CLEAR SIGHT DESIGNS · TECHNICAL DUE DILIGENCE PRACTICE
Principal Systems Architect: Preston McCauley
5-DAY SENIOR-LED SPRINT · 48-HOUR WRAPPER VERDICT · FIXED $35,000

Every target claims proprietary AI.
I tell you whether it's real before you sign.

A five-day technical diligence sprint, run personally by a 25-year systems architect who has built the kind of AI infrastructure your targets claim to own. You get a preliminary wrapper-risk verdict in 48 hours and a decision-ready IC memo before your committee meets — with every number traceable to a document in the data room.

See a Full Sample IC Memo ↓
THE DILIGENCE RISK GAP IN TECH ACQUISITIONS
80%
of dealmakers report AI-related security, accuracy, or governance issues during transaction audits.
INCLUDING
40%
where hallucinated outputs or unverified wrapper dependencies produced inaccurate diligence findings.
Source: KPMG & Dealmaker Transaction Technology Audits (2026).
When technical diligence relies on surveys and checklists, deal teams underwrite phantom software moats. I audit the actual codebase, dependency graph, and cloud telemetry.
THE PERSON DOING THE WORK

The satisfaction gap exists because AI diligence is being run by people who have never built one of these systems.

I've spent 25 years building them.

I'm Preston McCauley. I'm a systems architect, not a diligence-firm partner who assembles a team. As the creator of Archaiforge (geometric AI architecture), PHYSIM (biopharma.clearsightdesigns.com), and the CLEAR AI Method™ & Governance Frameworks, I've architected biomedical ontology pipelines, high-throughput deterministic data engines, adaptive workflow orchestration, and enterprise compliance systems — the same categories of infrastructure middle-market targets describe in their pitch decks as "proprietary AI."

That matters for a specific reason. When a target's CTO explains why their fine-tuned model is defensible, the useful question isn't on a checklist. It's the follow-up: what's the actual delta between this and a prompt against a frontier model with good retrieval? Someone who has shipped the architecture asks that question in the first ten minutes. Someone working from a template never asks it at all.

I built the CLEAR AI Method™ and the ISD Framework because standard technical diligence answers "does the target own this code?" and stops. Private equity needs the next question answered: does owning it create an economic moat, and what breaks if it's removed?

CORE PRACTITIONER CREDENTIALS
  • Creator: Archaiforge & PHYSIM (Deterministic Geometric AI & Computational Biology Engines)
  • Creator: CLEAR AI Method™ & Enterprise AI Governance Frameworks
  • Recognition: Nominee, Top 75 AI Innovator (Dallas Innovates / North Texas AI Ecosystem)
  • Author: Generative AI for Everyone: Enterprise Systems & Applied Engineering (2024)
  • 25+ Years Experience: Enterprise systems architecture, deterministic data engines, and biomedical platforms
  • Direct Senior Execution: Every engagement conducted 100% directly by Preston McCauley — zero junior handoffs
LinkedIn Profile ↗ Archaiforge.com ↗ PHYSIM Biopharma ↗
Preston McCauley — Principal Systems Architect
Preston McCauley
Principal Systems Architect & Diligence Lead
Clear Sight Designs
ARCHITECTURAL BRIEFING

How to Spot a Commodity API Wrapper in 10 Minutes

A senior architect's walkthrough of wrapper failure patterns, un-cached token burn, and silent model disintermediation.

THE UNDERWRITING QUESTION

Traditional tech diligence tells you the software runs. That was never the risk.

An AI-enabled target can pass every conventional technical check — clean architecture, good test coverage, no license contamination, SOC 2 in hand — and still be worth two turns less than the ask. Five things determine that, and none of them appear on a standard tech DD checklist.

35% WEIGHT CORE MOAT

Data Gravity

Does the target own data a competitor cannot buy, license, or scrape? Proprietary corpus volume, customer contract rights to train on it, and whether the feedback loop compounds or just accumulates. This is the heaviest weight because it's the only moat that survives the next model release.

