the second hundred · metaphor 206
There are two different questions you can ask about another person, and they are not the same question. The first: can you move them — with the words and warmth and pressure actually at your disposal, could you steer them from where they are to anywhere you'd want them to be? The second: can you read them — from the little that shows on the surface, could you ever infer the state they are hiding underneath?
These two freedoms come apart. Some people you can push anywhere and never understand; others you understand completely and cannot budge an inch. The system of him split in two along exactly this seam — whether he could be steered at all, and whether he could ever be read — and knowing the answer to one told you nothing about the other.
The levers you have are never the whole of a person. You get a few inputs — what you can say, offer, withhold — and a few windows onto what you can watch. Inside, their states are wired to each other in ways you didn't design: one mood drives another, a hidden ache leaks into a visible habit. Whether your levers can reach all of them, and whether your windows can reveal it, are two separate accidents of that wiring — and control theory gave each a name.
steer · controllability W_c = [ B | AB ]
read · observability W_o = [ C ; CA ]
The two matrices
Model the person as two coupled inner states, x₁ and x₂, that update in step: x → A·x + B·u, and what you observe is y = C·x. The matrix A is their inner wiring — here x₂ feeds into x₁ but not the reverse. The vector B is your lever: which state your one input actually pushes. The vector C is your window: which state your one reading actually sees.
Controllability asks whether, over time, your pushes can reach every direction of them. The test is a 2×2 matrix, the controllability matrix W_c = [B | AB]: if its rank is 2 — its determinant nonzero — you can steer them anywhere; if it collapses to rank 1, the reachable set collapses to a line, and everything off that line is forever out of reach. Observability asks the mirror question: does what you see pin down who they are inside? Its test is W_o = [C ; CA]. Full rank, and the readings determine the hidden state uniquely. Rank-deficient, and a whole direction of them moves without ever changing what you see — invisible.
The instrument computes both live from whatever B and C you choose — every rank, determinant, ellipse, and condition number read straight off the matrices at each click, nothing scripted.
What to notice
Change B and only the magenta shape moves; change C and only the cyan shape moves. The two yes/no chips flip independently — that independence is the lesson. Now find the surprises the coupling hides. Put your one input into x₂: you become fully controllable, able to steer the entire plane from a single lever, because x₂ spills into x₁. Put it into x₁ instead and you go rank-deficient — the reachable set snaps to a line, because nothing you do to x₁ ever reaches x₂.
Reading works the same, in mirror. Read x₁ alone and you can reconstruct both states — the hidden x₂ leaves its fingerprints in the x₁ you watch. Read x₂ alone and x₁ becomes an unobservable direction: the dashed cyan line. Drag the gold hidden-state point along it and its cyan twin follows — two genuinely different inner states producing the exact same reading, forever indistinguishable to you. The four preset corners set every combination. Watch the condition number, too: when it climbs into the tens, the direction is technically reachable but ruinously costly to steer.
The mapping
Your inputs are the levers you truly have on someone — a word, an incentive, your presence in a room. Your outputs are what of them you can truly see. Their coupling is how their inner states drive one another, wiring you never chose. From those three things follow two independent verdicts. The reachable set is the set of states you could actually steer them into — maybe the whole of them, maybe only a single line, with everything else off-limits no matter how hard you push. The unobservable direction is the part of their inner life that leaves no visible trace, that you could not detect if it swung to its extreme.
The insight the instrument makes physical is that these are separate. Being able to move someone is no evidence that you can know them; being able to read someone is no evidence that you can move them. The manipulator who can steer a person anywhere may understand nothing of what's underneath. The friend who reads every flicker may have no lever that reaches. And when the coupling is kind, a single well-aimed input or a single honest window can reach or reveal the whole — if you push and watch in the right place.
Read as life lessons
Influence and insight ride separate axes. You can hold the controls of a person and be blind to them, or see them entirely and hold nothing. Confusing the two is how people mistake power for understanding.
What matters is not how hard you push but where. In coupled systems the well-placed small input reaches everything and the clumsy big one reaches a line. The same is true of the window you choose to watch.
A whole part of someone can move without ever showing — an unobservable direction. It isn't hidden to spite you; it simply leaves no trace in anything you can measure. You'd never know it was there.
The mapping, exactly
| Mathematics | Life |
|---|---|
| input B | The levers you actually have on someone — the words, incentives, and presence you can bring to bear. |
| output C | What of them you can actually see — the surface signals you get to read. |
| coupling A | How their inner states drive one another — wiring you didn't design and can't change. |
| rank W_c = 2 | Your levers reach every part of them; rank 1 means they reach only a slice, a single line through who they are. |
| reachable set | The states you could actually steer them into — sometimes the whole plane, sometimes just that line. |
| rank W_o = 2 | What you see determines who they are inside; rank 1 means the readings can't tell certain states apart. |
| unobservable direction | The part of their inner life that leaves no visible trace — it can swing fully and you'd never see it. |
| near-singular Gramian | Technically possible, but ruinous — that direction can be moved or read only at astronomical cost. |
| rank deficiency | A whole direction of them beyond your reach, or beyond your sight, no matter the effort. |
Where the metaphor tears
Rank says yes or no; a person never does. A Gramian can be technically full-rank yet so ill-conditioned that steering one direction costs an order of magnitude more energy than another — the instrument shows a condition number of ten or more even when the answer is a clean "controllable." In a life, "can you move him" is almost never a yes/no; it's a question of at what cost, over what horizon, and whether you can bear to pay it. Naming the binary sharpens the real question: not whether you can steer, but how ruinously hard it would be.
Controllability and observability are computed from an assumed A, B, C — your theory of how the person is wired and what you can touch. If that theory is wrong, the state you confidently "cannot observe" may be the only one that matters, and the reachable set you drew is a fiction. The math is exact about a picture that may be nothing like the person. Its honesty is conditional on a guess.
The neat 2×2 plane is a local caricature. Real people are nonlinear — the same push helps at one dose and backfires at another — and time-varying: the lever that reached them last year is dead now, the window that was open has closed. Coupling that was one-way becomes mutual. The categories stay useful as a way of asking — can I reach this, can I see that — long after the tidy matrix stops being true.