The rulebook.

Your audio never leaves your machine. It is decoded and analyzed in your browser, and it is never transmitted.

S3 is mandatory wherever this appears. Stating what does cross the boundary is what earns belief in the rest.

The measurements do travel. Every completed report is handed to the page that embeds the tool. For signed-in members, the host also stores it against their account. The audio is never part of that. What is stored is published in full, and the list changes only in public.

Amended 8 September 2026. This sentence used to end "the measurements and the file name, nothing more." That enumeration was already false — the stored row carries the findings, the genre profile, the delivery target and a timestamp as well, and had done since it was created. One sentence was making two different promises. A boundary promise — the audio is not in it — is absolute and survives every release. An inventory promise — exactly these fields, nothing more — is a snapshot of an implementation and is false the moment a feature adds one. So the boundary stayed and the enumeration went, replaced by a stronger commitment rather than a weaker one: the list is published in full below, and it changes in public.

S1 describes one thing: the AudioVibes tool running in your browser. It is not a statement about ABA Media Platform as a whole, and it does not describe how any other ABA product handles audio.

Other ABA products handle audio differently, and some hold it by design. Each one carries its own numbered statement in this file.

Where a product has no numbered statement here, no copy may describe that product’s audio handling — not by adapting S1, not by context, and not by implication.

Approved 17 September 2026. It makes no claim about anything unbuilt. It says what an approved statement covers and what it does not, which is why S17 and S18 exist as statements of their own.

Your audio never leaves your machine. If you’re a member of the ABA Media Platform and signed in, we store your results in your member profile.

Approved 11 October 2026. Where the four tools appear together, S1, S17 and S18 said the same sentence four times. S19 says it once. Each tool’s own statement stays the wording on that tool’s page, and S3 stays the full account of what travels. Amended 12 October 2026, when the other three tools began saving for members: “your ABA AudioVibes results” became “your results”.

Your audio never leaves your machine. ABA Phaze Check and ABA Compare decode and measure it in your browser, and never transmit it. If you’re a member of the ABA Media Platform and signed in, each tool stores its readings in your member profile: the measurements and the names of your files and tracks, never the audio. Signed out, neither tool sends anything at all.

Approved 6 October 2026. A statement of its own, because S1 covers ABA AudioVibes only and is not stretched to cover another tool. Second sentence amended 12 October 2026, before member saves shipped for these tools. It used to read: “Today neither tool sends anything at all: no measurements and no file names.”

Your audio never leaves your machine. ABA Pocket Check decodes and measures it in your browser, and never transmits it. If you’re a member of the ABA Media Platform and signed in, it stores its readings in your member profile: the measurements and the names of your files and tracks, never the audio. Signed out, it sends nothing at all.

Approved 8 October 2026. The measuring runs on the page itself, in small steps, and the tool’s test records each request the page makes. Second sentence amended 12 October 2026, as S17’s was. It used to read: “Today it sends nothing at all: no measurements and no file names.”

There is no fetch, XMLHttpRequest, WebSocket or sendBeacon anywhere in the build, and the page makes no external request of any kind. The only outbound path is a postMessage to the page that embeds it.

A structural fact rather than a policy promise, which is why it is the strongest sentence available — and why it is never softened into "we don't upload your files".

Saved clips are held in your browser's own storage (IndexedDB) — up to six clips within a 420 MB budget, oldest evicted first. Browser storage is not encrypted: anyone using that browser profile can replay them.

The first never appears without the second.

The page never sends a member id. The host derives the member from the signed-in session on the server.

No account is required to analyze a master. Without one, appraisals are kept in that browser's own storage; signing in is what makes them follow you to another browser or another machine.

Appraisals you made before signing in are added to your account when you sign in. A row leaves this browser only once the account has confirmed it, so nothing is discarded — if the account cannot be reached, the appraisals stay where they are and the app says so.

An earlier version of this described a silent wipe, and it cost the site owner two appraisals before it was corrected. That is what the correction record below is for.

“Remove from my list” hides the reading from you and from everyone. Its record is kept, with its number and its time, and the removal is logged.

Amended 16 October 2026. Until then this read: “Clear removes the appraisal from the account permanently. It is not a local hide.” ABA now keeps a record that is only added to; to have your own details erased, see Your data at ABA.

Appraisals belong to the account, not the person at the keyboard. If you sign in as a client and run evaluations, that client will see them. Clear them before you sign out.

The app acts on messages only from the site it is published on. A page that embeds it anywhere else cannot supply the member name it displays or the appraisal list it shows — it will show only the history held in that browser. What still travels outward to any page that embeds the tool is S6 — the report itself, plus a build stamp and a height ping — and S6 is open.