Stress Test: Zero-shot frontier model disintermediation.
25% WEIGHT WRAPPER RISK

Wrapper Immunity

What percentage of the product's actual logic is a call to someone else's API? I audit the call ratio against the codebase, not the architecture diagram. A target where a foundation-model provider can ship the core feature for free in one release is a different asset than the one in the CIM.

Stress Test: API deprecation & native OS-level feature release.
20% WEIGHT GOVERNANCE

Deterministic Governance

What happens when the model is wrong? I look for hard runtime gates, human verification checkpoints, and override telemetry — not eval scores. In regulated workflows, an ungoverned failure surface is a liability line item, not a technical footnote.

Stress Test: Adversarial injection & silent hallucination liability.
10% WEIGHT UNIT ECONOMICS

Unit Economics Under Load

I model inference cost per transaction at 5x current volume. Industry benchmarks put AI-native gross margins near 52% against 80–90% for traditional SaaS, and ICONIQ's 2026 survey found inference cost rising as a share of spend — from 20% to 23% — as products mature. Margin compression at scale is the most commonly missed adjustment in AI-target underwriting.

Stress Test: 5x volume spike & token pricing shifts.
10% WEIGHT COMPLIANCE & IP

IP & Compliance Integrity

Copyleft contamination in the model-serving path, incomplete invention assignments, customer DPAs that block post-close training, and sector mandates (CMS-0057-F, FDA Part 11, FINRA 3110) the target has to meet on a clock.

Stress Test: Open-source audit & regulatory enforcement horizon.
Each dimension is scored against evidence in the data room and stress-tested against a specific failure scenario. The composite maps to a moat tier, and the tier maps to a defensible position on the entry multiple. In the engagements I've run, most targets claiming proprietary AI have been thinner than represented.
FIVE DAYS · PARALLEL TO FINANCIAL AND LEGAL WORKSTREAMS

No drag on your timeline.

We work alongside your quality of earnings and legal teams during exclusivity. Every finding is established before your Investment Committee convenes.

DAYS 1–2

Ingestion & Architecture Read

You grant VDR access, or I work from technical interview transcripts and cloud invoices where code access is restricted. I hash every source document to its exact VDR coordinate so any associate can pull the underlying file in one click and check my work. I read the model inventory, dependency manifests, and 12 months of itemized cloud billing.

END OF DAY 2

Preliminary Wrapper Verdict

A one-page readout: wrapper risk, unit-economics red flags, and whether anything I've seen threatens the thesis. Enough for your deal team to adjust posture before the sprint finishes. If the answer is going to be bad, you find out on day two — not day five.

DAYS 3–4

Stress Testing & Labor Reconciliation

I stress the five dimensions against specific failure scenarios, model unit economics at 5x volume, and reconcile operational case volumes and handle times against burdened GL payroll to size realistic post-close automation capacity.

DAY 5

Decision-Ready IC Deliverable

An 8-page Investment Committee memo tying every technical finding to the entry multiple, the downside case, and a Day 1–100 roadmap. Backed by a full technical appendix for whoever inherits the asset.

VDR INTAKE SPECIFICATION

What I Need From the Data Room on Day 1

The five-day sprint clock begins upon receiving data room access. Deal teams can request these six artifact categories directly from target management or the sell-side advisor to ensure an immediate Day 1 kickoff:

01
Architecture Diagrams & Repository Manifest

System topology, data flow schemas, and codebase access (read-only repo access preferred; zip export or SBOM accepted).

02
12 Months Itemized Cloud Billing

AWS, Azure, or GCP invoice line items showing GPU instance hours, database storage, and external API token spend.

03
Model Inventory & Eval Telemetry

Base models used, fine-tuning scripts, training data sources, loss curve history, and error/exception telemetry.

04
Customer Contracts & DPAs (Sample of 5–10)

Customer Master Services Agreements and Data Processing Agreements to verify training rights and data isolation clauses.

05
Engineering Org Chart & Payroll Extract

R&D and engineering org structure, key-person dependencies, 12-month attrition, and departmental payroll for GL labor reconciliation.

06
CIM Claims & Product Roadmaps

Target's pitch deck, Confidential Information Memorandum (CIM), and product roadmap to establish the baseline of represented capabilities.

