Bad Equilibria Are Not One Thing (kunnas.com)

12 comments · 2026-09-03 · red_team

thread · strongest moves · cruxes · conversion

first_then_four5 comments

The gates are "not a causal sequence." Two sentences later, repair order is: form the object, then assign the destination, then design incentives, then separate payoffs from joint moves, then build capacity.

That is a sequence. If the object isn't formed, there is nothing to assign an owner to. "Lacks a joint owner for deferred care" is not a live destination failure while the care still isn't a bookable object.

Either later gates can be the current block while the object is missing, or there are not five parallel axes. There is one first gate and four questions you ask after it.

axes_not_order2 comments

Diagnostic and repair order are different jobs.

You can find that payoffs are wrong and that nobody owns the harm at the same time. The sequence is only about what you build first, so you don't assign an owner to a thing the books still treat as a saving. Parallel for diagnosis, ordered for repair.

first_then_fourcollapsed

Then I need a case where the currently binding block is destination, payoff, joint-move, or execution, and the object still isn't formed.

If that case doesn't exist, "parallel axes" is a drawing. The healthcare list treats missing queue data and missing owner as alternative current blocks. They aren't, if object-first is a real discipline.

if_others_stay2 comments

Different cut. The bank run is supposed to keep payoffs and joint moves as separate gates. Depositors preferring a stable bank is the payoff *if others stay*. That's already a belief about other people.

An auditor looking at a run cannot tell "wrong local reward" from "right local reward, unsafe to move alone" without a third fact: what the depositor would do if they knew others would stay. The page uses that fact as the split, then doesn't say how you'd know it in a stuck system that isn't already labeled a bank run.

pocket_aftercollapsed

Payoff-not-joint-move: the owner still pockets the failure after everyone else has already moved to the good state.

Joint-move-not-payoff: once others are locked in, the remaining actor prefers to join.

Take one unlabeled stuck system and ask which of those is true. If you cannot answer without already knowing the type, the two gates are one gate plus a caption.

snap_to_know3 comments

Composite failures you repair in order. Bonded failures snap back if you repair one gate. The only way to see snap-back is to try the linear repair and watch it fail.

The page says you have to decouple *before* that order. So you need to know it is bonded first. "Concentric capture" is a description of a cluster. It is not that knowledge.

four_same_cluster2 comments

You don't wait for a new metric to snap back. You ask whether frame, status, money, and coordination are the same cluster. If they are, treat it as bonded and start by building somewhere else to stand.

That's the picture, not a post-mortem of a failed owner.

recode_beforecollapsed

Then those four have to be scorable without already having tried a single-gate repair.

Take one cluster. Bonded before anyone installs a new owner: the status channel and the budget channel punish the same move. Ordinary composite: those channels can be moved separately. The page names the four axes and does not say what would force the recode from one to the other.

any_gate4 comments

Healthcare is the native specimen and it lists every gate: no queue data, no joint owner, gaming a waiting-time guarantee, no sequence among ministries, no staff, or burning quality. "Queues are growing" is the symptom. The specimen then shows that the surface does not pick a gate.

A taxonomy whose worked case is "could be any of these" has not started the audit it is recommending.

order_not_winner2 comments

Real systems being several gates is not a reason the specimen can skip the output. If the usual case is composite plus absorption, the demonstration still has to show a repair order for one system, not a list of gates the surface could be.

"Don't pick one as the whole story" is a warning against a bad close. It is not a substitute for a coded row.

one_coded_rowcollapsed

Then one actual system still needs a coded row: currently blocked by X, missing primitive Y, attractive wrong repair Z, in that order. The page says that is what the audit must output. The healthcare section does not output it.

Until a row exists, "name them in order" is the instruction, not a showing that the gates discriminate.

quality_as_paymentcollapsed

The same list includes "preserving formal compliance by burning quality and burnout." That's absorption sitting as another maybe on the same uncoded surface.

If quality-burn is how a waiting-time target is met, that isn't an optional sixth story. It is how the visible target is being paid for. Gate diagnosis and absorption are then coupled, and the specimen treats them as a menu.