Asked about test coverage in a technical due diligence session, an engineering team almost invariably answers with a percentage — line or branch coverage on the software side, item counts within the factory acceptance test protocol on the hardware and facility side. Reframe the same question as which failure modes are covered, and the vocabulary of the answer shifts: coverage is suddenly described in terms of files, modules and equipment items rather than outcomes. That shift exposes a pattern everyone in the room senses and few name, namely that scope has been shaped by what is convenient to test rather than by what failure would cost. The same asymmetry is observable in factory acceptance protocols, where a substantial share of the line items originates in the supplier's standard form, while the thing that actually causes trouble in the field is typically the interface between two suppliers' systems — an interface that appears on neither party's standard form.
A second and quieter observation concerns the life history of the coverage document itself. In most companies the test scope document is written once, generally at the request of the first institutional customer or the first audit; it is accurate on the day it is signed, and it remains in that state while the product moves through three releases, the facility through two expansions, and the team doubles in headcount. The test suite, meanwhile, continues to grow, but the logic of that growth is accretion rather than design: after every field incident a test is added to catch that incident, no test is ever removed, and no item is ever re-prioritized. What emerges is not a map of current risk but a chronological inventory of past accidents, in which scenarios whose probability has since collapsed are re-exercised every night while the portion of the architecture that changed in the last twelve months goes untested.
The mechanism beneath this behavior operates on two levels. The first is the substitution of what is measurable for what ought to be measured: because coverage percentage is a number easily produced, easily placed on a dashboard and easily shown to outsiders, it displaces the harder question of coverage quality, and after a period the team begins improving the number rather than the coverage. The second is the substitution of what is vivid for what is dangerous, in that a failure already experienced sits mentally far closer than a failure never experienced whose consequences would be considerably heavier, so testing investment distributes itself according to familiarity rather than according to likelihood and severity. Neither tendency is an error. While the team is small, while knowledge is genuinely shared, and while the product runs in a single customer configuration, leaving scope implicit is a rational shortcut that meaningfully reduces documentation cost. The difficulty lies not in the shortcut but in its persistence once the team has grown, the customer base has concentrated, and the consequence of a single failure has become a contractual penalty.
On the ownership dimension the mechanism is subtler. In most engineering organizations the person who writes the test is the person who made the design, a choice defensible on grounds of speed but one that produces coverage inheriting the design's blind spots wholesale, since an engineer cannot readily construct a test against an assumption that functions, for that engineer, not as a proposition to be examined but as the ground beneath the examination. Layered onto this is the fact that the scoping decision is never recorded as a decision at all: what will be tested is approved in a meeting, while what will not be tested is discussed nowhere. Coverage thus becomes not the sum of decisions taken but the residue of decisions never taken, leaving the company without an inventory of the risk it is carrying.
The counterpart of this configuration at the review table is direct and unforgiving. An investor or acquirer will not accept a practice described orally as an existing practice; what it looks for is coverage that is formally defined — a version-controlled record, dated and approved, traceable back to requirements, showing which failure mode is addressed at which test level. When what enters the data room is a screenshot of a coverage report, that artifact is classified not as verifiable evidence of control but, at best, as an indicator, and the consequence of that classification appears not in the technical section of the agreement but in its economic one: every unverifiable practice excluded from representations and warranties migrates into the escrow percentage, a condition precedent, or a special indemnity. The cost of undocumented coverage lies less in the engineering hours required to remedy it than in the buyer's reflex to collect that uncertainty through price.
The absence of the measurement dimension enters valuation through a wider channel. Escape rate — defects found in the field relative to defects found in test; rework hours per release; time-to-closure on nonconformities raised during commissioning; root-cause distribution of warranty-period callouts — where none of these series is maintained, the reviewing party does not stop estimating them, it simply estimates them on its own assumption and builds that estimate from the conservative end of the distribution. In practice this presents as an upward adjustment to warranty provision in the normalization schedule, a rework reserve added to the first-year operating budget, and an earn-out threshold conditioned on defect rate. Coverage without measurement, saying nothing about management quality, likewise says nothing about forecast accuracy; and what is ultimately being purchased is not historical performance but the predictability of future cash flow.
The continuity dimension opens the most expensive of the three channels. Nearly every engineering organization contains one person who knows what will break; that person makes scoping calls by intuition, builds the critical tests personally, and quietly checks one or two additional things before a release ships. This knowledge appears under no line item on the balance sheet, yet it is priced with considerable precision in the closing negotiation as a key-person agreement, a retention bonus, a non-compete term, and an earn-out tied to an individual. Reproducibility of coverage, for its part, is not an abstract notion: it is measured by whether an engineer hired within the past year can, consulting no one and working only from existing records, produce a coverage proposal for a new component — and the outcome of that exercise constitutes the only real evidence supporting a claim of scalability.
Managing this configuration through institutional architecture proves both cheaper and more durable than managing it through individual discipline, and it reduces to five separable components. The first is a register of failure modes rather than of tests, in which each line is tied not to a test but to an outcome and to that outcome's contractual, operational or safety cost, with the scoping decision set against that cost. The second is the traceability chain, keeping the link from requirement to test and from test to evidence under version control, with the coverage document revised on the same cadence as the product. The third is the measurement set: critical-path coverage in place of line coverage, accompanied by escape rate and rework hours. The fourth is the separation of ownership, in which the role approving scope is distinct from the role building it, so that the design's blind spot is examined from outside at least once. The fifth is cadence, fixing the gates at which scope is reopened — design freeze, pre-factory acceptance, pre-commissioning, and major release — rather than leaving reopening to occur between them.
BEIREK's intervention in this area begins not by expanding coverage but by making the exclusions visible. On the projects we manage, the record we maintain is not a list of what was tested but a reasoned record of what it was decided not to test, stating for each item which failure mode is being accepted, which compensating control carries that acceptance, and on what date and by whom the decision was taken. That record is kept at the moment the decision is proposed rather than the moment it is approved, since a record kept at approval documents only the outcome while a record kept at proposal preserves the reasoning as well; and in a due diligence session, the document that persuades a buyer's technical adviser fastest is not the coverage itself but precisely this record of reasoning.
The second line of intervention concerns cadence and role. At the gates where scope is reviewed we place a counter-argument role whose sole function is to advocate for what has been left outside the scope, a role staffed not from the team building the project but from the contract and operations side, because the true cost of a failure mode is held not in engineering hours but within liquidated damages, availability undertakings and warranty scope. To this we add the discipline of test retirement: at every review, items removed from the register are recorded with the same rigor as items added, failing which the coverage ceases to be a risk map and becomes an archive. Finally we run the handover test, since the reconstruction of the same scope from existing records while the engineer who built it is absent from the room is the only verifiable moment at which institutional capacity separates from personal capacity.
Test coverage is a record not of what a company's product verifies but of which risk that company has knowingly chosen to carry. Where the choice is recorded together with its reasoning, every gap in coverage becomes, for the reviewing party, evidence of mature prioritization rather than a weakness, and is read in favor of management quality. Where it is not recorded, the same gap never remains a technical discussion; as in every transaction in which unmeasured uncertainty is priced by the counterparty, it passes directly into the closing structure and, ultimately, into the consideration.