
 | `phase1Version` | `string` | Phase 1 JSON schema version, e.g. `"phase1.v2"`. | identifier | Lets consumers detect the schema generation. v2 changed `truePeak` (now dBTP) and `bpmConfidence` (now 0-1) units vs v1. |
+| `fundamentalsQuality` | `object` | Local-only authority summary for tempo, beat grid, downbeats, meter, key, chords, percussion, and monophonic transcription. | `fundamentals-quality.v1` | Never produced by an LLM. Use its status and plain-English messages to avoid presenting weak measurements as settled facts. |
 | `bpm` | `float \| null` | Primary tempo estimate from `RhythmExtractor2013`. | beats per minute | Main tempo anchor for Ableton project tempo and clip warp assumptions. |
 | `bpmConfidence` | `float \| null` | Tempo confidence from `RhythmExtractor2013`, normalized to 0-1. | 0-1 (Phase 1 v2: raw Essentia confidence ~0-5.32 divided by 5.0 and clamped) | Higher = stronger rhythmic periodicity. Below 0.4 (raw < 2.0) suggests an ambiguous pulse or half/double-time content; hedge tempo claims there. |
 | `key` | `string \| null` | Global key label from `KeyExtractor` (`edma` profile), e.g. `"A Minor"`. | categorical | Starting point for harmonic reconstruction; validate by ear against bass/chord roots. |
@@ -163,6 +164,25 @@ Current server behavior that affects schema expectations:
 | `tuningFrequency` | `float \| null` | Estimated tuning reference frequency from spectral peak analysis. | Hz | Deviation from 440 Hz helps detect detuned material or concert-pitch variants. |
 | `tuningCents` | `float \| null` | Tuning offset from A440 in cents. | cents | Positive = sharp of A440, negative = flat. Useful for pitch-correcting reconstructions. |
 
+### `fundamentalsQuality`
+
+`fundamentalsQuality` is a summary of trust, not a separate analyzer. It reads
+the local Phase 1 output and labels each musical fundamental as
+`authoritative`, `ambiguous`, `failed`, or `not_run`.
+
+| Field | Type | Description |
+|---|---|---|
+| `fundamentalsQuality.schemaVersion` | `string` | Currently `"fundamentals-quality.v1"`. |
+| `fundamentalsQuality.targetProfile` | `string` | Currently `"electronic_ableton_v1"`. |
+| `fundamentalsQuality.localOnly` | `bool` | Always `true`; this block is derived from local DSP/model outputs only. |
+| `fundamentalsQuality.llmExcluded` | `bool` | Always `true`; Phase 2 must never write or override this block. |
+| `fundamentalsQuality.overallStatus` | `string` | Worst-case status across domains. |
+| `fundamentalsQuality.domains.<domain>.status` | `string` | Domain status for `tempo`, `beatGrid`, `downbeats`, `meter`, `key`, `chords`, `percussion`, and `transcription`. |
+| `fundamentalsQuality.domains.<domain>.plainEnglish` | `string` | Short user-facing explanation of how much to trust that domain. |
+| `fundamentalsQuality.domains.<domain>.source` | `string \| null` | Local algorithm/source label, e.g. `rhythm_extractor_confirmed`, `edma`, or `librosa_viterbi`. |
+| `fundamentalsQuality.domains.<domain>.confidence` | `float \| null` | Normalized confidence when a domain has a scalar confidence. |
+| `fundamentalsQuality.domains.<domain>.evidence` | `object` | Small domain-specific evidence object, such as BPM agreement, chord disagreement, hit counts, or full-mix fallback. |
+
 ---
 
 ## BPM Cross-Check

apps/backend/NEEDS.md

@@ -188,6 +188,17 @@ resolved the wire-or-demote choice: **demote**. Standing consequences:
    change (it remains a legitimate baseline-fairness tweak if the comparison
    ever needs it).
 
