Every Angstraum brief an office answers reaches for the same three moves: open the blind wall, scrub the graffiti, plant something at its base. On 31 August 2026, Nature Cities published the first study to randomise all three and measure what each one actually buys. Pablo Navarrete-Hernandez (University of Chile) with Tom Schlesser and Davide Luca (Cambridge Land Economy) ran a controlled online experiment — 18,150 perceived-safety ratings from 2,028 Chilean respondents, four space types, one prompt: “Imagine you are walking alone. How safe from crime do you feel?” Peer-reviewed by William McClanahan and David Mitre-Becerril, CC-BY, replication files on Zenodo. As of last week its Altmetric score was 1 and no outlet had run it. That gap is why this desk is running it now.

←TODAY: In 2026 a blind ground-floor wall is still argued as one opening percentage, reconciled after the fact by the façade consultant. →3012: By the Zurich-3012 horizon the perceived-safety curve of every public elevation is a queryable layer — dosed, and read differently by whom it serves. Fulcrum: The first opening buys the most, and more than half a wall gives some back — visible only if you plot the whole curve, never the endpoint.

The three fixes do not share a shape, and the shapes are the story.

• Windows — a curve that bends back. This is the oldest lever in the book: Jane Jacobs’ “eyes on the street” from The Death and Life of Great American Cities. It is significant at the smallest dose tested, 12.5% of the wall (Bonferroni-adjusted P=0.010) — then the lift flattens and, in the paper’s words, “is not linear, even decreasing at the 50% intervention threshold.” Their term for why is overviewing: past a point, more eyes read as surveillance, not safety. Jacobs complicated by her own logic.
• Graffiti removal — a step, not a curve. Nothing at 12.5%, then a jump at 25% that simply holds: “no effect before and a plateau thereafter.” You clear the tags or you don’t; there is no gradient to tune.
• Planting — the strongest average lift, 22.5% above the baseline safety level (coefficient 0.842), against 14.0% for windows and 13.6% for graffiti removal. Those are shares of the untreated safety reading, not shares of people.

One caution the authors raise themselves: intensity is measured as percentage of pixels changed, and “this metric may not capture perceptually equivalent changes across intervention types.” Windows at 25% and planting at 25% are the same pixel share, not the same change to a human eye. Compare the three with that in hand — and remember these are stated ratings on static photos, one online panel in one country.

Then the finding that makes it an article. At the untreated wall, men and women rate safety identically (P=0.509). Every intervention then lifts men more, significantly — windows P<0.001, graffiti P=0.001, planting P<0.001 — and the gap is widest on the planting. So the move the brief reaches for first is the one that least closes the gap the brief was written to close. The authors’ reading is careful and right: women’s insecurity is anchored in an assessment of vulnerability to gender-specific crimes “that environmental modifications may not adequately address.” That is an argument for also doing what a façade cannot — lighting, sightlines, staffing, reporting — never for doing less.

Atelier: A blind ground-floor wall on a public route is a live line item in every Vorprüfung and every ground-floor programme argument. Stop negotiating openings as one percentage reconciled later, and start asking what the first tranche buys against the fourth — the honest read being that a small, well-placed opening budget appears to do most of the perceived-safety work while pushing past half the wall is measurably counterproductive, but on a Chilean photo panel, so a hypothesis to test here, not a number to import. Monday move: pull Stadt Zürich’s own Sicherheitsempfinden series from the Bevölkerungsbefragung and mark which live projects sit on the four space types this paper simulates — underpass, stair, square, street.

Hack: Re-fit the dosage curve from the authors’ own archive rather than trusting the abstract. The Zenodo release (DOI 10.5281/zenodo.21061449, CC-BY, no login) ships URBANIAPS_Nature Cities.dta; a few lines of pandas and statsmodels put the male/female split back under your own eyes. Adjust the column names to the codebook, then read the interaction terms.

import pandas as pd, statsmodels.formula.api as smf
df = pd.read_stata("URBANIAPS_Nature Cities.dta")
m = smf.mixedlm("safety ~ C(dosage)*strategy + gender", df, groups=df["respondent"]).fit()
print(m.summary())

Download the .dta, re-plot the split, and bring one question to your next design review: for this wall, which shape are we buying — the curve, the step, or the lift — and for whom does it work? The paper hands you a dosage curve instead of a slogan. Treat it as one.

Sources & Further Reading

• Primary: Nature Cities — Urban design interventions reduce fear of crime in public spaces in gender-sensitive ways
• Reinforcing: Zenodo — replication files (.dta / .do)