Risks That Only Bite When Another One Fires

Every register has a column at the bottom. Probability times impact, summed: so many days, so many pounds. On this one it says 118.5 days and £1.98m, and everyone in the room knows it is not a forecast. Two risks on the same chain, one after the other, cost more together than either costs alone. Two on different blocks cost the time of the worse one. One with a month of float costs money and no time at all. The sum cannot tell those apart. The programme can, so I put eighteen ordinary hospital risks on one and ran them in pairs.

The programme

A generated acute hospital on a live estate: four blocks, an energy centre, a car park, a helipad, a link bridge, four phased handovers, 10,117 activities and £428,000,000 of priced work from 1 March 2027 to a logic-driven finish on 20 March 2031. Block C, the theatres, sits on the footprint of a ward block that has to be decanted and demolished first, and the critical path runs straight through it: asbestos, demolition, piling, frame, fit-out, commissioning, twelve theatre validations in turn, handover.

The welcome page a moment after importing the 11.5 MB XML — The 11.5 MB file imports in about ten seconds: 10,117 activities, 19,144 links, six trades on hourly rates.

The DCMA schedule check on import: gate pass, eleven of eleven applicable checks passing, three not applicable — The 14-point check on import: **11 of 11** applicable checks pass, three not applicable because nothing has started. High float 3.9%, no activity over 44 days, no constraints, no leads.

The board with the four blocks folded to their summary rows: Block A May 2027 to November 2030, Block B to October 2030, Block C October 2027 to January 2031 with its sub-phases, Block D February 2028 to November 2030 — The four blocks folded to their summary rows. Block C starts last, because the ward block has to come down first, and finishes last, because the theatres are validated one at a time.

The register

Eighteen risks, each with a likelihood band, a delay in working days, a cost, and the activities it is linked to. Asbestos beyond the survey, obstructions under Block C, winter concrete on Block A, a late steel fabricator on Block B, the DNO diversion, the switchgear, the air handling units, Gateway 2, a drylining shortage across 214 zones, a theatre design change, an M&E insolvency, lifts, imaging, the validations, the decants. Nothing exotic. Six have a mitigation written against them; two of those, the trial holes, are already done.

The risk register page: 18 risks, 3 critical and 4 high, expected 118.5 days and £1.98m, 110.8 days and £1.78m after the implemented mitigations — **3 critical, 4 high, 11 medium**. Expected across the register **118.5 days and £1.98m**, 110.8 days and £1.78m after the two mitigations already done, 96.1 days and £1.61m if the other four are carried out. Six mitigations, £430,000.

The probability by impact matrix with the eighteen risks placed on it — The matrix the register is scored on. Three in the critical corner: the ground under Block C, winter concrete, and Gateway 2.

One thing about how a linked risk counts. The drylining shortage is linked to 214 partition zones. If it fires it adds ten days to the job once, on the linked activity where it bites, not ten days to each zone. The form says so under the field.

The edit form for the drylining risk: Likely, 10 days, £150,000, score 12 High, and the helper text under the delay field saying it is added to the job once if it happens, not once per linked activity — "Added to the job once if it happens, not once per linked activity." Likely, 10 days, £150,000, a score of 12; a second drylining subcontractor proposed, 50% off the impact for £120,000.

The board with the switchgear risk open in the side panel and its two linked bars highlighted on the chart — The switchgear risk open on the board, its two linked bars lit on the chart: the delivery and the energisation it gates.

The whole job first

The whole job, 2,000 runs, all 18 risks: finish at 80% confidence 16 June 2031 without risks and 4 September 2031 with them, 80 calendar days later; cost at 80% £428,433,600 and £430,551,672 — All eighteen, two thousand runs, three and a half minutes. At 80% confidence the finish moves from 16 June to **4 September 2031**, 80 calendar days, and the cost from £428,433,600 to **£430,551,672**. The plan's own date: about 0% of the time.

Read the left-hand column before the right. With no risks at all, the 80% finish is already 88 days past the plan, and the best of two thousand runs is 67 days past it. That is the importer's default range doing what a default does: one day under, two over, on every activity, and a critical path of 361 short activities that can barely finish early and can all finish late. The risks are the 80 days and £2.1m on top. Keep that column in mind; it decides what the pairs show.

Two on the same chain

Sixteen risks closed, two left open: the asbestos beyond the survey and the ground under Block C. One sits on the demolition, the other on the piling that follows it. Zero float, one after the other. Then each on its own.

Three results stacked: the asbestos and ground risks together, 15 days later at 80%; asbestos alone 4 days; ground alone 8 days — Top, both: **15 calendar days** and £526,480. Middle, asbestos alone: 4 days, £171,520. Bottom, the ground alone: 8 days, £359,832. Four and eight make twelve; together they make fifteen.

The money adds, near enough: £531,352 apart, £526,480 together. The days do more than add, because at the 80% line a risk that fires half the time is partly hidden in the half where it does not, and two on the same chain leave fewer runs where neither fired. The register wrote 7.5 days and 10 days and summed them. The programme says the pair is worse than its parts.

