Inventory Reorder Forecast

"47 in stock" is not a number you can act on

Is that three weeks of cover or three months? Your supplier needs two weeks' notice either way. Working it out properly means a spreadsheet per SKU — so it gets done for the top five and guessed for everything else.

Those guesses cause both the stockouts and the dead cash.

Days of cover, with the arithmetic

days_of_cover = on_hand ÷ daily_velocity, and the reorder trigger is cover ≤ lead_time + safety. Every line prints its inputs, so you can check a recommendation instead of trusting it.

Four verdicts per SKU: REORDER NOW, REORDER SOON, OVERSTOCKED (cash tied up, not a reorder), or a data verdict.

The trap that makes naive forecasts wrong

A stockout corrupts its own velocity. If an SKU hit zero during the window, it couldn't sell what it didn't have — so its measured velocity is suppressed, and reordering off that number under-orders your best seller. Again.

Those SKUs come back flagged STOCKOUT_SUPPRESSED, with the true velocity declared unknown-and-higher rather than quietly used.

Same for divergence: when 28-day and 7-day velocity differ by more than 50%, you get both numbers and a flag. That gap is either seasonality or a stockout, and those demand opposite actions — so the agent won't pick one for you.

Under 14 days of history it returns INSUFFICIENT_HISTORY. It will not extrapolate a trend from one week.

Read Flagged before the reorder list

It's the section that tells you which numbers are lying to you.

What you supply

Your own Claude Code login — and two exports. Sales-by-SKU and inventory on-hand (Shopify: Analytics → Reports → Sales by product; Inventory → Export), dropped into ~/inventory/exports/.

Then your supplier lead time and safety days in the rules file. Leave those blank and every SKU returns a data verdict instead of a quantity — deliberately, since cover means nothing without a threshold to compare it against.

No store API key. Exports are the input, which is exactly why nothing can place an order.

The guard is a file, not a promise

~/inventory/policy/FORECAST-RULES.md carries NEVER_AUTOORDER, the cover formula, the stockout rule and your numbers. The skill reads it before every run.

Where this stops

It reads exports and drafts a brief. It does not place purchase orders, contact suppliers, or change stock levels.

It won't claim seasonality — it can't see last year in a 28-day export — and it won't compute margin unless your export carries cost.

Verified on build: the skill, the working tree, the NEVER_AUTOORDER guard, the cover formula and both data traps are present on a fresh fork. Running it on your own exports is yours to do.