Every number on this page is computed by the Chapter 7 notebook's own code. The 1,000 starting gaps of step 5 are the notebook's draws with the lab's random seed 26107, not Meridian records. Your notes stay on this page until you leave it; copy the memo before you go.
Meridian's executive committee is setting three standing policies for the next two years: how much to invest each year in capability, how aggressively to hire against the engineering gap, and what to report to lenders beyond the covenant package. It wants to know what each policy does over time, not just next quarter. You support the committee.
Use AXIOM to separate momentum from program, find the steady state each investment level buys, test what lenders can infer from each disclosure option, and choose a hiring rule that is robust to delays.
The hiring loop of Worked Example 7.1; the three-reading company of Equation 3.2; the covenant model of Online Appendix 7.B.2; the start-up's logistic law of Proposition 7.4.
In the time-semantics bench, run a continuous model, sample it quarterly, and compare the exact discretization (Equation 7.2) with a simple step-by-step approximation on a nonlinear law.
The three-reading company of Equation 3.2, investment held at 8 a year, read as the yearly sample of a continuous law. Dots: the yearly law. Line: its exact quarterly law, A1/4 (Equation 7.2). Dashed: a simple quarterly step of the continuous law.
The start-up's logistic law ẋ = 0.9 x (1 − x/100) from 2 (Proposition 7.4; Online Appendix 7.B.4). A nonlinear law has no exact translation: a step-by-step approximation runs behind it.
Theorem 7.1(ii): an error ε₀ can grow like eLt, so a forecast needed within τ holds for (1/L) ln(τ/ε₀). Capability needed within 15 points, sensitivity L = 0.5 a year (Section 7.3).
In the superposition splitter, separate the free and forced responses of the three-reading company, then change the held investment and watch the steady state move along the policy map (Equation 7.6).
Equation 7.4: the path is exactly the free response (momentum alone, Akx₀) plus the forced response (what the program adds, starting from zero). Figure 7.1 draws technology at 8 a year. Dotted: the steady state of the policy map, x* = (I − A)−1B ū.
In the observability probe, choose which readings the lender package reports, row by row, and watch the rank of the observability matrix (Equation 7.5) and the invisible directions respond.
The covenant model of Online Appendix 7.B.2: four readings in quarterly deviations from plan; arrows show which reading feeds which, with its weight a quarter. The covenant package reports liquidity and leverage. An outsider who knows the inputs can reconstruct the state exactly when the rank is 4 (Theorem 7.2(iii)).
One standard error of the least-squares reconstruction of technology, with one unit of independent error on every reported number.
On the gain bench, sweep the hiring gain through every regime of Table 7.4, then add a one-quarter delay and repeat.
dk+1 = 0.9 dk + 0.6 uk under uk = −γ dk, from 100 engineers short; with the delay, dk+1 = 0.9 dk + 0.6 uk−1. Right: the closed loop's eigenvalues against the unit circle (Figure 7.2).
In the Lyapunov sketchpad, test whether a composite scorecard of your choosing falls every quarter along simulated paths.
The program's contribution grows while momentum fades (Theorem 7.2(i)); readings that feed nothing reported stay invisible however long the record (Theorem 7.2(iii)); and the delay halves the safe range of hiring gains (Theorem 7.5).
Recommend the investment level and its steady state, the hiring gain and its margin from instability with and without delay, and one disclosure change with what it reveals.