+**⤴ REVERSED 2026-07-02 (owner-approved): wired as the Phase-2-off fallback.**
+`apps/ui/src/services/deterministicRecommendations.ts` now projects the real
+Phase 1 payload into `AudioFeatures` (the render-gated projection flagged
+above — it lives there, not in `analyzer.ts`), attaches per-card Phase 1
+citations, and hedges/skips cards whose fundamentals are weak;
+`DeterministicAdviceSection` renders it **only** when Phase 2 interpretation
+is unavailable (off, failed, interrupted, or terminally empty), labeled
+"Deterministic Baseline · NO AI". The eval bridge and its uncited-by-design
+scoring are unchanged. Standing consequence #2 stands for the Phase-2-on
+path: when interpretation runs, its output is the product surface.
+
 ---
 
 ## Sub-goal 3 — the Gemini verdict

apps/backend/analyze.py

@@ -197,6 +197,7 @@
     TorchcrepeBackend,
     analyze_transcription,
 )
+from fundamentals_quality import build_fundamentals_quality  # noqa: E402
 
 
 def analyze_structure(
@@ -1531,6 +1532,10 @@ def main():
             "perceptual": result.get("perceptual"),
             "essentiaFeatures": result.get("essentiaFeatures"),
         }
+        output["fundamentalsQuality"] = build_fundamentals_quality(
+            output,
+            analysis_mode="fast",
+        )
         _emit_progress_marker("complete", "Analysis complete.", 1.0)
         print("Done.", file=sys.stderr)
         print(json.dumps(output, indent=2))
@@ -1946,6 +1951,10 @@ def main():
         "perceptual": result.get("perceptual"),
         "essentiaFeatures": result.get("essentiaFeatures"),
     }
+    output["fundamentalsQuality"] = build_fundamentals_quality(
+        output,
+        analysis_mode="standard" if run_standard else "full",
+    )
 
     # Conditional MT3 namespace — *absent* (not null) by default. Only added
     # when ASA_ENABLE_MT3=1 AND the transcribe() call succeeded. See the gate

apps/backend/analyze_core.py

@@ -135,8 +135,14 @@ def analyze_bpm(
         }
 
 
-def analyze_key(mono: np.ndarray) -> dict:
-    """Extract musical key and confidence using KeyExtractor with EDMA profile."""
+def analyze_key(mono: np.ndarray, *, include_tuning: bool = True) -> dict:
+    """Extract musical key and confidence using KeyExtractor with EDMA profile.
+
+    ``include_tuning`` gates the per-frame spectral-peak tuning-frequency pass,
+    which is the dominant cost of key analysis on a full track. Fast mode does
+    not surface ``tuningFrequency``/``tuningCents``, so it passes ``False`` to
+    skip that work while keeping the return shape identical.
+    """
     try:
         extractor = es.KeyExtractor(profileType="edma")
         key, scale, strength = extractor(mono)
@@ -147,6 +153,11 @@ def analyze_key(mono: np.ndarray) -> dict:
             "keyProfile": "edma",
         }
 
+        if not include_tuning:
+            result["tuningFrequency"] = None
+            result["tuningCents"] = None
+            return result
+
         try:
             frame_size = 2048
             hop_size = 1024

apps/backend/analyze_fast.py

@@ -10,9 +10,15 @@
 
 import sys
 import numpy as np
-import essentia.standard as es
 
-from analyze import apply_bpm_correction
+from analyze_core import (
+    analyze_bpm,
+    analyze_key,
+    analyze_loudness,
+    analyze_time_signature,
+    analyze_true_peak,
+    extract_rhythm,
+)
 
 
 def analyze_fast(mono: np.ndarray, sample_rate: int = 44100) -> dict:
@@ -37,89 +43,27 @@ def analyze_fast(mono: np.ndarray, sample_rate: int = 44100) -> dict:
     """
     result = {}
 
