← Back to the blogEvolutions

The impact report: know what breaks before you write a line

Features, rules, screens and data touched, conflicts, tests, the decision to make: how an impact report moves the decision before the code.

October 7, 2026 · 4 min read

Most teams discover the impact of a change after it ships. A support ticket, a broken screen two modules away, a meeting about who should have known. An impact report moves that moment to before the first line of code.

Late decisions, expensive meetings

The pattern is familiar. The business asks for something that looks simple. The team estimates it, builds it, ships it. Then a side effect turns up somewhere else, and a dozen people spend an afternoon reconstructing what happened.

Usually nobody did bad work. The information existed, scattered across the code, a few Jira tickets, an old Confluence page and the memory of two colleagues. It just wasn’t anywhere you could read it in one go.

AI-assisted coding raises the stakes. Faros AI’s The AI Productivity Paradox report (July 2025, more than 10,000 developers across 1,255 teams) found that teams with high AI adoption merge 98% more pull requests, with review time up 91%. Output speeds up. The ability to foresee consequences doesn’t, so decisions slide later, often all the way into production. And every surprise makes the next change harder to approve.

A request that looks like one number

“Shorten the refund window from 30 days to 14.” On paper, that’s a find-and-replace.

Lyriks places the request on the model of your software, which already knows where the rule lives. The spec says “a customer is refunded within 30 days.” The same spec links that rule to refund.py, the Payments API and the refunds table. From there, the report comes in five parts.

Part 1, what’s touched

Features, business rules, screens and data. For the refund change, that means the refund rule itself, the customer-facing request tracker, the support dashboard, the refunds table and the API calls that check the deadline. Some of these are obvious. The API checks usually aren’t, because they sit in a service nobody associates with refunds. Each item links to its source, so anyone can click from a rule to the code, ticket or page behind it.

Part 2, conflicts

These can be with the existing system or with other changes running in parallel. Here the report surfaces something more interesting than a number. The knowledge base, which brings together code, Jira, Confluence, Notion, Figma and GitHub, shows the window is 14 days in Jira, 30 in Confluence and 21 in one branch of the code. So the request was never really about a number. The rule itself had never been settled.

Part 3, what’s not touched

People tend to skip this part, and shouldn’t. Knowing what stays put lets QA focus, reassures neighboring teams, and avoids freezing everyone for a change that turns out to be local. Here, for instance, invoicing and order history don’t move, and the teams that own them can be told there is nothing to do this time.

Part 4, what to test

The rules and flows QA should cover. Requests made on day 13, 14 and 15. Requests opened before the change and still pending after it. The countdown shown on the support dashboard. Each test points back to the rule it covers, so once the decision lands on one value, QA knows exactly which cases change.

Part 5, the decision

Stated plainly, for the person who owns it. Which value is the source of truth, 14, 21 or 30 days, and does it apply to orders placed before the change? That call belongs to the business. It gets made once, documented, and linked to its source.

CHANGERefund window30 days to 14TOUCHEDrefund.py, Payments API, refunds tableINCONSISTENTJira 14 days, Confluence 30, code 21SCREENSRequest tracker, support dashboardDECISIONWhich value is the source of truth?
An impact report for a small change: what it touches, where the sources disagree, which screens are involved and the decision to make. Example taken from the article.

Readers around the table

Several people read the same report, each for a different reason. The product lead reads the features and trade-offs. The engineering lead reads modules, data and conflicts. QA reads what to test and what to leave alone. Leadership reads the decision without having to decode a ticket.

Meetings change as a result. You stop debating what might happen and decide from a shared baseline, with clear owners. The business also gets an honest view of cost. A change touching one rule and one screen reads very differently from one that crosses several features and collides with work in progress, and that difference is visible before anyone commits to a date.

Stability matters too. The engine is deterministic, built on graph rewriting and developed in-house. Ask the same question twice and you get the same report. You can rerun it after the decision and compare the two versions line by line, which is hard to do with a document that shifts every time you generate it.

ONE IMPACT REPORTRefund windowPRODUCT LEADFeatures and trade-offsENGINEERING LEADModules, data, conflictsQAWhat to test or leaveLEADERSHIPThe decision: 14, 21 or 30one shared baseline, the same report on every run
One impact report on the refund window change, read by four people for four reasons: product, engineering, QA and leadership.

After the decision

The validated spec goes to your coding agent over MCP, with its rules and their sources. Lyriks checks the generated code rule by rule, and gaps go back to the agent until every rule holds. At each release, a compliance report says what is compliant, what was added without being requested, what was left out and what regressed. The impact report opens the loop and the compliance report closes it. The Evolve your software page covers the full approach.

To start, pick one or two modules where changes cost you the most meetings, and run a first report on your own code in a demo.

Keep your software in tune

Start with one module. Lyriks recovers its spec, then verifies every change. All it needs is your repository.