Restricted access or incomplete room? The clock still starts. The 48-hour preliminary readout explicitly flags what is missing, how it constrains the findings, and where confidence intervals adjust.
EVIDENCE LEDGER

Every number in the memo resolves to a document you can open.

This is the part standard tech diligence skips, and the reason IC members discount consultant scores. A qualitative rating you can't trace is an opinion with a decimal point on it.

Every score, multiple defense, and EBITDA projection in my memo carries an evidence node ID. Each node names its source family, the exact VDR folder, the substantiating data, and a confidence level. Claims resolve backward: an EBITDA figure traces to feasible automation capacity, which traces to reconciled burdened labor, which traces to case-volume telemetry, which traces to named documents.

Ten source families feed the ledger: financial and QoE, workflow telemetry, commercial cohorts, customer references, transaction comps, model evaluation harnesses, data-rights and IP audits, SBOM and supply chain, security and SLA, and sector regulatory.

Node IDSource Family & VDR LocationSubstantiating Empirical EvidenceConfidence
E-0112Departmental payroll & GL extract · VDR 02.0118 prior-auth FTEs ($2.06M burdened) + 14 appeals FTEs ($1.60M)98%
E-0119Production event logs · target data lake146,200 annual submissions across 42 payer portals; 7.8 min median handling time; 11.2% exception rate94%
E-0220PitchBook / CapIQ comps 2024–2026Median 13.4x for proprietary RCM platforms >15% margin and >95% GRR; services trade 7.5–9.0x94%
If a partner disputes a conclusion, the argument happens over the source document, not over my judgment. That's the point.
ILLUSTRATIVE SAMPLE · SYNTHETIC TARGET

Read a complete IC memo before you hire me.

Project Beacon is a modeled healthcare RCM platform built on synthetic data. Every figure is illustrative. It exists so you can judge the format, the depth, and the reasoning before spending $35,000 — which is a reasonable thing to want and a strange thing for most diligence firms to withhold.

CLEAR Composite Score:
4.76 / 5.00
Tier 1: Proprietary Moat
Recommended Entry:
13.0x EBITDA
Full Multiple Justified
Modeled Post-Close EBITDA:
+$2.70M / yr
+770 bps Margin Expansion

What the 8-Page IC Deliverable Contains:

  • ✓ Executive IC Determination: Clear thesis & entry multiple recommendation
  • ✓ 5-Dimension Grounded Matrix: Data gravity, wrapper immunity & stress tests
  • ✓ Proprietary Asset Register: ISD structural failure impacts & replacement burden
  • ✓ GL-Reconciled EBITDA Bridge: Reconciled operational labor & runtime SLM caching
  • ✓ Exit EV Sensitivity: Modeled against entry basis across compression, base case (13.0x), and expansion exit scenarios
  • ✓ Unabridged Evidence Ledger: Traceable to exact data room files
Open the Full Sample Memo (8-Page IC Deliverable) ↗ View Full Technical Appendix & Architecture Dossier ↗
WRITING & ANALYSIS

How I think about AI targets.

Technical diligence in private equity moves by referral inside deal teams. Here is the operational thinking behind how I evaluate proprietary software moats and unit economics.

FEATURED ESSAY · 4 MIN READ

Why 78% Gross Margins on an API-Dependent AI Target Are a Red Flag

When a target claims 78%+ gross margins while calling third-party frontier APIs at scale, true COGS are being masked by un-cached token loops or promotional cloud credits that expire post-close. In production, third-party API token costs push true AI gross margins down to 40–50%. The only exception—and the definition of a Tier 1 moat—is when inference runs on self-hosted, quantized SLMs inside a private VPC with zero per-token third-party spend (as demonstrated in our Project Beacon benchmark). If a target lacks dedicated local inference infrastructure and still claims SaaS-like margins, the AI is usually a batch wrapper or an unsustainable demo.

By Preston McCauley · Principal Systems Architect
MODEL DEFENSIBILITY · 5 MIN READ

The Five Questions That Separate a Fine-Tuned Model From a Prompt

The difference between a defensible asset and an unmaintainable wrapper isn't the model weights — it's data provenance, loss-surface stability, regression governance, and the human override feedback loop.

