# 2018 Contrapunctus candidate-source audit

## Claim tested

Whether the preserved Tengri video audio can be identified as the public Kimiko Ishizaka Contrapunctus I recording, rather than merely recognised as the same Bach work.

## Result

**Not identified.** The candidate is a real, publicly documented recording, but the tests do not distinguish it as the source of the Tengri track. Broad chroma is score recognition, not recording fingerprinting; the synthesized same-work control attains an equal-or-higher broad-score result. The direct linear-retiming clone check has low sample correlations for every candidate.

## Inputs and public sources

- **Kimiko Ishizaka — Contrapunctus I** — [public source](https://commons.wikimedia.org/wiki/File:Kimiko_Ishizaka_-_J.S._Bach,_The_Art_of_Fugue,_BWV_1080_-_01_Contrapunctus_1.ogg); OGG SHA-1 `a5513d2f6350ce9a354b1fb0eb13a80cc11aab4f`; OGG SHA-256 `4e637d8ff71cdc53b078b2feb95378da0646d9860880a73a4adc795c4d3c35fe`.
- **Mutopia-derived synthesized Contrapunctus I** — [public source](https://commons.wikimedia.org/wiki/File:The_Art_of_Fugue,_Contrapunctus_I.ogg); OGG SHA-1 `8e26008e3a94f5e2a67ab5c6c257d7fa9a31f951`; OGG SHA-256 `fa6ca72fe722ca05da861d9d57196e2d7edddc435d3cdee85595cfe028b7f610`.
- **Kimiko Ishizaka — Contrapunctus IV (wrong-work control)** — [public source](https://commons.wikimedia.org/wiki/File:Kimiko_Ishizaka_-_J.S._Bach,_The_Art_of_Fugue,_BWV_1080_-_04_Contrapunctus_4.ogg); OGG SHA-1 `3e7269e4316a28440fc304210ea15d0bcef8415c`; OGG SHA-256 `8a5cfe7c276fda6f5abb731b686ee9baaa22843574d471d1553321e2915da1e4`.

The Ishizaka-I OGG SHA-1 is `a5513d2f6350ce9a354b1fb0eb13a80cc11aab4f`, matching the checksum published by Wikimedia Commons for that file.

## Broad pitch-class comparison

The search permits a global tempo map, offset, and all twelve semitone shifts. Its result says only how closely a candidate follows the harmonic progression of the video. Search ranges differ by candidate duration, so the maxima below are not p-values and must not be read as a race between recordings.

| Candidate | Best cosine | Semitone shift | Reference seconds / video second | Offset (s) |
|---|---:|---:|---:|---:|
| Kimiko Ishizaka — Contrapunctus I | 0.774737 | 0 | 1.2700 | 2.200 |
| Mutopia-derived synthesized Contrapunctus I | 0.834523 | 0 | 1.0000 | 1.000 |
| Kimiko Ishizaka — Contrapunctus IV (wrong-work control) | 0.742339 | 0 | 1.0500 | 9.500 |

## Direct linear-retiming clone check

Seven fixed five-second video windows (5, 25, …, 125 seconds) are compared after the best broad map, with only ±0.2 seconds of residual source lag. This is sensitive to a straightforward sample-rate/speed-derived copy. It does **not** test a pitch-preserving time stretch, a remix, a different rendering of the same MIDI, or arbitrary editing.

| Candidate | Median maximum Pearson r | Range across seven windows |
|---|---:|---:|
| Kimiko Ishizaka — Contrapunctus I | 0.010886 | 0.007012–0.011538 |
| Mutopia-derived synthesized Contrapunctus I | 0.047491 | 0.023287–0.070090 |
| Kimiko Ishizaka — Contrapunctus IV (wrong-work control) | 0.027516 | 0.013521–0.032202 |

## Interpretation boundary

The video is still consistent with *Contrapunctus I*, as the uploader-labelled reply says. This audit gives no defensible basis to call the Ishizaka recording the video source, to identify the actual performer, or to turn a performance difference into a cipher. A source claim would need a stable recording-specific fingerprint or an independently sourced upload history; a hidden-message claim still needs an author-selected transformation and controls.

## Reproduction

Run `scripts/audit_contrapunctus_performance_source.py` with the preserved video WAV, three decoded control WAVs, their source OGGs, and an output directory. The input file hashes, declared search ranges, full per-semitone tables, and fixed window records are in the adjacent JSON.