How a replay works: quarter by quarter, with no hindsight.

The question

When could a committee have seen the overrun?

Large projects tend to overrun for structural reasons: a single-point estimate at sanction, threats treated as independent when they move together, and in-flight reviews that confirm an overrun once it is hard to act on.

A replay turns that into a question you can test. Suppose a committee had used Capibud at sanction and every quarter after, with nothing but the information public at the time. When would it have seen the overrun coming, how far ahead of the public disclosure, and what could it have done about it?

The method

Six steps, with no look-ahead

1. Assemble a point-in-time record. Filings, regulator documents, press reports and market prices, each stored with the date it was published. Nothing enters the replay before its publication date.
2. Set a range at sanction. Each project gets a P50 and P80 for cost and finish on its sanction date, from its reference class and the shared factors, before any construction evidence exists.
3. Step forward one quarter at a time. For each quarter the replay shows the public estimate, Capibud's P50 and P80, the rules that fired and the documents behind them.
4. Test for look-ahead. An automated test checks that no quarter's forecast uses any information dated after that quarter. If it finds one, the replay fails.
5. Measure the warning and price the responses. Time in hand runs from the first in-flight warning to the date half of the eventual overrun was public. At the first warning, responses are priced as they would be for a live project, and any of them can be adopted, routed for approval and recorded.
6. Score against outcomes. For completed projects, the range set at sanction and each quarter's range are compared with the final cost and finish, and with the sanctioned budget. Misses are reported alongside hits.

An invented example

One illustrative LNG export project, step by step

The invented project is a two-train LNG export plant sanctioned with a budget of $8.0bn. Quarters are counted from sanction.

Reading the table

In this invented case the overrun is $2.2bn on an $8.0bn budget, so half of it is public once the estimate reaches $9.1bn, in quarter 13. The first in-flight warning fires in quarter 6. The time in hand is the gap between the two: seven quarters, or about 21 months.

Pricing the responses at the first warning

At quarter 6 the replay prices the responses open to the committee. Three invented ones show the form:

• Bring train 1 forward and let train 2 follow: about $0.5bn of expected value in the example.
• Put the marine works on a fixed-price contract: about $0.2bn, with a narrower range.
• Hedge the foreign-currency share of spend: narrows the range but adds little expected value, about $0.05bn.

In the real product, each response carries its range and can be adopted, routed for approval and recorded, exactly as for a live project.

What the example does not show

An illustration of the method, not evidence for it

• No accuracy claim. We invented both the forecasts and the outcome, so the example cannot show whether the method works. It only shows what each step produces.
• Public information is coarse. A replay sees only what was published. A live deployment reads monthly actuals and contractor data, which a replay cannot see.
• A replay cannot show what people would have done. It prices the options a warning opens; it cannot say whether an owner's team would have taken them.
• The real results are behind sign-in. The megaproject replay, its projects, its hits and its misses are shown in a guided walkthrough. Our published forecast tests are on the case studies page.

What this means for you

Ask for the same test on your own history