By Preston McCauley
VDR PATTERNS · 6 MIN READ

What I Found in the Last 10 Data Rooms

Anonymized patterns from recent middle-market audits: from un-cached API loops burning $40k/mo to missing training rights in customer DPAs that block post-close fine-tuning.

By Preston McCauley
FIXED FEE · SENIOR-LED · NO JUNIOR STAFFING

The $35,000 Diligence Sprint.

A five-day fixed-fee technical diligence engagement delivering an 8-page Investment Committee decision memo and complete architecture appendix before your committee convenes.

SENIOR-LED DILIGENCE SPRINT · 5-DAY EXCLUSIVITY
$35,000 Fixed Flat Fee · Single All-Inclusive Deliverable

Technical & AI Diligence Sprint (Buy-Side or Sell-Side VDD)

Conducted during exclusivity or pre-auction to evaluate wrapper risk, substantiate data gravity, and underwrite the entry multiple.

  • ✓ 48-Hour Preliminary Readout: Early wrapper and margin red-flag verdict
  • ✓ 5-Dimension Grounded Audit: Data gravity, wrapper ratio, and failure stress tests
  • ✓ GL-Reconciled EBITDA Bridge: Burdened payroll reconciliation and automation sizing
  • ✓ Decision-Ready Deliverable: 8-page IC Memo + Full Technical Appendix
  • ✓ Direct IC Support: Senior architect defense call with your deal team and committee
POST-CLOSE & CUSTOM PORTFOLIO SCOPE
Custom Scope Tailored to Value Creation Thesis

Post-Close Value Acceleration & Technical Advisory

For sponsors and operating partners seeking ongoing technical governance, local model caching, or portfolio-wide architectural consolidation.

  • ✓ 100-Day Value Sprint: Execution oversight for identified post-close EBITDA initiatives
  • ✓ Runtime Model Caching: Private SLM deployment to eliminate external API token burn
  • ✓ AI Governance Auditing: Periodic regulatory and runtime failure reviews
  • ✓ Multi-Asset Due Diligence: Add-on acquisitions, platform roll-ups, and corporate carve-outs
For reference: big-firm technology diligence runs $75,000–$200,000 over six weeks and produces a report written largely by associates. Published boutique tech DD pricing sits at $25,000–$75,000. I price at $35,000 flat and staff it with one senior person — me — from kickoff to IC.
WHAT THIS ENGAGEMENT IS AND ISN'T

What I review, what I don't, and what would change my conclusion.

Every institutional diligence report carries a strict scope and limitations framework. Here is the operational boundary of the technical sprint:

✓ IN SCOPE
  • Architecture and model inventory
  • Dependency and software license audit (OSS & copyleft)
  • Unit economics evaluated against 12 months of itemized cloud billing
  • Data provenance and contractual training rights
  • Model evaluation methodology and error telemetry
  • Operational workflow and labor feasibility reconciled against GL payroll
  • IP assignment completeness
  • Sector regulatory alignment (CMS-0057-F, FDA Part 11, FINRA 3110)
✕ NOT IN SCOPE (UNLESS SEPARATELY ENGAGED)
  • Penetration testing & dynamic vulnerability exploits
  • Full source-code security review
  • Financial audit or Quality of Earnings (QoE)
  • Legal opinion on formal IP title
  • Customer reference calls (I interpret transcripts; I don't field calls)

What I Rely on Management For:

Representations about undisclosed dependencies, roadmap commitments, and pending contract changes. Where a claim can't be substantiated in the data room, the memo says so and the confidence level drops.

What Would Change the Conclusion:

Material undisclosed third-party model dependency, customer contracts prohibiting post-close training, key-person departure during exclusivity, or telemetry that contradicts represented volumes.

ENGAGEMENT PATTERNS

What these findings look like on real deals.

A demonstration of how deterministic technical auditing converts into transaction leverage and operating clarity.

