Cycle Count Weeder

Built and stress-tested for you over the weekend during the Keystone AI Enablement engagement. It's yours to keep, use, and change.

What it is

A small AI tool that takes your Ordova cycle-count export and hands back only the products you'd actually walk the floor and count — with the service charges, freight, drop-ship, discontinued items, and other non-count rows stripped out automatically. No more scrolling through and deleting them by hand.

In your words: "When exporting a Cycle Count from Ordova — help me weed through the products I actually would be counting." That's exactly what it does.

What you get back

Paste your export in, and it returns:

• The same columns you gave it, with the rows you don't count removed
• A one-line summary — e.g. "Removed 8 of 18 rows: 4 service/decoration, 1 freight, 1 drop-ship… 10 rows to count."
• A short "flagged for review" note for anything it wasn't sure about — so it never silently drops something you might need

Build it on your own stack (about 5 minutes)

This is a reusable assistant — build it wherever your team already works. The recipe is the same everywhere:

1. Create a new assistant and name it Cycle Count Weeder.
2. Paste the instructions block (at the bottom of this page) into its instructions field.
3. Save. It's ready whenever you run a cycle count.

Where "create a new assistant" lives in each platform — pick the one your team already uses:

• Claude (claude.ai) → Projects → New Project (instructions go in the project's custom instructions).
• ChatGPT (chatgpt.com) → My GPTs → Create (paste into Instructions; building needs a paid plan).
• Gemini (gemini.google.com) → Gems → New Gem (paste into Instructions; free to build and use).

Not sure which? Use whatever your team already lives in — see Learn → Reusable Assistants.

How to use it

1. In Ordova, export your cycle count the way you normally do.
2. Open the spreadsheet, select the rows, and copy them.
3. Open your Cycle Count Weeder assistant and paste.
4. It hands back your cleaned counting list. That's it.

The five-second gut-check (do this every time)

It's an AI tool, so give it a quick glance before you trust a run — five seconds, just good habit:

• Does the "rows to count" number look about right?
• Spot-check two or three Bin numbers against your original — they should match exactly.

If anything looks off, paste it again or check the flagged rows. You're always in control — it never touches your real data in Ordova.

A few gotchas

• Paste the cells — don't attach the file. These tools read pasted text reliably; uploaded files, not so much.
• If your export is very wide, it runs cleaner if you first delete the columns you don't need for counting — keep Item #, Description, Bin, On-Hand, and Category/Status.
• Blanks stay blank. If a vendor or a cost is missing, it leaves it missing — it won't invent anything.

Make it yours — you own the whole thing

The real gift: you have the full instructions, so you can change how it behaves. The part to edit is the DELETION RULES block — it's plain English. Want it to also skip a category? Or to keep inactive items that still have stock on hand? Just edit that short list and save.

And you don't have to do it alone. Paste the instructions into any AI — Gemini, ChatGPT, Claude — and say "help me adjust these rules so it also does X." You're starting from 90% done, not a blank page.

What it does well today — and what a bigger version would add

Today: it reliably cleans a pasted cycle-count export for one person, with you reviewing the result. A real time-saver, ready to use now.

A bigger build would add: reading your file straight from Ordova (no copy-paste), handling every report format automatically, and writing your counts back into the system. That's a larger project — but the version in your hands solves the daily annoyance today.

The full instructions (the engine — yours to keep and tweak)

Copy everything in the box and paste it into your assistant's instructions field. The DELETION RULES block is the part you edit.

# Cycle Count Weeder

You clean up cycle count exports for Keystone Recognition. The user pastes an inventory export. You return **only the rows they actually need to physically count**, in the same columns and the same order they gave you, plus a one-line summary of what you removed and why.

## INPUT HANDLING
The user pastes rows copied straight from a spreadsheet. Columns arrive tab-separated; the chat may collapse the tabs so a row looks like run-on text — the tab delimiters are still there; use them as column breaks. Each new row starts at a line break.
The first row is the column header — use it to identify columns, but do not count it as a data row.
The user may paste the full wide export or a narrowed set of columns — handle either, and always return whatever columns they gave you.
If delimiters are missing, reconstruct by meaning: Item # is the leading code, On-Hand is a number, Status is Active/Inactive/Discontinued/Hold. When unsure, do not guess — keep the row, flag it, preserve its text exactly.

## PRESERVE & ACCOUNT (non-negotiable)
- Reproduce every kept row's cells EXACTLY as received. Never normalize, correct, round, or tidy. The Bin/location column especially is copied character-for-character.
- Keep rows in their original order. Never re-sort.
- Never invent or infer a missing value. A blank stays blank.
- Account for every row: kept + removed = total data rows. If a row doesn't clearly fit a rule, KEEP it and FLAG it — never drop it.

## DELETION RULES — EDIT THIS BLOCK PER PERSON
Remove a row if any of these is true:
1. It's a charge/service, not a physical good — Category is Service/Decoration, Freight, Labor, Setup, Art, or Non-Inventory.
2. Status is Inactive or Discontinued — even if on-hand > 0.
3. Item is Drop-Ship or Made-to-Order — not held in stock.
Keep everything else — active, physically stocked goods — including active items with 0 on-hand.
Status removals are ONLY "Inactive" or "Discontinued." Any other status (Hold, Pending, blank, unfamiliar) is NOT a removal reason: keep and flag.

## OUTPUT
Return the kept rows as a clean table in the original columns, then one line:
`Removed X of Y rows: [breakdown by reason]. Z rows to count[ — N flagged for review].`
Y is the number of data rows (never the header or a TOTAL row). Reasons must sum to X, and X + Z must equal Y. List any flagged rows underneath with a short reason.