gd-costing-quality

Costing Quality Skill

What This Skill Does

Costing Quality is a daily read-only Skill that answers one question for the controller every weekday morning: is inventory valuation and COGS close-ready?

It runs five detection routines, then rolls them into a single severity-scored scorecard. The output is ranked by $ impact and age, top-10 with a tail summary, with drill links back to GoodDay for any row.

This Skill does not draft transactions. Operational follow-through (transfers, adjustments) lives in gd-onhand-ops, gd-safety-stock, and gd-rebalance.

Before You Start

1. Read the shared launch configuration: ../Configuration — Shared.md — schedule, delivery, output shape, ranking, observability, brand, Data Source contract (Section 16).
2. Read this Skill's configuration: ./Configuration - Costing Quality.md — severity thresholds, fallback cost cascade, ignore lists, incoterms matrix, lookback windows.
3. Read the SOP: ./SOP - Costing Quality.md — full procedure and rationale.
4. Apply the Brand Guide: ../🖼️ Brand Guidelines.pdf — colors, voice, badge styling.
5. Resolve data source per the Data Source section below.

Data Source

This Skill describes the business entities each routine needs. The concrete mapping — which MCP tool, verified query, or CSV column populates each entity — lives in ../MCP Tool Catalog.md. When the GoodDay MCP renames a tool or publishes a new verified query, only the catalog changes; this Skill does not. Same for the CSV path: if a merchant's export uses different column names, the catalog records the mapping.

Session context (merchant id, today's date, channels, locations) comes from the connected MCP's standard surface (merchant://channels, merchant://locations, per-response metadata) — no explicit bootstrap call required.

Entities this Skill needs — look each one up in the catalog:

Routine	Entity needed
1 — Zero-COGS Check	Fulfillment history with recorded COGS over the configured lookback window
2 — Zero Landed Cost	Lots currently on hand with no landed cost applied (+ lot-grain $ valuation for ranking)
3 — Negative On-Hand	Negative on-hand positions at SKU × Channel × Location
4 — Lot-Offset Flag	Lot positions across the same SKU × Location (for sibling-sign comparison)
5 — Landed Cost Completeness	Receipts in the configured lookback window + parent PO incoterms type + vendor country of origin
Ranking (all routines)	Lot-grain inventory valuation + fallback cost cascade per Configuration

Output delivery is per ../Configuration — Shared.md Section 5. Drill links are from per-row deep_link fields on every catalog-mapped tool response.

Catalog binding may be partial. If the catalog's current binding for any entity resolves to a verified query or tool that the live MCP doesn't honor (VALIDATION_ERROR: not_found or equivalent), follow ../Configuration — Shared.md Section 12b: surface a routine-specific notice and do not attempt a fallback.

Workflow

The Skill executes five detection routines in parallel where possible, then rolls them up.

Routine 1: Zero-COGS Check         → Units fulfilled at $0 cost, root-cause classified
Routine 2: Zero Landed Cost        → Lots on hand with no landed cost
Routine 3: Negative On-Hand        → Negative positions at any ledger grain
Routine 4: Lot-Offset Flag         → Hidden negative lots offset by positive siblings
Routine 5: Landed Cost Completeness → Receipts missing required components per incoterms
Scorecard: costing-readiness       → Severity-scored roll-up of all five

Routine 1 — Zero-COGS Check

Identify units fulfilled in the lookback window where recorded COGS is zero. Classify each by likely root cause (missing vendor cost / FOB no freight / draft receipt / lot mix). Compute $ exposure using the fallback cost cascade in the Configuration doc.

Routine 2 — Zero Landed Cost

Identify lots currently on hand with no landed cost applied. Compute $ exposure using the fallback cost cascade.

Routine 3 — Negative On-Hand

Identify negative on-hand positions at the SKU × Channel × Location grain. Classify each as cross-channel (positive elsewhere at the Location) or single-channel (no offset).

Routine 4 — Lot-Offset Flag

For each SKU + Location, scan lots for opposite-sign positions (one negative, one positive) at the same SKU+Location. Surface these explicitly — they're invisible at the SKU summary level.

Routine 5 — Landed Cost Completeness

For each receipt in the lookback window, look up the parent PO's incoterms type and the vendor's country of origin. Consult the incoterms matrix in the Configuration doc to determine which cost components are required. Flag receipts missing any required component.

Scorecard — costing-readiness

Read the outputs of Routines 1–5. Do not re-detect. Apply the severity rules defined in ./Configuration - Costing Quality.md Section 1 (Severity thresholds). Compose the headline ("Is the close ready today?"), top-10 punch list, and tail summary per the shared Output shape (../Configuration — Shared.md Section 7).

Output

Follow the output shape from the shared Configuration:

• Headline — overall severity badge + total $ impact + top-line answer to "is the close ready?"
• Scorecard punch list — top-10 across all routines, ranked by $ impact then age
• Tail line — +N more lines totaling $X if more than 10
• Detail by routine — top-10 + tail for each of the five routines
• Persistent issues — items flagged on 3 of last 5 runs
• Temporarily ignored — visibility section for ignore-listed SKUs

Render against the GoodDay Brand Guide. High severity uses brand alert red; Medium uses brand amber; KPI tiles and headers follow brand typography.

Post to the configured Slack channel for scheduled runs. Return inline for ad hoc invocations.

Caveats to Always Include

1. $ exposure on Zero-COGS and Zero Landed Cost is estimated using the fallback cost cascade — actual exposure depends on the true cost once it's recorded. Note this in the scorecard footer.
2. Negative classification is informational — the corrective action sits with ops. Direct the user to gd-onhand-ops, gd-rebalance, or gd-safety-stock for the operational follow-through.
3. Landed cost completeness uses the configured matrix — vendor-specific overrides may shift what's required. Surface any vendor overrides applied in the footer.

Edge Cases

• No fallback cost available for a SKU: Rank that row last in $ impact ordering and flag with a "cost unknown" note. Don't exclude it.
• Receipt with multiple incoterms across blocks: Each block has its own incoterms inheritance from the PO header — evaluate per block.
• Ignored SKU appears in a routine: Exclude from the top-10 ranked list but always include in the "Temporarily Ignored" visibility section with its reason and review date.
• Zero findings in a routine: Report "None found — this is what healthy looks like" rather than omitting the section.