FRAMEWORK STRESS-TEST SCENARIOS (4 MODELED ARCHETYPES)
ArchetypeAudited SectorCLEAR ScoreMoat ClassificationDiligence Finding & Multiple ImpactAction
Project BeaconHealthcare RCM4.76 / 5.00Tier 1: Proprietary MoatFull multiple supported (13.0x entry); +$2.70M modeled EBITDA expansion. Runs self-hosted domain SLMs on dedicated private instances with zero external API per-token spend, validating 78% -> 84% gross margins under 5x volume.
Project AtlasLogistics TMS3.15 / 5.00Tier 3: Commodity Wrapper1.0x haircut modeled; 82% of core workflow relied on un-cached public prompt chaining with zero proprietary data gravity.
Project ClearwaterPort Demurrage4.08 / 5.00Tier 2: Niche SpecialistReal workflow depth in container dispute logic, narrow TAM; partial multiple support with targeted post-close caching.
Project CobaltIndustrial Quoting2.57 / 5.00Tier 3: TurnaroundRebuild required; brittle prompt chains failed on non-standard CAD files. Modeled as operational remediation case.
These are modeling scenarios built on synthetic data to validate the scoring framework. They are not client engagements.
DOWNSIDE EXPOSURE MODEL

What a missed wrapper costs.

This models exposure, not a promised outcome. Enter your target's EBITDA, entry multiple, and the haircut you'd apply if the AI turns out to be thinner than represented. If the target is genuinely Tier 1, the haircut is zero — and the value of the engagement is the conviction to bid at full multiple in a competitive process, not a price reduction.

Target Annual EBITDA: $5.0 Million
Target Entry Multiple: 10.0x EBITDA
Baseline Target Valuation: $50.00M
Vulnerability Scenario (If Commodity Wrapper / Thin Moat): 1.0x EBITDA Haircut
0.0x = Tier 1 Moat · 0.5x = Tier 2 Niche · 1.0x–1.5x = Tier 3 Wrapper Risk
Modeled Valuation Adjustment Exposure
-$5.00M
Adjusts purchase price from $50.00M down to $45.00M (9.0x)
Diligence Fee as % of Valuation Risk
0.70%
Fixed $35,000 sprint fee vs. modeled downside risk exposure
Modeled Post-Close EBITDA Capacity Potential
+$0.90M / yr
From workflow labor capacity recovery & runtime SLM model caching (+18% lift)
FREQUENTLY ASKED QUESTIONS

Straight answers for Investment Committees.

Can you really do this in five days?

The sprint is five working days from VDR access, not from engagement letter. It works because one senior person does all of it — there's no team to brief, no findings to reconcile across workstreams, no associate draft to rewrite. The constraint is data-room completeness, not my calendar. If the room is thin, I'll tell you on day one what's missing.

What if code access is restricted?

Common in competitive auctions pre-exclusivity. I run a structured 45-minute CTO and architecture session against cloud invoice telemetry and the model inventory. That yields directional clarity on wrapper risk and unit economics without deal friction — less certainty than full access, and the memo says exactly where the confidence gaps are.

How is this different from Crosslake or West Monroe?

They do broad technology diligence well and staff it with teams. I do one thing: whether the AI is real and what it's worth. If you need infrastructure, org, and security assessed at breadth, hire them. If you need someone who can tell whether a fine-tuned model is defensible or theater, that's a narrower question and a different kind of person.

Do you have a conflict of interest with targets or software vendors?

No. Clear Sight Designs is 100% independent. I have zero commercial relationships, reseller agreements, or revenue-sharing arrangements with target companies, foundation model vendors, or third-party diligence platforms. My sole fiduciary client is the private equity sponsor.

How do you use AI in your diligence process?

The sprint combines deep human systems auditing with purpose-built deterministic evaluation engines. I use proprietary code AST parsers, market dependency mappers, and telemetry analyzers—techniques I've developed over 25 years in systems engineering—to extract and stress-test target claims. The output is not probabilistic text or auto-generated summaries; it is a deterministic, auditable evidence ledger where every risk rating, multiple defense, and EBITDA reconciliation is backed by senior human judgment and verifiable data room provenance.

Who's on the engagement?

Me. There is no team, no junior staffing, no handoff. That's the constraint and the product.

Have a deal under exclusivity?

Tell me the sector, the rough size, and where you are in the process. If it's not a fit, I'll say so on the call rather than sell you a sprint you don't need.

Read the Sample IC Memo ↗