Every module rewritten before a certification signature lands at least doubles the true cost of that project. Most teams building aviation software learn this not before the project ends, but well after it does. In the defence and aviation ecosystem around Ankara — the engineering firms supplying companies such as TUSAŞ, Roketsan and Aselsan — this pattern has become increasingly familiar over the past few years. Teams that arrive from commercial software backgrounds produce working products, then find themselves close to starting over when certification review begins. The problem is not technical. It is a fundamental misreading of what aviation software actually means.The governing rule in aviation software is this: working is not enough; demonstrable reliability is required. The DO-178C standard — which replaced DO-178B at the start of 2012 and is now referenced by both EASA and FAA — does not regulate how software behaves. It regulates how that behaviour is verified. The framework divides software into five criticality levels, from Level A down to Level E, and each level demands a different volume of evidence, a different depth of testing, and a different degree of independence. For a Level A module — one that directly affects flight safety — every line of code must be traceable to a documented requirement, and that relationship must be verifiable in both directions. Doing this retroactively, meaning writing the code first and then constructing the requirements to fit, is not a technical failure. It is an audit failure. A certification examiner does not read the code; the examiner follows the traceability chain from requirement to implementation and back again.For mid-sized software companies entering this space in Turkey, the sharpest breaking point is change control. In commercial software development, a developer who finds a defect in the morning fixes it by afternoon; no change log is kept, no impact analysis is performed. In aviation, the same correction requires a formal change request, a reassessment of affected requirements and test scenarios, and traversal of an approval chain. A representative case from Ankara illustrates this clearly: a defence software firm with 280 engineers delivered its first major aviation project, then spent eight weeks revisiting fourteen months of development work because of documentation gaps identified during certification preparation. What was lost was not time alone — it was credibility. The company waited two additional years before it was considered again as a sub-supplier. That cost does not appear in any project budget column.Which teams manage this pressure more effectively? Field observation points consistently to one differentiator: teams that treat certification not as a phase at the end but as a responsibility distributed across every step of the development cycle. In practice this means: every requirement carries an identifier, every test scenario links to that identifier, and every code change records which requirements it touches. No specialist tooling is required to make this work; the discipline is what changes, not the technology. An anonymous aviation maintenance engineering firm in Eskişehir built this structure using a standard requirements management tool already in the organisation. The early sprint cycles ran roughly longer than before — a real cost. By the second half of the project, certification preparation added close to zero additional time. The acceleration came not from better technology but from process maturity that compounded over the project’s lifetime.Test evidence is consistently the least understood part of this equation. In commercial software, code coverage is a quality metric; reaching a high line-coverage figure is treated as sufficient. Under DO-178C Level A, the standard is different: Modified Condition/Decision Coverage — MC/DC — is mandatory. This means demonstrating independently that each condition within a decision can and does affect the outcome on its own. That is not produced by running a tool and reading the report. It is produced by deliberately designed test matrices built from the start with this coverage requirement in mind. Most engineering teams in Turkey know this requirement theoretically. In practice they attempt to retrofit test scenarios after the fact. MC/DC coverage achieved through after-the-fact mapping is rarely clean, and a certification examiner reviewing the test design record can tell the difference. The gap is not between a test that passes and a test that fails. It is between a test that passes and a test that proves something specific about a specific decision path — with a documented record linking back to the original requirement.For managers and programme directors making build-or-partner decisions in this space, three practical criteria clarify the risk before contract signature. First: does the team have a working requirements management tool, and does every requirement in the current baseline carry a live traceability record — not a plan to create one, but an active record? Second: has the change control process been exercised on a live project, under real schedule pressure, or does it exist only as a written procedure? Third: who performs independent verification and validation — is there a genuine structural separation from the development team, or is the same engineer reviewing their own work under a different label? If none of these three questions has a direct, evidence-backed answer, working with that team will produce surprises during certification review. Those surprises arrive late, cost more than early discipline would have, and occasionally cost the programme itself. Aviation software development is not about delivering an application. It is about being able to show a qualified examiner, in under two hours, the unbroken chain from original safety requirement to verified implementation. Teams that build this chain from the start lose speed early. Teams that defer it lose far more at the end.
This article was originally published in Turkish by Gökhan MERCANOĞLU on January 10, 2013. The English edition has been reviewed and edited by the author.