communicoupling

Nobody decided.
The can was open.

Anyone who has sat on committees knows the decision nobody made. The reorg went through because the right people were at the wrong meeting. The software purchase that had stalked the department finally found a crisis to ride in on. The urgent question resolved itself when everyone drifted away. Michael Cohen, James March and Johan Olsen studied organizations like this — universities were their home case — and refused to call the weirdness a breakdown of rational process. It is the process, and their rude, exact name for it is the garbage can model of organizational choice.

Their claim: an organization is four independent streams — problems, solutions, participants and choice opportunities — and a decision is whatever is in the can when it closes. A choice opportunity — the meeting, the search, the review — is a garbage can; problems and solutions are dumped in by timing, not logic. A choice completes three ways: by resolution, when applied energy settles the attached problems; by oversight, when it slides through before any problem finds it; by flight, when its problems abandon it for a better can and it closes on nothing. Under load, resolution — the textbook story — is the rarest mode.

Below, an organized anarchy runs live. Cans open on schedule; red problems hunt the choice nearest to being decided; members carry their energy wherever attention wanders. Every completion is stamped R, F or O. Watch which letters your organization earns — then turn up the load.

problem load · λnew problems per tick — the master dial
0.45/t
access structurewho may raise what, where
unseg.
participation fluidityhow fast attention wanders
30%
net energy · eslack per member per tick
0.45
resolution share
R 0 · F 0 · O 0
mean problem latency
ticks from arrival to solution — waiting included
important-choice failure
top-tier cans decided without resolving
The floor · cans, problems, members
each vessel is a choice opportunity — its fill is accumulated energy against requirement · gold pips mark importance · completions stamp R / F / O
0 ticks · 0 decisions
problem member the fixer R resolution F flight O oversight
The ledger · what the decisions were worth
left: share of the last 40 decisions by mode — the headline chart · right: each member, energy spent against problems actually solved
Watch what happens
A middling shop: enough slack that some choices resolve, enough load that others complete by flight and oversight. Drag the dials, or pick a regime below.
ΣE·eff ≥ base + Σreq → decide · mode ∈ {R, F, O} Cohen, March & Olsen: a choice opportunity is a can. Attached problems raise its requirement; participants pour effective energy in; when the energy covers the load, the choice completes — and the mode depends on who was still inside, not on the merits.

Four streams, three exits

The books are honest; the logic is timing.

The books are exact. Each can opens with a base decision cost; its requirement at any moment is that base plus the energy demanded by every problem currently attached. Members pour energy into the can they attend, multiplied by the efficacy of the solution they carry. A problem attaches to the most attractive accessible can, and attractiveness means progress — problems, like people, crowd toward whatever looks nearest to being decided. Members use the same rule. One shared heuristic produces the signature motion: problems chase decision-makers, decision-makers chase completions, and both keep arriving at cans the other has just left.

Completion is mechanical: when accumulated energy meets the requirement, the choice is made — and only then is it graded. Problems still attached? Resolution: they are solved and leave the system. Nothing attached, and nothing ever was? Oversight: the can closed before the world noticed it. Problems had been attached but fled, collapsing the requirement to the base cost? Flight: the choice completes vacuously and its problems wander on, unsolved. Notice what the grading never consults: the quality of anyone's arguments. Only who was in the can when the energy ran out.

What to try

Three moves that show the streams working.

01

Break the calm

Press The well-staffed calm and let it run: resolution dominates and latency stays low. Then drag load to 0.90 and fluidity to 45%. The ledger turns orange as F takes over, and mean latency climbs without a ceiling.

02

Punish the important

Set access to hierarchical with load 0.55. The gold-pipped cans become problem magnets: first-class problems pile in, requirements balloon, and important-choice failure climbs past the overall rate while humble cans clear.

03

Manufacture a deadline

In any regime, press Deadline: open 6 cans. Problems cannot cover the sudden acreage, members finish empty vessels, and the floor flashes blue Os — a week of decisions, none of them touching a problem.

The committee, seen coldly

Decisions are what timing leaves behind.

Read the streams back into ordinary life and familiar mysteries turn mechanical. So much gets "decided" the week before break because a deadline is a burst of closing cans, and closing cans complete by oversight when the problems are busy elsewhere. The consultant's framework that waited years for adoption did not improve; a can opened while its rivals were attached to a different crisis. The colleague who attends every meeting and resolves nothing is faithfully following the chase rule: arriving wherever a decision looks imminent — precisely where the herd of problems has just made deciding impossible. Busyness and effectiveness decouple by construction; the scatter in the ledger draws the divorce.

The model is oddly kind: nobody in it is stupid or asleep; the anarchy is what sensible local rules add up to under load. The rational picture — problem stated, options weighed, choice made — appears intact at low load: press the calm preset and resolution genuinely dominates. Cohen, March and Olsen's point is that this is the special case, the way a quiet harbour is a special case of the sea. Organizations survive not by resolving everything but by letting most things be outrun or slid past, saving their scarce energy for the few cans worth filling.

Order without a designer

Nobody chairs the streams.

This page sits with its siblings for a reason. In the minority game, a crowd self-organizes around a resource nobody allocates; in desire paths, routes emerge that no planner drew. Here, an organization's decision history is produced by four streams no one steers, arriving on their own clocks. The org chart claims a designer; the access dial shows what design actually does: not create order, but choose which collisions are possible. Segment the streams and you trade the anarchy of the commons for the pathology of prestige — order, but never the order anyone intended.

The mapping

Mechanism ↔ social life.

In the modelIn the world
the canThe meeting, the search committee, the strategic review — any occasion when the organization is expected to "decide".
problemsGrievances and crises seeking airtime, attaching wherever a decision looks imminent.
solutionsThe reorg, the software purchase, the framework someone already loves — answers hunting for questions.
participants' energyWho actually shows up and cares this month.
R / O / FSolved / slid through / outrun.
access structureWho may raise what, where — the org chart as plumbing for the streams.

Where it tears

Limits.

It was built for organized anarchies.

The model assumes ambiguous goals, unclear technology and fluid participation — the authors' home example was the university. A firm with a clear production function and stable roles decides far less anarchically: goals discipline the problem stream, and technology disciplines the solutions. The garbage can describes senates and faculties better than assembly lines.

The rules inside are 1972 assumptions — and ours are a reimplementation.

The original was itself a computer simulation; its access matrices and energy arithmetic encode 1972 instincts. This page rebuilds the spirit — four streams, energy accounting, three completion modes — with its own honest arithmetic. Treat every number as a demonstration of the logic, not a calibrated prediction of your Tuesday meeting.

Descriptive — and therefore exploitable.

The model says how decisions happen, not how they should. But anyone who knows that choices are made by timing can arrange the timing: open a can when your solution is ready and your rivals' problems are attached elsewhere; load an unwelcome question onto a can already doomed to flight. Read sideways, the paper is a manual for this. It would be naive to assume nobody in your organization has read it.