You cannot win this job at P80.

To sit at P50, this job needs $119,788 above the base estimate. To sit at P80 it needs $183,167. To sit at P95, $260,370.

The project is carrying $47,553.

Now convert those into tender prices, because that is where the decision actually bites. Add the overhead and profit and P50 becomes a tender of $1,070,847. P80 becomes $1,134,226.

What P80 would have meant

Week 7 listed the six bids on this job. The lowest was ours at $1,000,000. The next was $1,032,000.

So a tender priced at P80 would have come in at $1,134,226 — a hundred and two thousand dollars above the next contractor. Not close. Not a negotiation. A different postcode.

And here is the part that should be uncomfortable. A tender priced at P50 would have been $1,070,847, which is still $38,847 above the second bid.

Price this job at the median outcome and you do not win it.

Now run the six bids backwards through the curve and ask what confidence level each one implies. Ours was P10. The second bid was P26. The third, P48. The average of all six landed at P58, and only the highest bidder in the room was anywhere near P80.

A competitive tender does not clear at the eightieth percentile. It clears wherever the most optimistic contractor happens to be. On this job that was us, at P10.

Which means the honest question is not what level should we hold. It is: at what level are we prepared to stop bidding?

The median is not the middle

The usual defence of bidding low is that it averages out. Win some, lose some, and across enough jobs the good ones pay for the bad ones.

There is a real mechanism underneath that, and there is also a hole in it that costs contractors their companies.

Take the mechanism first. Run this job on its own and the relative spread of the outcome is about seven percent. Run twenty jobs like it and the relative spread of the total falls to 1.6 percent. Forty jobs, 1.1 percent. That is exactly the independence effect from Week 10 — the one that does not work inside a project because everything shares a February, but does work across a portfolio, because your job in Leeds and your job in Bristol do not share a crane.

Diversification is a company-level property. It is not available to a project, and it is not available to a contractor running three jobs.

Now the hole. The cost curve is not symmetric. It cannot be — costs are bounded below and unbounded above, which is why Week 9 said to use a long right tail. On this job the median is $946,796 and the mean is $956,440.

That gap is $9,644, and it is not a rounding error. It is the shape of the distribution.

Bid at the median and you are not right half the time and wrong half the time. You are short by $9,644 on average, on every single job, forever. Across forty jobs that is $385,747 of entirely predictable loss, and no amount of diversification touches it, because averaging does not move an average.

Spreading the bet makes the total more predictable. It does not make it smaller. Predictably short is still short.

Where the cover actually lives

Which points at the thing Cost & Cash set up twenty weeks ago and never fully explained.

This job carries $47,553 of contingency inside the performance measurement baseline, and $50,000 of management reserve sitting above it. The project manager can draw on the first. He cannot touch the second — it is the company's money, held by the company, released by somebody more senior.

That structure is the answer to the whole problem, and most people carry it without knowing why it exists.

The project holds enough to run at a level it can be held accountable for. The company holds the rest centrally, across every job, where the diversification actually works. Put the whole P80 into every project and you have priced yourself out of the market. Put none of it anywhere and the first bad winter takes the company.

On this job, base plus contingency plus management reserve is $924,561, which is about P36. That is the real confidence level this project is being run at, and nobody wrote it down anywhere.

So who decides?

Not the planner. Not the project manager. Not the risk register.

Choosing a confidence level is a statement about how much capital the business will expose per contract and how often it is willing to lose money to stay in the market. On this job it means accepting a 76.5% chance of covering cost and a 10.2% chance of recovering the full overhead and profit. Those are not project controls decisions. Somebody with a balance sheet has to make them.

The planner's job is narrower and more important than choosing. It is to make the choice visible — to put the curve, the levels, the implied tender prices and the win probability in front of the person who does decide, and to make sure nobody can later claim they thought the number was $47,553 because it was calculated.

Which is why most risk appetite statements are worthless. The company will not accept high risks. That sentence commits nobody to anything and cannot be tested. A usable one names a number: we bid at P30 and hold portfolio cover centrally at P80; we do not bid below P50 on new clients; any job below P20 goes to the board. Those can be complied with, and more importantly they can be breached.

Practical insight

Do this once and it changes the conversation permanently.

Take your last completed job. You have the tender sum and you have the final cost. Build the crudest possible curve — the reference class from Week 7 will do, ten jobs and a spread.

Now work backwards and answer one question: what confidence level did we actually bid that job at?

Then ask the same of the job you are pricing this month, and put the two numbers on one slide with nothing else on it.

You are not asking anybody to change the bid. You are showing that a level is already being chosen every time, silently, by whoever adjusted the last line of the estimate. The only question on the table is whether the person choosing it knows they are choosing.

Key takeaways

✔ P50 needs $119,788 above base, P80 needs $183,167, P95 needs $260,370. The job carries $47,553.
✔ A P80 tender would have been $1,134,226, against a second bid of $1,032,000. A P50 tender still loses the job.
✔ The six bids map to P10, P26, P48, P62, P79 and P92. Markets clear at the most optimistic bidder, not at P80.
✔ Diversification works across a portfolio and not inside a project — relative spread falls from 7.2% to 1.1% over forty jobs.
✔ The mean sits $9,644 above the median. Bidding at P50 is short on average every time, and averaging cannot fix an average.
✔ Contingency sits in the baseline, management reserve sits above it with the company. Base plus both is $924,561, or about P36.
✔ A risk appetite statement that names no number cannot be complied with and cannot be breached.

What's coming next

That is the analysis complete. Fourteen risks, ranged, correlated, simulated, and a confidence level that is somebody else's decision to make.

And every bit of it is still just a number. Nothing on this project has changed. The rock is exactly where it was in Week 1, the wayleave is still unconfirmed, and the model has not moved a single cubic metre of anything.

Phase D is about doing something. Next week we take the fourteen and ask what can actually be done to each one — and put the two boreholes from Week 8 through a decision tree, to find out whether spending $6,800 to find out about the rock is a good trade or just an expensive way of confirming bad news.