# The rights chains, read at the source Two corpora were acquired for this article — Sousa and Bach — and they are cleared by two completely different mechanisms. Part one's Chopin chain is in that article's own kit and is not repeated here. This file is written for this kit rather than copied from the internal corpus receipts it summarises; **every quotation inside it is verbatim**, and each names the receipt and the timestamp it was read at. --- ## Sousa — a work of the United States federal government **What the recordings are:** the complete marches of John Philip Sousa performed by the United States Marine Band — 106 recordings, acquired 2026-08-30 between 00:43 and 00:49 UTC, 703,056,088 bytes. Two layers, both clear, and a third independent stamp: - **The compositions.** Sousa lived 1854–1932; the latest work in this set dates to 1925. Public domain by age. - **The recordings.** Works of the United States federal government, prepared by members of the U.S. Marine Band in the course of their official duties. Under **17 U.S.C. § 105**, copyright protection is not available for any work of the United States Government — the recordings never carried a copyright to waive. - **The mirror's own declaration.** The archive.org mirror carries a **Creative Commons Public Domain Mark 1.0** (`https://creativecommons.org/publicdomain/mark/1.0/`) on every one of the 106 items — **verified per item, not sampled.** **The corpus receipt, verbatim** (`music/corpus/sousa-shakedown/PROVENANCE.md:6-16`): > These recordings are works of the United States federal government, prepared by > members of the U.S. Marine Band in the course of their official duties. Under > **17 U.S.C. § 105**, copyright protection is not available for any work of the United > States Government. The archive.org mirror carries them under a **Creative Commons > Public Domain Mark 1.0** declaration on every one of the 106 items — verified per > item, not sampled. ### Where the bytes came from, and the 403 that sent us there **Verbatim** (`PROVENANCE.md:18-27`): > The U.S. Marine Band's own page for the series is > `https://www.marineband.marines.mil/Audio-Resources/The-Complete-Marches-of-John-Philip-Sousa/`. > That host **returned HTTP 403 to an honest user-agent** when this corpus was acquired > (2026-08-30), so it was not the fetch path — the estate does not spoof a browser > user-agent to get past a WAF. The bytes below were taken instead from the > **archive.org mirror**: the `InternetJukebox.JPS.*` items in the `PublicJukebox` > collection, each of which credits the Marine Band in its own item notes — *"Audio and > scores courtesy of the one and only, the President's Own, The U.S. Marine Band."* The run doc says the same thing in its own words (`SOUSA-SHAKEDOWN…:71-77`): the host *"returned HTTP 403 to the honest user-agent this lane sends. It was not the fetch path, and no browser user-agent was spoofed to get past the WAF."* The user-agent that was sent is recorded in the corpus receipt at `PROVENANCE.md:32`. It carries the operator's name and address and is **not reproduced here**; it identifies the project, the version and the purpose of the fetch, honestly, which is exactly why the host refused it. ### Integrity at intake **MD5 and byte length were compared against the values archive.org publishes: 106 of 106 matched, zero mismatches.** The corpus's own manifest hashes: manifest.sha256 9b60f30188d622279acefedcc6f21c3519793ee4328c7926c8422fca1ddadd5f manifest.json 9aa11cb9236f4d941b14b15f56be5db52f2dfa6c503aa7b839c16b60ca4c9a11 ### What was deliberately left out Section stems and per-instrument tracks were excluded, and not for tidiness (`SOUSA-SHAKEDOWN…:104-106`): > **None of those were taken, and the reason is not tidiness.** A trumpet-section stem is > not a march: training on one teaches the model something false about what a Sousa march > sounds like. And because they are all recordings *of the same work*, sweeping them in > would have put a single march into the corpus up to **124 times**. --- ## Bach — the performer freed her own recordings **What the recordings are:** Kimiko Ishizaka's *Open Goldberg Variations* and *Open Well-Tempered Clavier, Book 1* — 79 recordings, 24-bit FLAC, 3,162,705,680 bytes. Two independent layers (`music/corpus/bach-shakedown/PROVENANCE.md:12-20`, verbatim): > 1. **The compositions** are by Johann Sebastian Bach (1685–1750) — public domain by age. > 2. **The recordings** — normally the encumbered layer — were dedicated to the public > domain under **CC0 1.0 Universal** by the performer, Kimiko Ishizaka, as the whole > point of the crowd-funded "Open" Bach projects. That dedication is what makes this > corpus usable; a public-domain score with an ordinary commercial recording would not > be. ### The dedication, read per item and stored verbatim `PROVENANCE.md:37-44`: > **Licence evidence, stated exactly.