Cities: plan and place

The plan-place gap, AI in participation, and candidate first-study places.

Planning theory holds most of what the Cities method needs on the gap between plan and place. Alexander and Faludi separate three views of a plan, each with its own idea of failure. Healey asks what a strategic plan does, not what it says. Talen, and Laurian and colleagues, show that implementation is rarely evaluated and that the worked method compares plans with permits, not with the walked place. Nordic work adds that planning offices keep some tensions unspoken, which bears on how a Test will be received. On AI in participation the evidence is narrow and mostly about reading consultation text: PlanAI reports a 90 per cent efficiency gain but no accuracy metric, and the Greater London Authority has authorised it for the draft London Plan consultation; a Swiss multilingual note finds 90 per cent accuracy but 70 per cent completeness on substance; a 2026 preprint treats disagreement between models as the signal. Simulated residents are scored by the authors' own metrics, the design Holagram rules out. In practice Tallinn offers a live document stream: a new general plan (initiation sent to the council on 12 May 2026; ideas survey closed 31 August), the Linnahall detailed plan of 2017 with the city's strategy centre due to report this autumn, and plans in progress in Lasnamäe and Ülemiste. Helsinki offers a city plan built on one-hectare cells and a site under construction at Malmi. No work was found that ties a planning text to a walked place by method.

Sources

What it implies for the proposals

Ideas to take

Build the first study as a genre ladder around one place: a terms of reference or general plan (territory), a detailed plan (site), a strategy or investor presentation (institution), and the walked street (street and body). Each Entry opens with a header that names the document's genre and which of Alexander and Faludi's three views of a plan applies, so that what 'broke' means is fixed before the walk.

Changes: Adds a genre-and-view field to the Trace and Test registers. It stops one Test being applied across genres, and it answers the planner's first objection (a framework is not a promise) inside the method instead of in a footnote.

Risk: The authors of a plan may not agree on its genre (framework or binding), and a reader may pick the view that makes the Test come out as wanted. Have the outside reader set the view before the walk, and log any disagreement about genre as data.

Put a documentary step between plan and place. For one site, list the permits, design conditions and council decisions issued under each plan and compare policy text with permit (after Laurian and colleagues) before any walk. The walk then tests the permit-to-place step and the plan-to-permit step separately, by different hands.

Changes: Splits 'text against place' into two smaller Tests and gives the Trace register a routine source that is not a plan. An unbuilt or lapsed plan becomes a result, not a failure of the method.

Risk: Permit records may not be open or easy to read in Estonian or Finnish, and the step can grow into a conformance study that eats the three weeks. Cap it at one site and one afternoon per plan. Linnahall has had little built, so use Ülemiste or Lasnamäe for this idea.

Treat lapsed plans as their own class of Trace. Tallinn ended 52 stalled detailed-plan procedures in two weeks in November 2023, 12 of them in Lasnamäe, and said in April 2025 that about 100 old ones remained. For a sample of closed Lasnamäe procedures, set the plan text against the plot as it stands: what did the document promise, and what is there?

Changes: Gives the method a plan-without-place case, the cleanest form of the text-place gap, which needs no judgement about anyone's intentions. Adds 'never built' as a documented result beside held, broke and unchecked.

Risk: A closed procedure may have been replaced by design conditions or a later plan, so an empty plot is not a break. Check the register for successors before writing the Reading. Plots are privately owned, so use public documents and the street view only.

Give the consultation record its own set of Entries. Take the ideas-survey responses if the city releases them (the published summary if not), have three or more models code themes in each language, keep every divergence in the log, and have a person who did not run the models recode a stratified sample of about 40 and compare both with the city's own summary.

Changes: Gives the Adversary a measurable job and makes disagreement between models a recorded kind of data. The human recode is the Test, run by different hands from the Reading, and the official summary is itself treated as a document with a view.

Risk: The raw responses may not be public and may contain personal data, so the study may be limited to the summary. Greater Cambridge's opt-out for respondents (about 5 per cent used it) is a precedent to ask about before any model touches free text (cities-13). The benchmark papers were team efforts, so this can eat the three weeks. The author's company builds participation software; models and prompts should be neutral and published, and the interest declared in the Entry header.

Test a government programme claim at the street. The Interreg page (30 April 2026) records the city's account of its participation practice, including on-site 3D displays with QR codes at plan locations. Visit a fixed sample of sites with active plan procedures (Linnahall, Mustakivi tee, the Ülemiste Keskus area) and record whether a notice exists, in which languages, whether the code opens, and what it shows.

Changes: Shows how the Cities method tests an innovation programme's claims, and gives the Mandates method a worked model: a claim in a document, conduct on the ground. Uses fixed-time observation for the street level.

Risk: The claim may apply to only some plans, and notices come down after the display period, so absence is not a break. Fix the sample and the display dates from the register first, and accept 'unchecked' where the period has ended.

Write three Entries about single one-hectare cells of Helsinki's city-plan grid: one where the code fits the ground, one where it does not, one at a boundary. The Trace is the cell code and legend; the Reading is what the code assumes (one main use; detail left to detailed plans); the Fiction is the cell twenty years on under that code; the Test is a counted walk at fixed times and a short conversation. Tallinn's planning office says it will use a similar grid.

Changes: A small repeatable unit of text against place that does not depend on a contested decision, and a direct contribution to a method Tallinn is about to adopt. It tests a text that generalises on purpose.

Risk: The plan says boundaries are settled in detailed planning, so a mismatch is not a break: a category error unless the header names the view (first idea). If the researcher picks the cells, the choice is a Reading; draw them by a rule or by the outside reader.

Treat the Linnahall file as a moving document set. Fix a snapshot date at the start, give every new document (a council decision, a strategy-centre concept, a competition brief) and every physical change (the temporary public space) a new Entry instead of editing an old one, and add 'overtaken by events' as a reason in the failure log for Readings that a later decision makes moot.

Changes: Makes time part of the Entry: the ground and the text both move. Keeps the log honest about the difference between a Reading that broke on the ground and one that an event removed.

Risk: Adds a category to a protocol that already has three results; 'unchecked' plus a reason would do the same work. The autumn 2026 decisions may land mid-study, so plan the schedule around them.

Counter-arguments

Gaps