Every number on this page is computed by the Volume II Chapter 15 notebook's own code with the book's seeds (Online Appendix Remark 15.B.1); the ten illustrative programs in step 5 are new draws from the lab's seed, 26215. Your notes stay on this page until you leave it; copy the memo before you go.
The chief operating officer proposes to retire the engineering office's count of capability and run Chapter 7's hiring rule on a digital twin fed by the skills survey. The audit committee wants to know what the twin would cost, which readings are worth buying, and how the twin would notice a change in the world. You support the team preparing the answer.
Use AXIOM to build the twin, size its rule's margin, price better readings, and set its monitor and envelope.
Worked Example 15.1, Step 1: the law, costs, cap, and discount are Chapter 7's, and the program starts from an audited 58. The survey runs twice a year, at the start of every second quarter, and its error is normal with a standard deviation of 2.5 points, so that 95 percent of readings fall within ±5 (Volume I, Table 5.2). The twin knows the law and its own hires but not the swings. It keeps its estimate by the Kalman filter and runs Chapter 7's rule as a target for planned capability on its estimate; the target is chosen by simulation on 20,000 design paths and valued on 100,000 fresh ones, all on common random numbers. Its envelope is Chapter 14's guardrail, which every target here keeps.
In the twin panel, run the Kalman filter on simulated surveys; compare its error with the survey read as exact and with no readings.
(a) One simulated program from an audited 58 (the book's seed 1503): capability, the skills survey's readings twice a year, the twin's estimate with a band of two standard deviations, and the survey read as exact and carried forward by the hires and the decay, on the same readings. (b) The estimate's error by quarter.
In the decision bench, sweep the target for the count, the twin, and the survey read as exact; mark each best target and its cost on fresh paths.
Best target chosen on the 20,000 design paths (seed 1500); cost from 58 and its standard error on the 100,000 fresh paths (seed 1501), $ million; above the count, the same; below floor, the long-run share of quarters below 55.
In the readings panel, vary the survey's frequency and error, and price each design against the count.
The readings study: each case's best target searched on the first 10,000 design paths and valued on the first 40,000 fresh paths, above the count on those paths (Online Appendix Remark 15.B.1).
In the monitor, set the cumulative sum's threshold for a false-alarm rate, and test it on a drift in the law.
In the governance panel, decide what the twin may do alone and when it escalates.
Ten programs under the drift, each run by the monitored twin: new draws from the lab's seed, 26215. (a) The error each program decides with, quarter by quarter; (b) the cumulative sum G at each survey, until the alarm sends the engineering office to audit.
Target: planned capability (points); error: the estimate's standard deviation just after and just before a survey; cost above the count: present value at 10 percent, $ million, on 100,000 fresh paths; below floor: long-run share of quarters below 55.
An error that settles between surveys (Theorem 15.1); a margin that must grow with the twin's error (Theorem 15.2(iii)); readings priced by what they save (Theorem 15.2(i)); a monitor that trades false alarms for speed (Theorem 15.3(ii)); and an uncertainty that runs on a calendar (Proposition 15.3(iii)).
Whether to retire the count, the twin's rule and margin, the readings worth buying, and the monitor and envelope the twin should run under.