Audit Process Questions That Separate Structured Debt Discovery From Sales Theater
A technical debt assessment before hiring an engineering partner should connect findings to delivery delays, incidents, repair effort, and risk, then leave the client with a written, ranked repair queue. The questions below separate that work from sales theater by asking vendors to show their method, explain what they would leave alone, and define what the client will own afterward.
Why the audit question exists in the first place
Technical debt is deferred engineering work that buys speed now at the cost of slower or riskier change later. Ward Cunningham’s borrowing metaphor shows why taking a shortcut on purpose can make sense for a business, as long as someone tracks the interest. An audit is supposed to make that cost visible. Yet the same “assessment” label can describe a polished presentation that names a prospect’s pain and proposes a fixed scope without examining the codebase, delivery records, or team behavior. The format alone proves nothing, especially when the person who offers the diagnosis gets paid for the cure they propose.
What should a technical assessment cover before signing with a software development partner?
Start with evidence that exists before the engagement. Lead time for a typical change shows whether delivery is getting slower, while change failure rate and recovery time help show where testing and architecture debt lie. The share of engineering time spent on incidents and unplanned work shows how much capacity goes to firefighting. These are business costs. Dependency age and known vulnerabilities bring a different kind of pressure because components past end-of-life set deadlines that untidy code may not have.
Ownership concentration belongs in this baseline as well. If only one engineer understands a system, the time another engineer needs to make a safe change is a measurable cost. That context changes how code metrics should be read. High test coverage in a module nobody changes says little about the expense of modifying a frequently touched area with no tests. A useful assessment therefore covers code, architecture, infrastructure, testing, documentation, and process debt.
To build that baseline, pull at least six months of delivery records before a vendor arrives. Include cycle time, change failure rate, incident tickets by system, support escalations, and changes estimated as small that finished late. For AI-built or rapidly shipped systems, also ask whether code was AI-generated and by whom, without treating that answer as proof of a defect. Hard-to-read logic, duplicated functions, and high complexity still need to be assessed against how the system is changed and operated.
The specific questions that reveal whether a vendor has done discovery or prepared a deck
“What does your discovery process look like before any code is written?” A credible answer describes a sequence, beginning with the business, then explaining what the vendor examines, what the team is asked, and which records inform the assessment. Theater offers a generic process slide or asks for requirements so a quote can follow.
“Can the vendor show the audit template, threat-model document, secrets-rotation runbook, and phased proposal it uses?” A credible answer includes actual documents, even if they are redacted, that demonstrate a repeatable method. Theater promises those artifacts after the engagement starts or substitutes a scope document that jumps straight to price and timeline. The documents make the method open to inspection. Without them, a founder has only a description of a process.
“Which parts of a system do you typically avoid recommending changes to, and why?” A credible answer distinguishes expensive debt from ugly-but-harmless code and explains why stable, rarely touched, cheap-to-run areas may be left alone. Theater treats every defect as urgent and every repair as billable. Not every shortcut deserves repayment right away. The useful distinction is between high-interest debt in code the team actively changes and debt in areas nobody has touched in a year.
“How do you cost a finding before ranking it?” A credible answer connects the finding to slower changes, repeated incidents, or blocked work, then considers whether that cost is growing, how much repair would take, and what risks are involved. Theater offers a ranked list without its rationale or compresses the assessment into one composite debt score. Ranking cannot substitute for costing the individual findings. Otherwise, the loudest request for a refactor can outrank the work that consumes the most engineering time.
“What would cause you to pause or narrow the scope mid-engagement?” A credible answer names conditions such as a dependency finding that changes the risk picture, a security issue that reorders priorities, or a team constraint that makes a proposed fix premature. Theater offers no pause conditions or only a vague promise to adapt. Discovery must be allowed to change the plan. A vendor who can explain when to do less is showing how evidence constrains the work being sold.
“Who owns what after the engagement ends, and what does the written report include?” A credible answer describes the deliverable, its recipients, and how another engineer could act on its findings without calling the vendor. Theater leaves the output undefined or withholds the report until a follow-on engagement is signed. The handoff is part of the assessment itself. The client needs to be able to inspect findings, assign them, or take them to another team to make a sound choice about what happens next.
