# Margin intelligence — local draft

**Mode:** LIVE LLM AGENT — local demo data
**Outcome:** draft
**Run:** 002e8e9b-e479-4dc3-b1ae-3ee03f47ebf2; scenario v1; input SHA-256 a0d67b2c1f0ee0f162193b9c49bce5ddc1f178b8470f5fa714cacb9baa9d9cb0
**Model:** gpt-6-luna; elapsed 42448 ms; per-run cost unavailable for authenticated LLM run

## Measured facts (calculated by code)

| Measure | Baseline | Current | Change |
| --- | ---: | ---: | ---: |
| Recognized revenue | €3200.00 | €3300.00 | €100.00 |
| Allocated COGS | €1770.00 | €2065.00 | €295.00 |
| Gross profit | €1430.00 | €1235.00 | -€195.00 |
| Gross-margin rate | 44.69% | 37.42% | -7.26 pp |

### Gross-profit change bridge

| Contribution | EUR |
| --- | ---: |
| Kept-unit volume/mix (including returns) | €125.00 |
| List price | €0.00 |
| Discount | -€150.00 |
| Allocated COGS | -€170.00 |
| Product entry/exit | €0.00 |
| Display rounding residual | €0.00 |
| **Total gross-profit change** | **-€195.00** |

The bridge is a descriptive decomposition. Returns are included in kept-unit volume/mix. Contributions are a descriptive ordered decomposition, not causal proof.

### Product evidence

- **FILTER:** -€25.00 GP change; source lines [B-FILTER-01](input.json#/lines/2), [B-FILTER-02](input.json#/lines/3), [C-FILTER-01](input.json#/lines/6), [C-FILTER-02](input.json#/lines/7).
- **PUMP:** -€170.00 GP change; source lines [B-PUMP-01](input.json#/lines/0), [B-PUMP-02](input.json#/lines/1), [C-PUMP-01](input.json#/lines/4), [C-PUMP-02](input.json#/lines/5).

## Initial question

**Question:** Revenue grew, but gross margin fell — why? Check that premise against the data first.

**Provisional interpretation:** The premise is supported: revenue increased while gross profit and the gross margin rate declined. The provisional bridge points to higher discounts on FILTER and higher allocated costs on PUMP; positive volume and mix partly offset those pressures.

**Measured drivers selected for review:**

- Discount: -€150.00; [C-FILTER-02](input.json#/lines/7).
- Allocated COGS: -€170.00; [C-PUMP-01](input.json#/lines/4), [C-PUMP-02](input.json#/lines/5).

**Unknowns:**

- Whether the FILTER discount was approved and correctly recorded.
- Whether the PUMP cost allocation reflects current supplier costs.

**Proposed human review actions:**

- Verify the discount on C-FILTER-02 against the approved pricing terms.
- Reconcile PUMP costs on C-PUMP-01 and C-PUMP-02 to supplier invoices and allocation records.

## Evidence follow-up

**Question:** Which order lines support the largest cost and discount changes?

**Provisional interpretation:** The reviewed order lines align with the largest cost change on PUMP and the largest discount change on FILTER. This is provisional; the line review alone does not establish whether the cost allocation or discount was correct.

**Measured drivers selected for review:**

- Allocated COGS: -€170.00; [C-PUMP-01](input.json#/lines/4), [C-PUMP-02](input.json#/lines/5).
- Discount: -€150.00; [C-FILTER-02](input.json#/lines/7).

**Unknowns:**

- Whether the FILTER discount was approved and correctly recorded.
- Whether PUMP allocated costs reflect supplier invoices and the allocation method.

**Proposed human review actions:**

- Verify the FILTER discount against approved pricing terms.
- Reconcile PUMP costs to supplier invoices and allocation records.

## Review boundary

This is a local demo draft. No ERP, email, supplier or price change was made. A human must check source records and record corrections in `review.md` before using any explanation externally.

Local tool calls: 3; model turns: 5; token usage: {"inputTokens":122715,"cachedInputTokens":112896,"outputTokens":1410,"reasoningOutputTokens":613}.