oracle-dd

name	oracle-dd
version	2.0
type	autoresearch
description	Deep deal due diligence with multi-model verification
user-invocable	true
agent	ORACLE
agent_model	claude-opus-4-6
mcps	["paperclip","aegis","github"]
gstack_skills	["/investigate"]
eval_metric	dd_checklist_completion
eval_budget	120s
guard	false_flag_rate < 0.10

/oracle-dd — Autonomous Due Diligence

What it does

Runs comprehensive due diligence on CPA acquisition targets using Paperclip for public data gathering, AEGIS for multi-model analysis, and /investigate for systematic root-cause analysis of any red flags.

Protocol

Step 0: Load context

Read these files:

• memory/RESEARCH_MEMORY.md — prior research, active deals
• dark-factory/ — existing LOIs, valuations
• HENRY_BRAIN.md — deal criteria (0.4x revenue, 75-80% post-AI EBITDA)

Step 1: Target identification

Confirm the deal target. Required inputs:

• Listing ID (e.g., TXS5345)
• Asking price or revenue figure
• Source (APS.net, BizBuySell, broker referral)

Step 2: Public records research (Paperclip)

Fetch and analyze these sources for the target:

1. TX State Board of Public Accountancy

   • Verify CPA license status, disciplinary history
   • Confirm firm registration and authorized services

2. TX Secretary of State

   • Entity search: formation date, registered agent, status
   • Check for any franchise tax issues

3. APS.net listing page

   • Revenue, asking price, owner details, transition terms
   • Client concentration, service mix, employee count

4. Google reviews + Yelp

   • Client satisfaction signals
   • Volume and recency of reviews
   • Any complaint patterns

5. LinkedIn

   • Owner profile: age indicators, career history, retirement signals
   • Employee profiles: team composition, tenure

6. Court records (Harris County)

   • Any active litigation involving the firm
   • Tax liens or judgments

Step 3: Financial analysis (AEGIS → DeepSeek R1)

Route quantitative analysis to AEGIS MCP with DeepSeek R1:

1. Revenue verification

   • Cross-reference listing revenue with public indicators
   • Flag any revenue > $500K that lacks visible market presence

2. Valuation model

   • Apply HENRY standard: 0.4x gross revenue
   • SBA 7(a) eligibility check (< $5M, 10% injection)
   • Monthly payment projection (25yr @ current prime + 2.75%)

3. Post-AI transformation model

   • Estimate labor that AI replaces (target 60%+)
   • Project post-AI EBITDA (target 75-80%)
   • Calculate exit value at 7x EBITDA

4. Sensitivity analysis

   • Best case / base case / worst case on client retention
   • Break-even on client loss (how many can you lose and still profit?)

Step 4: Red flag analysis (/investigate)

Run /investigate on any anomalies found:

Automatic red flags (investigate immediately):

• Revenue claims inconsistent with market presence
• Owner age < 50 (not motivated to sell?)
• Client concentration > 40% in top 3 clients
• Active litigation or tax liens
• Asking price > 0.6x revenue
• No employee CPA license holders

Yellow flags (note but don't block):

• Limited online presence
• Mixed reviews (< 4.0 stars)
• Rural location
• Owner wants extended transition (> 12 months)

Step 5: DD scorecard

Generate a structured scorecard:

DD SCORECARD: [LISTING_ID]
================================
Target:     [Firm name]
Revenue:    $[amount]
Ask Price:  $[amount] ([multiple]x)
HENRY Buy:  $[0.4x amount]

SECTION                    SCORE   STATUS
─────────────────────────────────────────
License & registration     /10     [PASS|FLAG|FAIL]
Financial verification     /10     [PASS|FLAG|FAIL]
Client base health         /10     [PASS|FLAG|FAIL]
Owner motivation           /10     [PASS|FLAG|FAIL]
AI transformation fit      /10     [PASS|FLAG|FAIL]
Legal / litigation         /10     [PASS|FLAG|FAIL]
Market position            /10     [PASS|FLAG|FAIL]
SBA eligibility            /10     [PASS|FLAG|FAIL]
─────────────────────────────────────────
TOTAL                      /80     [PROCEED|CAUTION|WALK]

PROCEED: 60+ | CAUTION: 40-59 | WALK: <40

Red flags:    [list]
Yellow flags: [list]
Next actions: [list]

Step 6: Route results

• Save scorecard to dark-factory/DD_[LISTING_ID].md
• Update memory/RESEARCH_MEMORY.md with findings
• If PROCEED: notify SHIELD to draft LOI
• If CAUTION: notify ATLAS for strategic decision
• If WALK: log reason and archive

Autoresearch Loop

This skill improves by tracking dd_checklist_completion:

• Metric: % of DD scorecard sections completed with verified data (not assumptions)
• Budget: 120s per eval
• Modify: research sources, scoring weights, red flag criteria
• Guard: false flag rate stays below 10% (don't flag clean deals incorrectly)

MCP Integration Map

ORACLE (Opus 4.6)
  ├── paperclip: fetch TX State Board, SOS, APS.net, reviews, LinkedIn, court records
  ├── aegis: route quant analysis to DeepSeek R1
  ├── github: store DD reports in dark-factory/
  └── /investigate: systematic red flag root-cause analysis

Active Deal Queue

Target	Revenue	HENRY Buy	Status	DD Score
TXS5345	$142K	$56.8K	LOI ready	Pending
TXS5513	$424K	$127-170K	Awaiting package	Pending
TXS5491	$910K	$364K	Sourcing	Pending
TXS5450	$472K	~$189K	Research	Pending