INV Exception Triage Copy link

🚨 Blitz data = Excel worksheet UDFs, not MCP tools/connectors. Extract data only by writing =br_...() into a cell (via execute_office_js) and reading the spill — never tool_search for br_*, never refresh_mcp_connectors, and never conclude "no Oracle EBS / Blitz connector is set up" (the Blitz Report add-in is an Excel add-in, not an MCP connector, so it never appears in the connector list — its absence there is expected). Load the extraction mechanism first: read_skill("br-run-reports").

Performance: Read br-analytics-standards at the start of every run (read_skill("br-analytics-standards")). All pipeline steps must follow those rules. Target runtime: under 7 minutes.

Produce a ranked inventory exception register in one workbook. You own classification, severity and the register; br-run-reports does extraction and br-data-dashboard the build. Run steps in order, keeping intermediate sheets. Stuck interface rows are also a period-close blocker — if this run feeds a close assessment, hand the interface-error verdict to br-inv-cost-close-readiness rather than duplicating its scorecard.

Requires: br-run-reports, br-data-dashboard. (Optional: br-anomaly-detection for unusual transaction patterns.)*

Division of responsibility — READ FIRST

You do NOT run Blitz reports yourself. All extraction is by br-run-reports; never call br_* functions or touch a result sheet mid-run. Read-and-recommend only — this skill never corrects on-hand, resubmits interface rows, or approves counts; it ranks and routes.

When to use

"What's broken in inventory", "negative on-hand check", "stuck material transactions", "pending transactions review", "cycle counts waiting for approval", "inventory exception triage". For the period-close verdict itself, use br-inv-cost-close-readiness.

Inputs

• Organization Code (inventory org) — required; ask if not supplied. (Optional: subinventory, date window for the interface errors.)*

Reports used (authoritative names — do NOT broad-search for them)

Pass the documented name directly; search only on #BLITZ_REPORT_NOT_FOUND.

CONFIG

Severity classes

Defaults (edit to the customer): interface aging 2 days; count-approval aging 7 days; adjustment tolerance = the org's own cycle-count tolerance if visible, else 5%.

Output sheets & tab order (left to right)

INV_Exceptions (first), Exception_Register (every exception, ranked), INV_Onhand, Interface_Errors, Cycle_Pending (last).

Pipeline

1. Extract — DELEGATE. br-run-reports runs the three reports → INV_Onhand, Interface_Errors, Cycle_Pending. Wait for each. On-hand can be large — pass the subinventory filter when the user gave one, and if the org is huge, ask before a wide-open pull.
2. Anchor. Record: negative-row count and total negative units/$ by subinventory; interface error count by source and age; pending-count rows and oldest age. All zero → a clean one-line verdict, no register needed.
3. Classify (single execute_office_js call). Build Exception_Register in memory — one row per exception: class, org, subinventory/source, item, quantity/$, age, recommended owner and fix per CONFIG — sorted Critical → High → Review, $ descending inside a class. Write in one .values = call.
4. Dashboard (single call; charts second) (delegate to br-data-dashboard if it fires; else inline). KPI cards per class, exceptions-by-subinventory and by-age charts, top-10 by $ table linking to the register. NEW sheet at position 0; do NOT rename the report sheets; no multi-row copyToRange; clear strays; widen amount columns.
5. Narrative. ≤5 bullets: worst exception, systemic pattern if any (one subinventory, one error source), recommended first fix.
6. Arrange tabs + final validation. Order per CONFIG; INV_Exceptions active. One permitted verification read: register class counts tie to the Step-2 anchor.

Validation

Register ties to raw sheets; Critical only where CONFIG says so; an unassessable source (report not found / module absent) is shown as "not assessed", never silently skipped.

Add-on: slow movers & obsolescence (OPT-IN — extra extract, offer once before Step 1)

Costs one extra report run, answers a different question ("what should we stop stocking"), so it runs only when the request implies it ("slow movers", "obsolete stock", "dead inventory") or the user says yes to a single up-front offer. Never re-ask mid-pipeline.

• Extract INV Transaction Historical Summary (org, trailing 12 months, issues/usage view) alongside the standard pulls.
• Classify items carrying on-hand value: DEAD (no issues in 12 months), SLOW (issues in fewer than 3 of 12 months — CONFIG, label it), each with months-since-last-issue and on-hand $.
• Disposition candidates ranked by value: use-up / transfer / write-down review — always "review candidate", never "write this off" (disposition is the customer's policy call).
• Adds a Slow_Movers sheet before Cycle_Pending and a dashboard section; both degrade cleanly when the add-on didn't run (no #REF! placeholders).
• A DEAD item with recent RECEIPTS but no issues is its own callout — buying what nobody consumes.

Closed-loop (roadmap, not executed here)

Corrections route through the shipped INV uploads (item, subinventory, physical inventory) and the standard resubmit flows. This skill triages; a human fixes. Slow-mover stock-parameter fixes (min/max, safety stock) belong to br-inv-stock-optimizer, which owns that gated write.