What the System Cannot Yet Represent

Every formal account of human work leaves something out, and one that says what it leaves out can be used with its gaps in view. A system that claims to write every maneuver invites its user to force the ones it cannot write into a shape that fits, and the record then misstates what happened without anyone noticing. A system that names its gaps lets the user see where the record will fall short and decide what to do about it.

The Lam System for Political Interaction holds itself to a strict test here, which The Boundary of Computation sets out. A point where a maneuver could go more than one way belongs to the actor only if every legitimate way of proceeding can be written through it. When something a competent actor does cannot be written, whatever the actor chooses, the fault is the system's, and calling it discretion would hide it. The test produces a short list of moves the system cannot yet represent in full, and several essays in this series stop at one of them and point here.

Where the limits sit

A cycle with one lever, at one level, is written completely, from the opening to the flow back. So is a cycle with several levers on different resources, and so is escalation nested to any depth. The limits all sit where two of something meet: one lever played twice, a second attempt at one failed gate, two aims for one sub-cycle, two levers drawing on one resource, two resources missing at one gate.

None of them makes the record less trustworthy. Where a maneuver runs into a limit, the system either cannot write the move or writes what mechanically follows from it, and in either case what is written stays well-formed, every resource stays counted, and the maneuver still ends, because giving up is open at every point. What a limit costs is fidelity: the record says less than the actor did, or says it in a shape the actor would not have chosen.

Illustrative. The people and the company are invented. Leyla Demirci heads procurement at a food manufacturer and wants the operations director, Rohan Tessier, to approve moving the plants' maintenance contracts to a single provider. Her levers are a cost case built on her savings model, and the engineering function's sign-off on the new provider, which the head of engineering, Jun-seo Moon, has not given. Variations on her maneuver run into each of the limits below.

One lever, two sittings

Leyla puts the cost case to Rohan, and he asks to hear it again with the plant managers in the room. She means it as one lever, the same argument made twice, and each sitting spends some of her standing with him.

The system cannot yet write that. A lever's pass runs in a fixed order, and once the lever has been executed, what comes next for it is the target's reading. A second execution of the same lever has nowhere to go. Leyla can write the second sitting as a lever of its own, which records two commitments where she made one, or write Rohan's reading after the first sitting, which records a verdict he had not reached. The move looks like one the system forbids and actors plainly make, so under the strict test it is the system's to fix. It has been named and is not yet settled.

Trying again after a sub-cycle fails

The sign-off lever's gate fails, and Leyla opens a sub-cycle with Jun-seo. He declines, and she abandons the sub-cycle. Control returns to her cycle with Rohan, at the gate where it stopped, still short of the sign-off. She now wants another route to the same thing, through the group engineering director, who could sign in Jun-seo's place.

The system intends that move and cannot yet write it in full. From that point Leyla can set the lever aside if another lever can carry the cycle, moot the cycle, or give it up, and giving up is always available, so the maneuver is never stuck. Declaring a fresh lever aimed at the same need, from the position the failed sub-cycle leaves behind, is the step the rules do not yet admit. What it would look like is plain enough. Each new attempt would be a new lever with its own number and, if its gate fails, its own sub-cycle, so a lever would still never have more than one sub-cycle beneath it, and every attempt would count against the levers the actor decided to allow, which keeps the maneuver finite. Resumption, Mooting and Abandoning each stop at this point.

A sub-cycle that changes its aim

Suppose instead that halfway through the sub-cycle Leyla learns that Jun-seo will never sign and would agree to a trial run on one line. She is tempted to revise the sub-cycle's aim from the sign-off to the trial.

The system can write the revision, and what follows is mechanical. The sub-cycle closes holding a trial its parent never asked for, the handback check finds the sign-off still missing, and the parent is left at its failed gate, to be carried by another lever or given up. The counts stay exact. What the system has not settled is whether to allow the revision at all, and the answer depends on what kind of sub-cycle this is.

If it is an escalation, opened by the failed gate to fetch exactly what the parent lacked, its aim is pinned by definition, and changing it means it has stopped doing the one thing it exists for. Should Jun-seo need something before he will sign, that is a further escalation, one level down. If it is a pursuit Leyla chose to open beneath the parent for its own sake, its aim is hers to change, and it would hand back by some route other than the gate's retest. The system has no such second kind of sub-cycle yet. Adding one is the larger change, since the guarantees about counting and ending would have to be checked again for it. Revising Mid-Cycle takes no side, and neither does the system so far.

Two levers, one resource

Leyla adds a lever: a trial run Rohan can watch. It needs a slot in the one planned shutdown of the line the new provider would take on first, and the cost case needs the same slot, because she planned to demonstrate the savings there. If she holds one slot, playing either lever leaves the other's gate short.

The system can record what happens: the first lever spends the slot, and the second lever's gate fails. What it has not done is treat the choice of which lever spends the slot as one of the actor's forks. Several Levers in One Cycle shows that levers on different resources can be played in either order with the same result, and the guarantee about order rests on the resources being different. Here the order changes what the cycle can still do, so it is a real choice, and until the system names it as one, the guarantee stops short of this case.

Several things missing at once

The sign-off lever, as Leyla declared it, needs two things: Jun-seo's sign-off and the provider's audit report. She holds neither. The gate fails naming one of them, and the actor chooses which; a sub-cycle goes after that one, and the other comes to light when the gate tests again. The Capability Gate follows the case from the gate's side.

Which resource to fetch first can matter. The audit report may be what Jun-seo needs before he will sign, and fetching the sign-off first may waste a sub-cycle. The choice belongs to the actor, and the system lets her make it. What it has not yet done is enter that choice among the forks the order guarantee accounts for, so a maneuver that exercises it sits outside that guarantee as it stands. In the figure below, three of the five limits leave every guarantee standing, and two fall outside the guarantee that order matters only at the forks.

Named for later

Beyond these, the system names some larger extensions and leaves them undesigned. An executive often runs two pursuits at once, each with levers in play at the same moment, while within a single record the system keeps an agenda's pursuits in sequence, as Agenda explains. An actor whose errand fails sometimes gives up the whole maneuver in one decision, where the system has each level given up separately, as a decision written where it was made. Whether a named part of a maneuver behaves in every respect like what it stands for is left open by the sameness guarantee, which compares expressions only. And letting a name stand for a recurring pattern of moves, so that a maneuver can refer to the pattern, is named and not yet built.

What any extension has to keep

Every one of these will be judged against the rule that governs the record: nothing in it is overwritten, and whatever supersedes an entry is written beside it. A proposed move that would rewrite an earlier entry or an earlier binding of a name is refused on that ground alone, however much it would simplify the maneuver it was designed for. Whatever eventually fills the limits will add entries to the record and leave the existing ones where they are, and a maneuver written before the extension will read back unchanged after it.

An actor can check a maneuver against the list before writing it. Where a check below fails, the maneuver can still be written and run, and the actor knows in advance which part of the record will say less than happened, or which guarantee will not reach it.