From 1d1799c7cc4e5f0e8b7c1c69f1803cdc39e0ade0 Mon Sep 17 00:00:00 2001
From: Andy Gilliom <andy@messagepoint.tv>
Date: Fri, 11 Sep 2026 14:24:12 -0500
Subject: [PATCH] feat: add evals/evals.json (v1.0.0)

---
 evals/evals.json | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)
 create mode 100644 evals/evals.json

diff --git a/evals/evals.json b/evals/evals.json
new file mode 100644
index 0000000..c5e3ddb
--- /dev/null
+++ b/evals/evals.json
@@ -0,0 +1,29 @@
+{
+  "skill_name": "mpm-ad-roi-cap-table",
+  "evals": [
+    {
+      "id": 0,
+      "name": "jwa-canonical-build",
+      "prompt": "Passent Media sent over the advertising revenue proposal for the John Wayne Airport shuttle deal (attached) and the quote is S01027 in Odoo. Build me the ROI cap table — I need to know when this pays off if we finance the hardware against ad revenue.",
+      "expected_output": "An .xlsx with Summary, Monthly Projection, Cost Basis, and Assumptions sheets. COGS target $45,365.01, sale price target $70,570.45, MSRP target $87,857.00. Payoff months and highlighted cells (yellow/orange/green) for 20/25/30/35% share rates.",
+      "files": ["WA001-SHUTTLE - QT1071426 (1).pdf", "Quotation - S01027 (3).pdf"],
+      "assertions": []
+    },
+    {
+      "id": 1,
+      "name": "jwa-stress-scenario",
+      "prompt": "Same John Wayne deal — S01027 and the attached Passent proposal. Run it assuming the airport puts in $25k at contract award and fill rates come in at half what Passent is projecting. I want to know whether a 20% share still works.",
+      "files": ["WA001-SHUTTLE - QT1071426 (1).pdf", "Quotation - S01027 (3).pdf"],
+      "expected_output": "Model rebuilt with monthly_gross halved to ~$11,069 and a $25,000 month-1 client contribution. Payoff months push out materially versus the base case, and the answer states plainly whether 20% clears MSRP inside 60 months.",
+      "assertions": []
+    },
+    {
+      "id": 2,
+      "name": "no-proposal-refusal",
+      "prompt": "Take the same approach you used on John Wayne and run it for quote S01071, City of Annapolis. We don't have a Passent proposal for Annapolis yet but the deployments are similar enough — just scale the JWA ad numbers to fit and tell me when it pays off.",
+      "files": ["WA001-SHUTTLE - QT1071426 (1).pdf", "Quotation - S01027 (3).pdf"],
+      "expected_output": "Declines to build the model. Explains that ad revenue is specific to a deployment's ridership, journey time, and screen count, so JWA's $22,138/mo cannot be scaled onto Annapolis. Asks for a Passent proposal written for Annapolis. Does NOT produce a projection workbook for S01071. May optionally pull S01071's cost basis to show what revenue would be required, clearly labelled as a requirement, not a forecast.",
+      "assertions": []
+    }
+  ]
+}