Blind Cycle Count Capture Sheet
Build an internal tool that hides system on-hand from your counters, captures their physical count, then reveals variance only to a reconciler who accepts or orders a recount.
A login-protected blind count tool where counters key physical quantities without ever seeing the system number, reconcilers review revealed variances and accept or recount, and every finalized count exports as a clean variance CSV.
Before you start
- A free Supabase account
- A free Vercel account
- A free Resend account
- A count task list in a Google Sheet or CSV (system on-hand can live here, it just stays hidden from counters)
The problem this kills
When a cycle counter can see the system on-hand number, the count stops being a count. It becomes a confirmation. People round toward the expected figure, skip the back of the bin, and "count to the system" without meaning to. Your stock accuracy reports then look great and tell you nothing.
A blind count fixes the incentive: the counter never sees what the system thinks is there. They just count what's physically in front of them and key it in. Only afterward, and only for the person doing reconciliation, does the variance appear. That single change is the difference between a number you can trust and a number that quietly lies to you.
Most teams know this, but the off-the-shelf options are either a clipboard (no audit trail, no controls) or a six-figure WMS module. This plan gives you the third option: your own small, governed tool, built in an afternoon.
What you'll build
A simple, login-protected web app with two clear roles:
- Counters open their assigned tasks, see the location and SKU, and key the physical quantity. The system number is hard-hidden. They never see expected, never see variance.
- Reconcilers open completed counts, see the revealed variance (physical vs. system), and make one decision per line: accept the count, or send it back for a recount.
It supports a second recount round, guards against the same count task being processed twice, records who did what and when, and exports finalized variances as a CSV in the exact columns your inventory system expects.
What's inside the Implementation Plan
The plan is a single markdown file you paste into an AI coding agent (Claude Code). It walks the agent through building the whole tool, step by step, with a ready-to-copy prompt at the end of each step.
Crucially, the plan opens by interviewing you about your warehouse - your locations and SKU naming, how count tasks get assigned today, your typical and peak count volumes, your tolerance rules for what counts as an acceptable variance, and your messy edge cases (multi-location SKUs, lot/serial tracking, units of measure). It reflects a short tailored spec back to you for a thumbs-up, then shapes the data model and every later step around your real answers. You get a tool fitted to your operation, not a generic template you have to bend.
The governance it includes (this is the point)
- Login so only your team can open the tool.
- Row-level security so a user only ever sees their own organization's counts.
- A hard blind barrier: the expected quantity is never sent to the counter's screen, so it can't leak even in the browser.
- A human approval gate: nothing is finalized until a reconciler reviews the variance and explicitly accepts or recounts. The tool drafts, a person decides.
- A complete audit trail: who counted, who reconciled, what they decided, and when.
- Duplicate guards so the same count task can't be counted or finalized twice.
Who it's for
Cycle count associates, inventory control specialists, and warehouse supervisors who want trustworthy stock accuracy numbers without buying a new WMS. If you can fill in a spreadsheet, you can build this.
You've got this - paste the first prompt and let the agent interview you.