Part 5 of this series argued for deception and cost imposition. This version argues the other side first, as hard as it can be argued, because a defense budget spent on theater is worse than a defense budget unspent, and deception is the easiest category in security to buy as theater.
The skeptic's case has four planks.
First, the numbers are marketing. The headline figures in this space, 87% of agent traffic diverted, attacker costs up 2.3x, come from the vendor's own testing, with no published methodology, against opponents the vendor selected, in an environment the vendor built. Security has a long, embarrassing history of exactly this evidence shape, and the correct prior is that the real-world effect is a fraction of the claimed one.
Second, the best attacker already detects the maze. The vendor's own account concedes its test ARTEX flagged the honeypot at minute six. Frontier harnesses are improving at exactly the skill deception depends on them lacking: telling real systems from staged ones. Every model generation narrows the window in which a decoy is indistinguishable from production, and the defender must re-stage the entire theater each time. That is an arms race on the attacker's home field, perception, where the model labs' progress accrues to the attacker for free.
Third, deception defends the perimeter of an estate whose real problem is inside. The incidents that define this era run on identity chaining through legitimate credentials. A harness walking your internal systems on a stolen CI/CD token does not meet your decoy endpoints. It meets your real ones, with authority. Money spent on mazes is money not spent on the short-lived, narrow credentials that actually break the chain.
Fourth, decoys generate their own operational debt. Someone must keep the fake estate plausible: patched-looking, data-consistent, topologically believable. A stale decoy is worse than none, because it teaches the attacker your deception style at no cost and tells your own team a comforting story about coverage that does not exist.
Now the rebuttal, because the skeptic's case proves too much. It proves deception is oversold, perishable, and secondary. It does not prove deception is worthless, and the distinction is where the budget should land.
The tax is real even when the maze is detected. Six minutes and a pile of tokens, spent by the attacker, on nothing, is a cost imposed, and against volume operations running thousands of targets, small per-target taxes are the whole game. Detection-resistance was never the product. The product is cost per attempt and telemetry quality, and on telemetry the skeptic has no answer: a decoy credential has no legitimate user, so its use is the cleanest intrusion signal that exists, and a thorough harness cannot probe everything and avoid decoys at the same time. Thoroughness is the attacker's engine and the tripwire's trigger, in the same motion.
And the ordering objection is a agreement in disguise. Yes: identity hygiene first, the verified fix loop second, deception third, funded from what is left, sized as a tax and a sensor network rather than a wall. The skeptic's version and the advocate's version converge on the same program, differing only in how much of the budget line the third station deserves. That is a calibration question, and it should be settled the way everything in this series should be settled: run the harnesses against your own estate, with and without the deception layer, and measure the attacker's cost and your detection time in both configurations. Proof over claims, including the claims made for the defense.
The final part of the series assembles the operating model, deception included, at its measured size.