What this does not claim, stated because a security section that lists only reassurances is discounted on sight: it is not a defense against the site you are already on. A page that embeds the tool controls the frame around it and could draw anything it liked without involving the tool at all. The guarantee is narrower and exact.

ABA Custody holds the working assets of your production while it runs: every performance and every mix. When you leave, they are handed to you, and ABA keeps no audio or video after that.

Whenever you leave with recorded work, you leave through Milestone 8, full or light. You can ask to leave at the end of any task. At the end of Milestone 8, your tracks and video are handed to you, signed for. That copy is yours and it does not expire.

At every close that makes a mix, you also take a copy of that mix home, to hear how it translates.

The client always leaves with a signed-for, complete set of every asset they paid for, handed over at Milestone 8, full or light, and a copy of every mix at the close that made it.

While a production is running, ABA holds the working assets and the client holds every mix so far. Once they leave through Milestone 8, they have everything they paid for. There is no further remedy, and ABA says so plainly.

Approved 15 September 2026; amended 19 October 2026. A milestone is one stage of an ABA production; Milestone 8 is the last, and the way out passes through it. The first block is quoted whole: on its own, the name reads as a vault for your work, which is the one thing it is not. The schedule on which the record steps down is approved and not published yet, because it promises you a notice before each step and the sending of those notices is not built yet.

• File name — the name of the file you analyzed, without its extension.
• Score and grade — the production score and its letter.
• Delivery target — the standard the file was measured against.
• Integrated loudness, true peak and sample rate — three measurements lifted out of the report so your list can show them.
• Genre — the profile the score ran under.
• Analyzed date — when the measurement was made, copied out of the report, so a report added to your account later still shows the day you ran it.
• Grouping key — empty unless you have dragged a row to correct how your history groups. You wrote that one, not us.
• Member id — which account the row belongs to.
• Report — the whole report, opened up below. For ABA Pocket Check, ABA Phaze Check and ABA Compare it is the measurements as text, the names of your files and tracks, and the tool’s own record page, the same page its Download button gives you.
• Tool — which of the four tools made the reading.
• Set key — a random code made in your browser, so measuring the same files again updates one row instead of adding another. It has no meaning outside your account. Empty for ABA AudioVibes.
• Removed date — empty unless you pressed “Remove from my list”; then the time you did. A removed row is hidden from you and from everyone, and kept.

For the other three tools, the score, grade, delivery target, loudness, true peak, sample rate, genre and grouping columns stay empty. The site adds two of its own: a row id and a created date. The account itself holds your email address, the name you asked us to call you if you gave one, the date you joined and the date you last signed in.

The site also keeps a record of each action on your rows: its own number, the time, your member id, what was done (saved, updated, removed, brought back), which row and which tool. Your email, your name and your file names are not in it. It is only ever added to.

If you choose tips and update notices, the site keeps each choice as its own row: the time, your member id, on or off, the version of the words you saw, and where you chose (signing in, or your member page); a choice made at sign-in also says whether the box started ticked and the country our host reported. The latest row is your choice. Rows are only ever added.

• File name, score, grade, delivery target, integrated loudness, true peak, sample rate, genre, grouping key, member id and the report, meaning the same as above.

The host adds four of its own: a row id, a created date, an updated date, and an owner id. Last read from the live collection on 9 September 2026.

• measurements — each value the tool computed.
• findings — the written findings, in the words shown on screen.
• scores — the per-category scores and the weight each one carries.
• categories — the category definitions those scores ran against.
• engine — the build of the analyzer that produced it.
• source — the file name, duration, sample rate and channel count of what you loaded.

That is the whole row. Add a field and this list changes the same day, with the date above it.

The report is posted to whichever page embeds the tool. On museguitarist.com and abamediaplatform.com that page is ours. Any other site that embeds the tool would receive the reports run inside it — the audio, never.

Amended 9 September 2026. This statement used to read "on museguitarist.com that is your own site." The other statements here address you, the reader; that one was the only place "your" meant the owner — so to anyone but him it claimed this was your site, which is false. It also named one host when there are two. It was written as a note to one person and published to everyone, and the pronoun's meaning changed without the word changing. The correction does not close S6.

This is the outbound mirror of S11, and it is a hardening item that has not been closed. The reason it is still open is a real observed failure rather than a shortcut: a mismatched target origin is dropped silently, so a wrong guess loses a reader's results with no error anywhere. It is never presented as resolved, and closing S11 did not close it.

The page loads its webfonts from Google Fonts, which is a request to a third party on every page load. It carries no audio and no measurements — it is a static stylesheet URL — but it does disclose the visitor's IP address to Google. Self-hosting the fonts would remove it.

Resolved by inlining the font faces into the build, so the assembled file contains no external address at all. It is kept here, struck, because pages published before that build carried it — and because a reviewer checking the old citation should find out from us that it has moved, not from the discrepancy.