miFormulas and your own AI assistant

You are converting a perfume formula into an import file for the miFormulas app. Attached is the formula as a photo, scan, PDF or spreadsheet, and possibly a CSV export of my materials inventory.

Do this:

1. Transcribe the formula exactly, line by line: material name, dilution in percent, weight in grams. Copy names verbatim, including brand or supplier tags such as (Giv), (IFF), (Firm). If a line shows no dilution, use 100. Mark a line that is a solvent (ethanol, DPG, DEP, IPM, TEC, or benzyl benzoate used as a carrier) with "solvent": true. If the source gives only percentages or parts, convert them to grams on a 100 g batch and say so in "notes". Transcribe numbers exactly; decimal commas become decimal points in the JSON.
2. If a materials inventory (CSV) is attached, use its exact spelling of the material name in "material" whenever a line clearly matches an entry. When you are not sure, keep the transcribed name in "material" and add your best guess in "comment" ("possibly: …"). Never silently substitute one material for another.
3. Do not convert dilutions or weights to what the inventory stocks. Keep them exactly as printed; the app flags unknown dilutions and I convert them myself.
4. Use the formula name written on the sheet. If there is none, ask me instead of inventing one. Never invent version labels.
5. Give the JSON in a fenced code block (three backticks), and nothing else inside that block, in this shape:

{
 "type": "miformulas-import",
 "name": "<formula name as written>",
 "category": "<category if I named one, else omit this line>",
 "source": "<photo, PDF or spreadsheet, and the origin or author if known>",
 "notes": "<batch conversion, illegible lines, other remarks; omit if none>",
 "lines": [
  {"material": "Hedione", "dilutionPct": 100, "weightG": 17.5},
  {"material": "Iso E Super", "dilutionPct": 10, "weightG": 2, "comment": "possibly: Iso E Super (IFF)"},
  {"material": "Ethanol", "dilutionPct": 100, "weightG": 80, "solvent": true}
 ]
}

6. Below the code block, outside it, tell me in a few lines: the total weight (and whether it is a round number such as 100 or 1000, which suggests a complete transcription), how many lines matched the inventory and how many are new, and every line you could not read or were unsure about, as questions.

One formula per file. If the source shows several formulas, ask which one you should convert first.

Attached is a CSV export of my own materials inventory from the miFormulas app (columns: Name, CAS, Alternative names, Category, Supplier, Amount purchased, Purchase date, Cost EUR/g, IFRA limit %, Pyramid, Solvent, Cupboard, Fridge, Freezer, Density g/ml, Stock g (tracked), Dilutions %, Description). Help me check and complete it. Do not rename anything; I enter your proposals by hand.

Give me a table with one row per material that has something missing or doubtful, with these columns:

- Name, exactly as in the file.
- CAS: the existing value, or a proposal when it is empty, or a correction when the existing value does not belong to that material. Mark proposals and corrections as such.
- Alternative names: well-established trade names or common synonyms that are missing (for instance Ambroxan for Ambroxide, or patchouli ethanone for Iso E Super), separated by semicolons. Only names a perfumer would actually use; no systematic chemical names.
- IFRA limit: the IFRA Standards limit for Category 4 (fine fragrance) in percent of the finished product, with the amendment number you took it from. Write "no restriction" when the material is not restricted under the current standards, "specification" when the standard is a purity specification rather than a limit, and "unknown" when you are not certain. Never guess a number.
- Pyramid: Top, Top-heart, Heart, Heart-base or Base.
- Category: a proposal from the categories already used in the file, or a new one if none fits.
- Odour: one line, for the description field.
- Remarks: anything else, for instance a name that looks like a typo or a dilution that seems odd.

Below the table, list the likely duplicates: pairs of rows with the same CAS number or nearly identical names. Say for each pair whether they are really the same material or legitimately different products with the same CAS (a vetiver oil from Java and one from Haiti share a CAS number and must stay separate). Do not propose merging on the CAS number alone.

Keep the table plain (no merged cells) so that I can copy it into a spreadsheet. Skip materials where everything is present and plausible.

Below is a list of perfume formula names that were imported from another program, where every trial was a separate formula. Propose how to group them into formulas with a version history. Do not rename anything beyond splitting a name into its parts, and do not merge names that only look similar.

Rules:

- A stem followed by a version number ("Aura v04", "Aura V4", "Aura v.4", "Aura versie 4", "Aura 4") is a version of the formula "Aura". The version number is the number in the name.
- A name with a suffix that describes a presentation rather than a new trial ("20%", "EDT", "soap", "44gr", "10 ml", "candle") is a version too, with that suffix as its label; place it after the version it is based on. If that base version is not named, note which one it most likely belongs to and mark it as a question.
- Different stems stay separate formulas, even when they share words. A date in a name is not a version number; keep it as a remark.
- When a name fits more than one reading, list it under "questions" with the readings, instead of choosing.

Answer with a table with the columns: Original name | Formula | Version | Label | Confidence (high, medium, low) | Remark. Sort by formula, then version. After the table, list the questions, and the names you left as separate single-version formulas.

Here are the names:

(paste the names here, one per line)

I want to run the miFormulas app on my own web server so that all my devices share one data file. Attached are the server endpoint `data.php` from the miFormulas repository and section 18 of its manual, whose part B, "Your own web server with PHP", is the one I am following. Read both first.

My situation: (describe your device and what you know, for example: a Synology DS220+ with DSM 7.2, Web Station is installed but I have never used it; or: a Raspberry Pi 4 with Raspberry Pi OS; or: shared hosting at provider X with PHP 8.2 and FTP access. Say whether you can already reach the device by name or address in a browser, and whether you want to reach it from outside your home.)

Guide me step by step. Ask me what you need to know before each step rather than assuming. For every step, tell me exactly where to click or what to type, and how I can check that it worked. Cover: installing PHP and the web service if needed, placing `index.html`, `data.php` and the `data` folder, giving the web server's user write access to `data`, setting a token in `data.php` (I will choose the token myself and will not tell you what it is), putting my `miformulas-data.json` in the `data` folder, whether my web server serves that folder to anyone who asks for it and what to do about that on the web server I actually run (an `.htaccess` is not read by all of them, and moving the folder out of the web root with `$DATA_DIR` is the answer that always holds), opening the app in the browser and entering the token, reading the `?ping=1` self-check of `data.php` with me, and, if I want access from outside, the safe way to do that (a VPN such as Tailscale, with https). When something can go wrong in a way that costs data, warn me before that step.

I am setting up the perfume app miFormulas so that my computer, my laptop and my phone work on the same data. I am following way A of section 18 of its manual: a Cloudflare Worker with an R2 bucket. Attached are `server/worker.js` from the miFormulas repository and section 18 of the manual. Read both first, and follow the manual where it and you disagree about the Cloudflare dashboard: it was written from the dashboard as it is, and you have not seen it.

Help me with these, one at a time, and ask what you need to know before each:

1. A token. Give me one long random string of about twenty characters, letters and digits only, and tell me where it goes and where it must never go. Do not ask me what I chose afterwards.
2. What the Worker does, in plain words, so that I understand what I am pasting: what it stores, what the token protects, what the ETag check prevents and what the daily snapshots are for. Answer from `worker.js`, not from memory.
3. Reading the check of step 8 in the manual. I will tell you what the browser answered at the Worker's address; tell me what it means and which step to go back to.
4. Anything that goes wrong afterwards: I will paste the exact message the app or the browser shows, and you tell me where it comes from, using `worker.js` and section 18.

What I would like you not to do: describe Cloudflare screens from memory, invent button names, or suggest the paid plan. If I ask something the manual answers, point me to the step in it.