From Model to Plan

Begin by feeling how wide the gap really is, because computation's polish tends to disguise it. A computational study is an artefact of *exploration*. Its whole nature is provisional: it embodies a set of assumptions, optimized for a chosen objective, on a particular dataset, and its status is 'here is what we found when we looked this way'. It can be wrong, partial or superseded, and nothing bad happens if it is - you run another. It carries no authority beyond the quality of its reasoning, and it binds nobody. Crucially, it is *unaccountable* in the formal sense: no one can be held answerable to a citizen for what a study proposed, because a study decides nothing.

A plan is a categorically different kind of thing. It is an instrument of *decision and law*. When a development authority sanctions a masterplan, that plan governs: it determines what may be built and forbidden across real land owned by real people, sets the development-control regime, and stands as the legal reference against which every future building permission is judged. It is not provisional; it is binding, and undoing it is slow, contested and costly. And unlike a study, it is *accountable* - there is a body, answerable to law and to the public, that made it and must own its consequences.

The danger in computational urbanism is the silent collapse of this distinction - the moment a study is treated as if it were already a plan. It happens easily: a generative masterplan is so detailed, so rendered, so confidently optimized that it *feels* decided, and the provisional 'here is an option' quietly hardens into the binding 'here is the plan', without anyone having consciously crossed the line. The polish does the collapsing. A hand sketch obviously invites revision; a photorealistic, data-backed model radiates finality, and finality is exactly the property a study must not have. So the first discipline of turning a model into a plan is to *keep them apart in your mind and your language*: to insist, at every step, that what you hold is still a study - evidence toward a decision that has not yet been made and does not belong to you - until it has actually passed through the process that alone can make it a plan. The gap is not a technicality. It is where democratic control over the city either lives or dies.

So how does a study *legitimately* cross the gap? Through a defined statutory pipeline, and in India that pipeline is a legal process run by a development or planning authority under state town-planning law. Understanding its stages shows exactly where computation belongs and where it must stop.

First, evidence and drafting. The authority studies the city and site - and here computation earns its keep, feeding the evidence base: reading data, generating and comparing structural options, testing scenarios, checking against the rules. Out of this a *draft* plan is prepared. Note the word draft: at this stage the computational study has informed a proposal, nothing more. Second, publication and notice. The draft development plan is published and made available for public inspection - the moment the plan leaves the expert room and enters the democratic one. Third, objections and suggestions. The affected public - residents, landowners, civic groups - may file objections and suggestions within a statutory period, and the authority is obliged to hear and consider them. This is the public's formal purchase on the plan, thin though it often is in practice, and it is precisely where the participation of the previous lesson must be made real. Fourth, consideration and modification. The authority weighs the objections and may modify the draft - the plan changes, or should, in response to the public. Fifth, sanction. The competent authority, often with state government approval, formally sanctions the plan, at which point it acquires the force of law. Only now is there a *plan*. Finally, the sanctioned plan is given buildable teeth by the applicable development-control regulations and the National Building Code of India, which translate its intent into the rules every future permit must meet.

The shape of this pipeline carries the whole argument. Computation is dense at the front - evidence, options, scenarios, rule-checking - and must be entirely absent from the operative act at the end. Sanction is not a computation; it is an accountable legal and political decision. The stages between - notice, objection, consideration, modification - exist precisely so that a public can bend the expert proposal before it becomes law. Skip or hollow them and you have not sped the process up; you have severed the plan from its only source of legitimacy. The pipeline is not bureaucracy to be optimized away. It is the machinery by which a proposal earns the right to bind people's lives.

Watch carefully what happens to a computational study as it moves down that pipeline, because the translation is where the honest planner does their most important work. What a study offers is *evidence and options*: this configuration performs this way, these are the trade-offs, here is a range of possibilities and their consequences. What survives into a good plan is exactly that - the study's findings enter the deliberation as informed argument, sharpening the choice the authority and public must make. What must *not* survive is the study's implicit claim to have already made the choice. The model's ranking, its 'optimal' output, its air of settled answer - these must be stripped away at the threshold, leaving the evidence and surrendering the decision.

This is why the translation is inherently *political*, not technical, and rightly so. Between the draft and the sanction, real negotiation happens: objections reveal interests the model never weighed, the public surfaces values no metric captured, competing claims are balanced, and the plan is amended - sometimes in ways that make it 'worse' by the model's score and better by the measure that actually matters, which is public legitimacy and justice. A plan that emerges unchanged from the process, identical to the computational study that seeded it, has almost certainly failed: it means the objection-and-consultation stage did no work, that the public's voice bent nothing, that the model effectively decided after all. A healthy translation *deforms* the study - it must, because the study could only see the measurable, and the process exists to admit everything the study could not.

The deeper point is that legitimacy is conferred by the process, not carried by the evidence. A brilliant study does not make a legitimate plan; passing through notice, objection, consideration and accountable sanction does. This is why the same computational output could become a legitimate plan in one process and an illegitimate imposition in another - the difference is not the quality of the model but whether the democratic machinery genuinely operated on it. So the planner's task in the translation is twofold: carry the study's real evidence faithfully into the deliberation so the decision is well-informed, and just as firmly refuse to let the study's polish pre-empt the deliberation, so the decision stays genuinely open to everything the model could not see. Evidence survives; the pretence of a decision must not.

All of this converges on one boundary, and it is worth stating in the plainest possible terms because it is the load-bearing principle of the whole module: the plan is decided by the authority and the democratic process, never by the model. A computational study can inform a plan, sharpen it, test it, and make the choice better understood - but the act of deciding, of turning a proposal into a binding law over people's lives, belongs irrevocably to accountable human institutions and the public they answer to. No model, however good, may cross that line, and no amount of optimization earns the right to.

The reason is not technophobia; it is the nature of legitimate power. A plan binds people, including people who disagree with it, and in a democracy the only thing that entitles anyone to wield that binding power is accountable authority exercised through a process the affected public could contest. A model has none of these properties. It cannot be held answerable to a displaced family; it cannot be voted out; it cannot be argued with in a hearing; it did not weigh the interests it never measured. To let it decide is to let an unaccountable artefact exercise accountable power - which is precisely the thing constitutional government exists to prevent. This is why 'the algorithm says' is not a justification but an evasion: it names no responsible decider, and a decision with no responsible decider is illegitimate no matter how good the algorithm.

In the Indian statutory frame this boundary is not merely ethical but legal: the master-plan or development-plan process vests the sanctioning power in a named authority, exercised under law, after public objection and consideration - and a computational study has no standing in that act whatsoever. It feeds the evidence; it does not sanction the plan. Every binding decision the plan contains - land use, rezoning, density, displacement, whose city this becomes - is deferred, absolutely, to that authority, that democratic and participatory process, the affected communities, and the governing planning law and development-control regulations. The honest computational urbanist internalises this until it is instinct: build the best evidence you can, argue it openly, improve the decision however you can - and then hand the decision, whole and undiminished, to the people and institutions who alone have the right to make it. That handover is not a limitation on computational urbanism. It is what keeps computational urbanism on the side of a free and just city.