** The Goldberg item carries > `licenseurl = http://creativecommons.org/publicdomain/zero/1.0/` — the CC0 1.0 deed > itself. The WTC item carries > `licenseurl = http://creativecommons.org/publicdomain/mark/1.0/` (Public Domain Mark) > **and** an explicit `rights` field reading **"Released under CC0 1.0 Universal"** — so > the CC0 dedication is asserted in the item's own metadata, not inferred from the > project's reputation. Both were read from `https://archive.org/metadata/` > at acquisition time and are stored verbatim in `_work/items_meta.json`. | | Open Goldberg Variations | Open Well-Tempered Clavier, Book 1 | |---|---|---| | archive.org identifier | `OpenGoldbergVariations` | `bach-well-tempered-clavier-book-1` | | creator | Kimiko Ishizaka | Kimiko Ishizaka | | date | 2012-05-28 | 2015-03-19 | | `licenseurl` (verbatim) | `http://creativecommons.org/publicdomain/zero/1.0/` | `http://creativecommons.org/publicdomain/mark/1.0/` | | `rights` (verbatim) | `None` | `Released under CC0 1.0 Universal` | | files taken | 31 | 48 | **The honest user-agent worked here.** `PROVENANCE.md:8-10` records it as *"honest throughout; no browser user-agent was spoofed and no host returned 403."* The UA string itself carries the operator's name and address and is not reproduced here. ### Integrity at intake 79 files, 3,162,705,680 bytes; **MD5 79/79, byte length 79/79, integrity failures 0.** Corpus `manifest.sha256` = `d86db1ab7f5d5d1d906f4f43011c53ce6ac6b2ae09d5bb9e53be066c7b904201`. ⚠ **A byte-count disagreement inside the Bach material itself**, disclosed because the house shows both readings rather than picking one: the corpus receipt splits the total as Goldberg 1,388,680,423 / WTC 1,774,025,257, while the run doc splits it as 1,391,486,584 / 1,771,219,096. **Both pairs sum to the same 3,162,705,680.** One file is attributed to a different item in the two tables. Use the total, not the split. ### The one defect in the source, caught by two receipts `PROVENANCE.md:59-70`, verbatim: > **The Goldberg FLAC item ships 31 files for 32 works.** Its track 28 is > **Variatio 27 and Variatio 28 concatenated into a single file**. Two independent > receipts agree, so this is measured rather than assumed: > > * **Duration.** The FLAC track 28 runs **257.45 s**. On the 32-track MP3 item > (`The_Open_Goldberg_Variations-11823`) Variatio 27 runs 111.43 s and Variatio 28 > runs 146.10 s — **111.43 + 146.10 = 257.53 s**. From track 29 on, the FLAC item's > numbering runs exactly one behind the MP3 item's, and the two re-align work-for-work. > * **The item's own filename**, which names both works: > `...-28Variatio27A2Clav.CanoneAllaNona-Variatio28A2Clav.flac` That defect is why the split machinery had to grow: the unit of splitting became the connected component of the file↔work graph, which handles both a work recorded twice (Chopin's case) and a file carrying two works (Bach's) at once. ### What was declined, on purpose `PROVENANCE.md`, "What was NOT taken": > * **`jamendo-113109`** — a Goldberg copy under **CC BY 3.0**, not CC0. Declined because > the brief asked for public-domain material and CC BY carries an attribution > condition. > * **The official sites** `opengoldbergvariations.org` / `welltemperedclavier.org` — > both answered HTTP 200 to the honest user-agent, but both route their downloads > through a Bandcamp "pay what you choose" checkout rather than serving files directly. ### The self-refutation the Bach lane wrote against its own choice `BACH-SHAKEDOWN…:125-135`, verbatim: > **An honest note against this choice.** `AUDIO-WORKFLOW-LESSONS-2026-08-30` lesson 3 — > written earlier today, after the corpus-bandwidth audit — **refutes the lossless-first > filter**: Sousa's MP3s reach ~20 kHz while lossless solo piano rolls off at ~9 kHz > naturally, so the rule is now *"reject on provenance; store the measured ceiling per > file."* Under that rule the 32-track MP3 Goldberg would have been an equally defensible > pick and would have avoided the fused file entirely. Lossless is not *worse*, so this > run is not compromised — but the tie-breaker used here (audio quality) is weaker than > it looked, and the next intake should break it on file cleanliness instead. --- ## The standing rule both chains obey **An aggregator's licence *facet* is not a licence.** Neither corpus was intaken from a site's "public domain pool"; each is a **named item** whose rights chain can be read independently of any flag — a composer long dead, plus either a statute (Sousa) or a dated dedication published by the party who made the recording (Bach). ## What is NOT claimed - The generated clips on the article page are **not** Sousa or Bach recordings and are not public domain by inheritance. They are machine-generated audio, and the page says so beside every player. - Neither chain demands credit. The article gives it anyway, which is the point.