The internal questions to ask your own team before any vendor arrives
Ask engineers which areas they avoid touching. Ask what would happen if one had to change this week. Then ask which deployments make them nervous and what, specifically, concerns them. The answers give a vendor something concrete to investigate rather than a general complaint that the codebase is difficult.
What would an engineer fix with two weeks free of feature demands if nobody would know how the time was spent? Follow that with a question about the last change that took far longer than estimated. Ask for the actual reason it slipped. The two answers serve different purposes. One reveals work engineers want room to do, while the other anchors the conversation in a delivery failure that already happened.
What the team does adds evidence that a code review can miss. Engineers may keep spreadsheets to bridge gaps between systems or copy data by hand between tools. Sprint planning may route work around fragile code rather than through it, while a security review or investor diligence process may expose questions the team cannot answer confidently. These are useful clues. Features stretching from one sprint into multiple cycles, or deployments stretching from hours into days, give the audit a delivery problem to explain.
That distinction is especially useful when an MVP moves toward production. Much of its debt may come from reasonable launch trade-offs rather than poor engineering, with the problem being that nobody recorded or tracked them afterward. A working screen is not enough evidence. The assessment needs to consider missing authorization behind a login, exposed API keys in client-side code, duplicated logic and unclear architecture, uncontrolled dependencies, and missing test coverage. In particular, a functioning login screen does not establish that access control is secure.
The red flags that appear in the billing model and the contract, not just the pitch
Even when the answers are convincing, the contract needs to back them up. Hourly-only billing with no fixed scope is a warning sign because it leaves the client with open-ended invoices and no clear boundary for completion. A low estimate at the start can still lead to a higher cost in the end. A credible plan explains what is included, what is excluded, which assumptions drive the price, and how scope changes will be handled.
A request to host the code in vendor-controlled accounts “for security” creates lock-in. Insist on client-owned accounts and repositories. The agreement should also require a written audit report at the end that the client can get without having to buy remediation. The client needs that report to check another vendor’s work against the original assessment.
The roadmap reveals whether discovery has a real checkpoint. A sound sequence moves from discovery to a baseline audit and priority ranking, then to a pilot or proof of concept small enough to finish quickly but detailed enough to expose risk. A direct jump to full implementation skips that check. It also removes the opportunity to reveal errors in the proposed scope before the larger commitment. A phased proposal is therefore more than a convenient way to divide invoices.
The same standard applies if the vendor presents a scorecard. Each item needs evidence rather than a promise, supported by examples, workflow diagrams, service-level expectations, migration steps, and a realistic timeline before work is approved. A score without support adds little protection. By contrast, a partner who can describe the first work, the areas left untouched, the pause conditions, and the first months in concrete terms makes the engagement inspectable. That is what a good commercial structure preserves after the sales conversation ends.
What a founder should have in hand before and after a credible audit
Before the conversation, assemble the delivery records and internal answers into a baseline the vendor must either match or explain. Bring a defined outcome as well, whether that is a business process that must improve, a risk that must fall, or a metric that will demonstrate progress. Otherwise, the engagement can settle on a technical solution before establishing which business problem it is meant to solve.
Afterward, the report should list what was found across the six debt categories and include dependency findings so they do not get lost between code and infrastructure. Each finding needs to show what it costs now, whether that cost is growing, how much work the repair is likely to take, and a note on risk. This is where a scored register becomes useful. Its entries retain the evidence and costing rationale behind the ranking, so a different engineer can understand why one repair precedes another.
The repair queue should lead into a phased roadmap whose first phase is small and quick enough to test the proposed approach against real risk. Each finding also needs a named owner after the engagement and a reporting cadence for what happens next. Responsibility cannot end at delivery of the report. The resulting document should let leadership decide what to fund, what to defer, and who will be accountable for the next step without relying on the vendor to reconstruct the reasoning.
At that handoff, a one-time audit remains a point-in-time assessment, while the more durable arrangement includes remediation and continued monitoring of delivery signals. Regular audits can help reduce debt systematically as business priorities change. Launching the product and keeping it reliable in production require more than the same document.