-    # Run RhythmExtractor2013 for BPM and beat data
-    try:
-        rhythm_extractor = es.RhythmExtractor2013(method="multifeature")
-        bpm, beats, bpm_confidence, _, _ = rhythm_extractor(mono)
-        result["bpm"] = round(float(bpm), 2) if bpm is not None else None
-        # Phase 1 v2: normalize RhythmExtractor2013 confidence (~0-5.32) to 0-1.
-        result["bpmConfidence"] = (
-            round(min(max(float(bpm_confidence), 0.0) / 5.0, 1.0), 3)
-            if bpm_confidence is not None
-            else None
-        )
-    except Exception as e:
-        print(f"[warn] Fast mode BPM analysis failed: {e}", file=sys.stderr)
-        result["bpm"] = None
-        result["bpmConfidence"] = None
-        beats = None
-
-    # Percival BPM for cross-check
-    try:
-        percival = es.PercivalBpmEstimator()
-        bpm_percival = percival(mono)
-        result["bpmPercival"] = round(float(bpm_percival), 2) if bpm_percival is not None else None
-    except Exception as e:
-        print(f"[warn] Fast mode Percival BPM failed: {e}", file=sys.stderr)
-        result["bpmPercival"] = None
-
-    # BPM agreement
-    if result.get("bpm") is not None and result.get("bpmPercival") is not None:
-        result["bpmAgreement"] = abs(result["bpm"] - result["bpmPercival"]) < 2.0
-    else:
-        result["bpmAgreement"] = None
-
-    # BPM correction (shared helper)
-    correction = apply_bpm_correction(result["bpm"], result["bpmPercival"], result["bpmAgreement"])
-    result.update(correction)
-
-    # Key extraction
-    try:
-        key_extractor = es.KeyExtractor(profileType="edma")
-        key, scale, strength = key_extractor(mono)
-        result["key"] = f"{key} {scale}".title() if key and scale else None
-        result["keyConfidence"] = round(float(strength), 3) if strength is not None else None
-    except Exception as e:
-        print(f"[warn] Fast mode key analysis failed: {e}", file=sys.stderr)
-        result["key"] = None
-        result["keyConfidence"] = None
-
-    # Time signature (default to 4/4 if we have rhythm data)
-    if beats is not None and len(beats) > 0:
-        result["timeSignature"] = "4/4"
-        result["timeSignatureSource"] = "assumed_four_four"
-        result["timeSignatureConfidence"] = 0.0
-    else:
-        result["timeSignature"] = None
-        result["timeSignatureSource"] = None
-        result["timeSignatureConfidence"] = None
-
-    # Duration
+    rhythm_data = extract_rhythm(mono)
+    result.update(analyze_bpm(rhythm_data, mono, sample_rate))
+    result.update(analyze_key(mono, include_tuning=False))
+    result.update(analyze_time_signature(rhythm_data, mono=mono, sample_rate=sample_rate))
     result["durationSeconds"] = round(float(len(mono) / sample_rate), 3)
     result["sampleRate"] = sample_rate
 
     # Basic loudness (LUFS integrated and range)
     try:
         # Need stereo for LUFS — Essentia expects shape (N, 2), not (2, N)
         stereo = np.stack([mono, mono], axis=-1)
-        loudness_algo = es.LoudnessEBUR128(sampleRate=sample_rate)
-        _, _, lufs_integrated, lufs_range = loudness_algo(stereo)
-        result["lufsIntegrated"] = round(float(lufs_integrated), 2) if lufs_integrated is not None else None
-        result["lufsRange"] = round(float(lufs_range), 2) if lufs_range is not None else None
+        result.update(analyze_loudness(stereo, sample_rate=sample_rate))
     except Exception as e:
         print(f"[warn] Fast mode loudness analysis failed: {e}", file=sys.stderr)
         result["lufsIntegrated"] = None
         result["lufsRange"] = None
 
-    # True peak (from stereo), emitted in dBTP (Phase 1 v2). NOTE: fast mode uses
-    # a plain sample peak (np.max(np.abs)), NOT the oversampled inter-sample true
-    # peak the standard path computes via Essentia's TruePeakDetector — so this is
-    # a sample-peak approximation and may read lower than the standard path on
-    # material with real inter-sample overs.
+    # True peak (from stereo), emitted in dBTP (Phase 1 v2).
     try:
         if stereo is not None:
-            max_peak = float(np.max(np.abs(stereo)))
-            result["truePeak"] = round(20.0 * np.log10(max_peak), 1) if max_peak > 0 else None
+            result.update(analyze_true_peak(stereo))
         else:
             result["truePeak"] = None
     except Exception as e: