The "Magic Box" Problem
"Our AI improved your audio" — every service says this. It differentiates nothing.
The deeper problem: a service that cannot explain why it sounds better cannot earn the trust of enterprise engineers. Only individual users pay monthly fees for a magic box. When TuneCore's engineering team makes an integration decision, they look at architecture — not sound.
Transparency is not philosophy. It is engineering strategy. Showing the process is itself the differentiation.
Three Layers, Fully Exposed
The aimastering.dev playground shows three layers in sequence.
| Layer | Component | Shown to User |
|---|---|---|
| 1 | Gemini Scan | Full scan output — LUFS, RMS, frequency balance, section structure, genre estimate in raw JSON |
| 2 | TRIVIUM | GRAMMATICA / LOGICA / RHETORICA deliberation log and weighted consensus result (consensus_minutes) |
| 3 | 14-Stage DSP | Every parameter per stage — LR crossover coefficients, EQ gain, limiter threshold, dither settings, all disclosed |
Layer 1: Gemini Scan
Gemini receives MP3/WAV and returns JSON conforming to the dynamic_mastering_formplan_v2 schema. The playground renders this JSON as-is — no hiding, no reformatting.
{
"track_identity": {
"genre": "Club / Techno",
"bpm": 134,
"key": "Am"
},
"whole_track_metrics": {
"integrated_lufs": -14.2,
"dynamic_range_lu": 8.1,
"peak_dbtp": -0.4,
"spectral_balance": { "sub": -2.1, "low": 0.3, "mid": -0.8, "high": 1.1 }
},
"macro_form": {
"sections": [
{ "label": "Intro", "start": 0, "end": 32, "energy": 0.42 },
{ "label": "Drop1", "start": 32, "end": 96, "energy": 0.91 },
{ "label": "Break", "start": 96, "end": 128, "energy": 0.55 }
]
}
}Layer 2: TRIVIUM Deliberation Log
Three agents submit independent positions; consensus_arbiter.py integrates them via weighted median. The full deliberation record (consensus_minutes) is displayed in the playground.
{
"target_lufs": {
"GRAMMATICA": -9.0,
"LOGICA": -8.5,
"RHETORICA": -8.0,
"consensus": -8.6,
"method": "weighted_median",
"weights": [0.55, 0.25, 0.20]
},
"unresolved_tensions": [
{
"field": "high_shelf_gain",
"GRAMMATICA": "+0.5 dB (ceiling constraint)",
"RHETORICA": "+1.8 dB (presence target)",
"note": "GRAMMATICA veto applied. Final: +0.5 dB"
}
]
}Layer 3: 14-Stage DSP Physics
Once the Blueprint JSON is finalized, control_layer.py converts it to per-section DSP parameters, and the 14-stage engine processes in sequence. The playground shows input/output at every stage.
| # | Stage | Exposed Parameters |
|---|---|---|
| 01 | DC Offset Removal | offset_db |
| 02 | LR Crossover Split | crossover_hz × 3 (complementary Linkwitz-Riley) |
| 03–06 | Per-Band EQ | gain_db × 4 bands × section |
| 07 | Multiband Compressor | threshold / ratio / attack / release × 4 bands |
| 08 | MS Processing | mid_gain_db / side_gain_db |
| 09 | Stereo Width | width_factor (0.0–2.0) |
| 10 | Harmonic Saturation | drive_db / character (tube|tape|solid) |
| 11 | LR Crossover Merge | merge_gain_db |
| 12 | True Peak Limiter | threshold_dbtp / ceiling_dbtp (stereo-linked) |
| 13 | Oversampled TP Safety | oversample_rate / final_ceiling_dbtp |
| 14 | TPDF Dither | bit_depth / noise_shaping: none |
Transparency as Sales Material
The first question in a white-label negotiation is not "what AI are you using?" It is: "can you explain why it sounds this way?"
| Criterion | Magic Box | Transparent Studio |
|---|---|---|
| Enterprise trust | "It sounded better" — subjective only | Explainable via architecture |
| Audit readiness | Impossible | Full trace via Blueprint JSON + logs |
| Adoption barrier | "No idea what's inside" | "Everything is visible" |
| Differentiation | None (everyone claims the same) | The process itself is a proprietary asset |
| IP valuation | Low (black box) | High (explainable pipeline) |
The Transparent Studio Declaration
This playground does not sell an "experience." It sells the complete public disclosure of an engineering process.
What Gemini read, what TRIVIUM argued and what it rejected, which parameters the 14-stage DSP operated with — all of it is visible. Not because there is nothing to hide, but because hiding it is the greatest weakness.
The message of the transparent studio:
"Every decision is visible. Every parameter is yours to override."