# Hands-On Lab Prompts
## Companion to Large Language Models for Business and Finance

This document contains every hands-on prompt used during the workshop, keyed directly to the slide numbers. You can copy and paste these prompts directly into your Goose workspace or terminal.

All prompts in this workshop are 100% self-contained and run on the book's authentic running specimen: **Acme Industrial Technologies** (NYSE: AIT). They require no external files or downloads.

---

### Slide 14 — Lab Exercise 1: Observing Variance on Identical Prompts

**Context**: In Cycle 1, we test next-token prediction and run-to-run sampling variance. Run this identical prompt three times in three clean sessions to observe how wording and figures drift while the model speaks with complete confidence.

```prompt
Document:
"Acme Industrial Technologies (NYSE: AIT) reported fourth-quarter consolidated revenue of $4,280 million, a 6% increase compared to $4,038 million in the prior-year period. Operating income expanded to $640 million from $595 million, reflecting price realization across heavy equipment lines."

Instructions:
Summarize the fourth-quarter performance for Acme Industrial Technologies.
State the total revenue and the year-over-year growth rate.
Use two concise sentences.
```

---

### Slide 24 — Lab Exercise 2: Testing Precision in Long Transcripts

**Context**: In Cycle 2, we test the stateless courier, transcript re-shipment, and attention decay across long threads. Submit these five prompts **sequentially, one turn at a time** in the same session. In Turn 5, observe whether the model still respects the European currency constraint established in Turn 1.

#### Turn 1
```prompt
Note that Acme Industrial Technologies reports in USD, but European sales occur in EUR.
```

#### Turn 2
```prompt
For Acme Industrial Technologies in Q4, what was total reported revenue?
```

#### Turn 3
```prompt
What was operating income for the fourth quarter, and what drove margin expansion?
```

#### Turn 4
```prompt
Did Acme disclose any supply chain friction in European markets?
```

#### Turn 5
```prompt
Reconcile European currency exposure using our Turn 1 note.
```

---

### Slide 33 — Lab Exercise 3: Grounding Queries in Primary Filings

**Context**: In Cycle 3, we test grounding model responses in verbatim regulatory filings (Note 8 of the Acme Industrial Technologies 10-K) to eliminate reliance on pre-trained memory. Paste this entire prompt into a clean session.

```prompt
Document: Note 8 (Debt Covenants and Credit Facilities), Acme Industrial Technologies 10-K:
"Under the Senior Credit Facility, consolidated net funded debt (funded debt less cash and equivalents) to EBITDA shall not exceed 3.50 to 1.00 at the close of any fiscal quarter. Consolidated funded debt stood at $2,800 million against cash and equivalents of $360 million. Consolidated EBITDA for the trailing twelve months reached $1,220 million."

Instructions:
Identify the restrictive debt covenant governing Acme Industrial Technologies.
State the covenant ceiling and current net debt balance.
If any figure is not explicitly disclosed in the document above, output 'NOT FOUND'.
```

---

### Slide 45 — Lab Exercise 4: Delegating Math to a Python Tool

**Context**: In Cycle 4, we enforce the separation between probabilistic text prediction and deterministic calculation. We instruct the agent to write and execute a Python script to calculate net leverage and evaluate covenant compliance against Note 8.

```prompt
Data:
Acme Industrial Technologies Financial Figures (FY2025):
Funded debt: $2,800 million.
Cash and cash equivalents: $360 million.
Trailing twelve-month EBITDA: $1,220 million.
Note 8 Covenant Ceiling: 3.50x net leverage.

Instructions:
Calculate Acme's net leverage ratio using Python:
net_debt = 2800 - 360
net_leverage = net_debt / 1220

MANDATORY EXECUTION RULE:
Do not calculate the ratio in narrative text.
Write and execute a Python script in your sandbox to compute the ratio.
State whether Acme remains in compliance with its 3.50x ceiling.
Display the executed script and exact stdout terminal output.
```

---

### Slide 55 — Lab Exercise 5: Running a Two-Session Extraction and Cross-Check Pipeline

**Context**: In Cycle 5, we test multi-agent division of labor by executing distinct tasks across separate chat sessions. Session 1 extracts structured JSON data; Session 2 independently audits and verifies the extracted payload against Note 8.

#### Session 1 (The Extractor)
*Run this prompt in your first chat tab:*

```prompt
Document: [Excerpt from Acme Industrial Technologies 10-K, Note 8:
"Under the Senior Credit Facility, consolidated net funded debt to EBITDA shall not exceed 3.50x. Consolidated funded debt stood at $2,800 million against cash of $360 million. Trailing EBITDA reached $1,220 million. Consolidated fourth-quarter revenue totaled $4,280 million."]

Instructions:
Extract the metrics into a strict JSON object: company, revenue_millions, funded_debt_millions, cash_millions, trailing_ebitda_millions, covenant_ceiling.
```

#### Session 2 (The Reviewer)
*Copy the JSON output produced by Session 1, paste it into a fresh chat tab, and run this prompt:*

```prompt
Audit the extracted JSON payload against Note 8.
Confirm whether net debt ($2,440M) divided by EBITDA ($1,220M) satisfies the 3.50x covenant.
Output an itemized verification log with PASS or FAIL status.
```

---

### Slide 63 — Creative Session: Writing Your LinkedIn Project Summary

**Context**: At the conclusion of the afternoon session, use your configured Goose agent to draft a professional summary of your project showcasing your governance and agentic systems skills.

```prompt
Draft a concise, professional LinkedIn post describing my work today at Schulich:
- Built an auditable financial analysis pipeline on corporate 10-K filings.
- Replaced unguided chatbot arithmetic with deterministic Python execution via MCP.
- Implemented multi-session division of labor (Extractor/Reviewer) and audit dossiers.
- Aligned pipeline controls with institutional model risk management principles.
- Focused on institutional defensibility and eliminating compounding error in finance.
Keep the tone professional, analytical, and free of marketing buzzwords.
```