Two on different blocks

Three results stacked: winter concrete on Block A and the Block B steel fabricator together, 8 days later at 80%; winter concrete alone 8 days; steel alone 1 day — Winter concrete on Block A and a late fabricator on Block B. Together **8 days** and £408,016. Winter concrete alone 8 days and £190,280; the fabricator alone 1 day and £232,528.

Eight and one make eight. The two blocks run side by side and the finish only hears from whichever is worse, which is the winter pours seven times in ten. The fabricator's money still lands, all of it. The register summed 8.4 and 6 days and was wrong by most of it.

Float swallows days, not money

The DNO diversion risk alone: finish at 80% unchanged at 16 June 2031, cost up £68,608 — The DNO diversion, thirty working days of float behind it. The 80% finish does not move. The cost goes up **£68,608**.

Possible, ten days, £80,000, a 9 on the matrix. On the finish it is nothing: the 80%, the 90% and the worst case of two thousand runs sit where they sat with no risks at all. On the money it is every pound, because the engine always adds the cost and only lands the delay where the logic lets it.

The single point that was not one

Two results stacked: the switchgear and the air handling units together, 1 day later at 80%; the air handling units alone, finish unchanged — The switchgear and the Block C air handling units together: **1 day** and £115,120. The air handling units alone: no days, £54,000.

I expected this pair to hurt. On paper the energisation has zero float and the switchgear risk is twenty days. The panel gave it one day, and the worst case of two thousand runs did not move at all. The reason is the left-hand column from earlier. The simulation finds the critical path again on every run from the sampled durations, and three hundred short fit-out activities, each leaning late, outgrow a dozen long energy-centre ones by weeks. By the time the switchgear fires, the energisation has float it does not have on the plan, and the engine does what it says: float absorbs. The plan's float table and the simulation's are not the same table, and the risk I was surest about did least.

One approval, four blocks; and the last one

Gateway 2 alone: finish at 80% 22 July 2031, 36 days later; cost up £425,304 — Gateway 2 late, linked to the first frame activity of each block: **36 days** at 80%, 39 at 90%, and £425,304. More than the asbestos and the ground together, because there is no block it is not in front of.

The theatre validation risk alone: finish at 80% 23 June 2031, 7 days later; cost up £76,736 — The theatre validations failing first time, twelve links at the very end: **7 days** and £76,736. The same scores as the DNO diversion, which moved the finish by nothing; here there is nothing after the twelfth validation but the handover.

What the mitigations buy

The register hero with all six mitigations implemented: expected 118.5 days and £1.98m, 96.1 days and £1.61m after implemented mitigations — All six marked Implemented. The register's own arithmetic: **118.5 days and £1.98m become 96.1 days and £1.61m**.

The whole job with six mitigations implemented: the lede says 6 are shown after the mitigations marked Implemented; finish at 80% 26 August 2031, 71 days later; cost at 80% £430,373,280 — The same run with six implemented, and the lede says so. The 80% finish comes in from 4 September to **26 August 2031**, 71 days past the no-risk line instead of 80; the 80% cost from £430,551,672 to **£430,373,280**.

Nine days and £178,392 at the 80% line, for £325,000 of winter working, an early switchgear order, a second drylining gang and theatre pre-validation. The register promised 14.7 expected days. Nine of them reached the finish; the rest went to risks it was not listening to.

Finance, reading the same run

Finance tab, how might this finish: likely finish August 2031, 140 days later than plan; P10 July 2031 104 days later, P50 August 2031 140 days later, P90 September 2031 183 days later; 0% of runs finish on or before the planned date — The Finance tab on a second run of the register as logged: most likely **August 2031, 140 days later than plan**, best case July, worst case September; 0% of runs on or before the plan. The second run put the 80% finish on the same 4 September and the 80% cost within £3,456 of the first.

What this does not tell you

The ranges are the importer's defaults, nobody's judgement, and they lean late; that is why the no-risk finish is already June and why the switchgear did nothing. Type ranges from the people doing the work and the pairs will move. Every risk fires on its own: a bad winter on Block A does not make the fabricator later, and on a real job it might. A delay lands once, on the linked activity where it bites; a risk that would stretch 214 zones by a day each is not modelled that way. And two thousand runs of ten thousand activities took three and a half minutes here, past the three-minute cut on the live server, so a job this size needs fewer runs there.

Run it on your own register

GeneraGantt reads MS Project, Primavera P6 and Asta Powerproject. Log the register against the activities it would hit, close everything but the two you are arguing about, and run them. Then open the two. Free while it is in testing, with a sample programme loaded. Nothing is saved: your schedule leaves with you.

Open GeneraGantt How the register prices a job

The programme in this article is a generated one, built so that risks of each kind sit where they would on a real acute hospital: in series on the demolition chain, in parallel on two frames, behind float, and in front of everything. The trust and the town do not exist. Every figure quoted is read off the app after importing the file, logging the register through the app's own service and running the panel; nothing is quoted that is not on a screenshot.