Case Study 40.1: Three Offers

The situation

A software engineer with six years of backend and distributed-systems experience has spent a year learning quantum computing — roughly the content of this book. They have three offers.

All three are real jobs at funded organizations. All three say "quantum" in the title. The engineer wants to know which is a career and which is a bet.

Step 1: apply question 1 to each

The engineer's first move is Chapter 21 §21.7's question, asked of the employer: compared to what?

Offer A. The interview described benchmarks against "classical heuristics." Pressed for specifics, the team named a greedy allocator and a random baseline.

>>> from vqelab.claims import is_tuned_baseline
>>> is_tuned_baseline("greedy"), is_tuned_baseline("random")
(False, False)

The engineer asks whether they have benchmarked against Gurobi or CPLEX — the solvers a bank's existing desk actually runs. The answer is that a comparison is "on the roadmap."

That is the finding. Not that the team is dishonest — they are enthusiastic and technically strong — but that a two-year-old company selling optimization to banks has not compared itself to the incumbent. Chapter 37 measured what happens when you do: QAOA lost 0–6 to an algorithm from 1994 on the standard benchmark, and Case Study 37.1's incumbent was printing a 0.046% optimality gap every day.

Offer B. The baseline is their own previous compiler version, measured on circuit depth and two-qubit gate count across a benchmark suite. Chapter 39 §39.6 measured why this is real work: the transpiler seed alone moved 14-qubit fidelity from 0.5755 to 0.7911, a 2.03× swing in error, driven by a 49-to-112 spread in two-qubit gate count.

Offer C. The baseline is the decoding threshold in the literature, and the metric is latency under a microsecond budget. Chapter 15's measurement is the context: one T gate takes a circuit from 450 to 2,882 physical qubits. Decoders are on the critical path of every fault-tolerant proposal.

Step 2: which of §40.1's roles is this actually?

• A is applications research, the second-smallest category, at an organization whose thesis this book's Chapter 37 measured directly and found behind.
• B is compiler engineering. The engineer's existing skills transfer almost entirely; the quantum content is the coupling map and the noise model.
• C is real-time systems and classical algorithms, with a physics-literate specification.

All three are largely classical engineering jobs. The difference is what happens if the field's timeline slips.

Step 3: the falsification test

What would have to be true for this work to have been worth doing, and what would show it was not?

Offer A. Worth doing if QAOA beats Gurobi on a real portfolio instance within the funding runway. Falsified if it does not — and nothing in the company's current benchmarking would reveal that, because the benchmark is against greedy. Chapter 37 §37.9's honest version: no instance family has been demonstrated on hardware where the classical guarantee is tight and QAOA does better.

Offer B. Worth doing if the hardware improves enough to run useful circuits. Falsified if the hardware stalls — but the compiler work is publishable, transferable, and the skills are the same ones that make a person employable at any compiler shop. The downside is bounded.

Offer C. Worth doing if fault tolerance arrives. Falsified if it does not — but decoder work is real-time systems engineering, and Chapter 15's numbers make the problem well-posed regardless of whether any particular machine gets built.

The asymmetry is the whole analysis. Offer A's thesis and the engineer's career succeed or fail together. Offers B and C build skills that survive the thesis failing. That is not a statement about which is more likely to succeed — it is a statement about correlation, and correlation is what decides how much of your career you are wagering.

Step 4: what the engineer does

They take B, and the reasoning is not that A is doomed.

Offer A is the most interesting problem and the most money, and I think there is a real chance the team is right. What I could not get comfortable with is that their benchmark cannot tell them if they are wrong. Comparing to greedy will always look good. If I join and we spend two years beating greedy, I will not have learned whether the approach works, and neither will they.

Offer B is a compiler job where the quantum part is the coupling map. Chapter 39's layout measurement — 2.03× fidelity swing from the transpiler seed — says there is real, unclaimed engineering there, and it is engineering I can already do. If quantum computing stalls for a decade, I am a compiler engineer. If it does not, I am a compiler engineer in the right place.

They also send Offer A's team a note suggesting a Gurobi benchmark, with Chapter 37's methodology attached. Two years later that team pivots to a hybrid product using the classical solver for the optimization and quantum sampling for a subproblem — which is a better product, and it exists because somebody finally ran the comparison.

What this case study is about

The engineer did not evaluate quantum computing. They evaluated three specific organizations' relationships with evidence, and one of them had a structural blind spot that no amount of talent would correct.

The transferable move: ask what the employer's benchmark is, and whether it could return a negative result. A team that cannot describe the measurement that would falsify its thesis is not doing research — and this applies well outside quantum computing.

And the honest counterweight: Offer A might work. Betting on it is a legitimate choice. The mistake would be making that bet without knowing you were making it, which is what a greedy baseline guarantees.

Questions

1. The engineer asks Offer A about Gurobi. What are three other questions that would reveal the same blind spot, and which is the least confrontational?
2. Offer C is at a national lab with the lowest compensation. Construct the strongest case for taking it, using Chapter 15's numbers.
3. The analysis rests on "correlation between the thesis and the career." Is that the right frame for someone at a different career stage — a new graduate, or someone twenty years in?
4. [measure] Using vqelab.claims, write the Claim object for Offer A's pitch as described. How many of the eight checks does it pass?
5. The hard one. The engineer's note leads to a pivot two years later. That is a flattering ending. Construct the realistic version where the note is ignored, and say whether it changes any of the reasoning above.