Two modules in this course have already named the testing period as the most exposed point in a project's calendar, without ever treating it in its own right. It is the moment when the works are worth the most, since everything has been bought, delivered, erected and connected, and it is the first moment when something circulates inside: steam, current, a pressurized fluid, torque on a shaft. Until the day before, the risk was that of an inert assembly, exposed to theft, site fire and natural events. On the day of the tests, the energy the structure is built to transform is added to it, and this is done on an installation that nobody has yet verified will withstand it. Peak value and first energy loading therefore fall on the same day, and it is that coincidence, more than technical difficulty, that drives the period's loss record.
What fails at that moment is not random, and the repetition of files eventually draws up a short list. Rotating machines lose their balance because an alignment made cold does not survive thermal expansion. Steam and oil circuits give way at flanges, bellows and tie-in points, where erection left a stress that no low-temperature hydraulic test would reveal. Control chains react to quantities they are seeing for the first time, on provisional settings entered by people in a hurry. And above all, part of the protection system has been inhibited to allow the test itself, because a controller that trips on the first deviation makes it impossible to ramp up. The installation is therefore put under stress at the precise moment when its safety systems are least active.
Policies bound this period in three ways, which combine more often than they exclude one another. The first is a fixed duration, a few weeks of testing included in the erection cover, beyond which the cover ends for anything arising from testing. The second is a restriction as to subject matter: the testing cover applies only to equipment that is new and has never been in service, which leaves out second-hand items, machines relocated from another site and those restarted after a long shutdown, although these are statistically the most fragile at first start-up. The third is a derogating regime during the period: a separate deductible, higher than the one applying to the rest of the works, a dedicated sub-limit, or cover cut back to sudden and accidental damage, excluding whatever arises from adjustment and fine-tuning.
The first of those boundaries carries a counting trap that its wording never flags. The testing duration almost always runs in calendar days from the first test, whatever that test was, and not in days of actual testing. It therefore keeps running through shutdowns, through the wait for a part, through an erection rework, through a contractual argument between contractor and project owner. Yet a testing campaign that stretches out is exactly the campaign of a project going badly, that is, the one with the highest probability of damage. The cover's calendar and the risk's calendar thus run in opposite directions: cover is consumed all the faster the more the installation holds, and it has expired by the time the decisive test finally takes place.
Extending that duration is a contractual act, it requires the insurer's agreement, and it is almost always requested too late. A project learns that it has run past its testing window on the day it notifies a loss, because nobody on site keeps the count: it appears neither in the works schedule, nor in the punch list tracker, nor in the progress reviews. It is also worth seeing that declaring after the fact a period already elapsed and extending a period still running are not the same request. The second is negotiated on a future exposure, which the insurer can assess; the first asks it to cover an interval whose outcome is known to the party asking.
Resuming tests after a defect has been corrected raises a neighboring question, and wordings rarely answer it explicitly. After a shutdown and a repair, is a new test the continuation of the same program, still consuming the same count, or a new test that reopens a period? The answer matters twice over, because a successful test is frequently the event that triggers acceptance, and therefore the switch described in the second module. A failed test thus prolongs exposure under the construction regime and pushes back by as much the start of the maintenance period and of the statutory warranties, while the testing window keeps emptying. Three clocks run at once, at different speeds, and only one of them is watched.
That leaves the interface with delay in start-up cover, which turns an average file into a heavy one. The delay indemnity is counted from the commissioning date set in the contract, so that a testing incident hits both covers on the same day: it destroys equipment and it postpones production. But delay cover is attached to the material damage, and it responds only if that damage is covered. If the testing window has closed, it therefore carries away both, and the gap between the two outcomes is not the amount of the damage, it is that amount plus several weeks of lost margin. A single calendar date decides two covers, one of which often weighs more than the loss itself.
A paper mill has a new paper machine installed. The erection all risks policy includes four weeks of testing, that is, twenty-eight calendar days from the first test of whatever nature, with a deductible specific to testing. The first cold test, no-load rotation of the wet end, takes place on April 2, 2025. On April 11 an alignment defect appears on a dryer roll; the site shuts down, orders the part and waits nineteen days. Testing resumes on April 30. On May 6, at the first full-load start-up with steam, a dryer section flange fails: 2.4 million euros of damage, of which 1.6 on the dryer section and 0.8 on the steam circuit. Delay in start-up cover is attached to the policy, with contractual commissioning on May 15, a thirty-day deductible and a daily indemnity of 34,000 euros. Actual commissioning takes place on July 28, that is, seventy-four days late. What is left?
The reflex is to say the damage occurred during testing, that it is therefore covered, and then to argue over the flange itself as an inherent defect. That argument will not happen. The testing window started running on April 2, it was worth twenty-eight calendar days, so it closed on April 29, seven days before the loss, and it closed while no testing was going on: the nineteen-day wait for the part consumed two thirds of the cover with nothing being tested. The decisive fact in the file is therefore not the May 6 flange, it is the April 11 shutdown, and nobody read the contract that day, because the testing count appeared in none of the trackers the site was keeping. The reading that matters is that of three words in the clause: calendar days or days of actual testing. Both wordings exist, and here they differ by nineteen days, which is the gap between a covered file and one that is not. The rest follows and adds up: delay cover being attached to the material damage, it falls with it, and forty-four indemnifiable days after the deductible, that is 1,496,000 euros, disappear on top of the 2.4 million. A spare part lead time, which is not a loss and which nobody notifies, will have cost close to 3.9 million euros.
- 01The testing period stacks two peaks on the same day: site value at its maximum and the first circulation of energy through an installation never proven under load.
- 02What fails at first start-up comes from a short list: alignments made cold, flanges and tie-in points left stressed by erection, provisional settings, protections inhibited to allow the ramp-up.
- 03Three boundaries combine: a fixed duration, a restriction to equipment that is new and never in service, and a derogating deductible or sub-limit regime during the period.
- 04The duration runs in calendar days from the first test, not in testing days: waiting for a part consumes the cover exactly as a test does, and the count appears on no tracker the site keeps.
- 05Delay cover is attached to the material damage: a closed testing window carries away both, and the delay portion often weighs more than the damage itself.