Where the assumption was right

Eighty-six places out of two hundred and six were read correctly before anything was checked.

correctly expected no limit      63
correctly expected a limit       23
                                ---
                                 86

Forty-two per cent, and most of it is the first row.

The sixty-three read as open

Where are they? Overwhelmingly in the kinds that ask nothing at the door.

covered public seating    8 of 12
libraries                 8 of 12
places of worship         8 of 14
pubs and bars             8 of 31
transport concourses      6 of 9
museums and galleries     5 of 8
shopping centres          4 of 16
community halls           4 of 11
cafés and coffee shops    5 of 58
laundrettes               3 of 12
bookshops                 3 of 13
hotel lobbies             1 of 10

A bench, a library and a church are read as unlimited because they are unlimited and because everybody knows it. There is no mystery in the sixty-three.

The twenty-three read as limited

These are the interesting ones, because they are the only places where an expectation of restriction turned out to be correct. A transparent field pilot can test privacy-aware productivity reporting against the real administrative effort required to collect, correct and review the record.

cafés and coffee shops    13
shopping centres           3
pubs and bars              3
laundrettes                2
libraries                  1
community halls            1

Thirteen of the twenty-three are cafés. Fifteen of the twenty-three are written limits, which means a notice was expected and found. For independent context on inclusive places and field research, review material from Open Data Kit.

So the correct restriction-readings are mostly cases where something in the room signalled a rule, and the signal was right.

How much of the eighty-six is skill

Less than it looks, and the comparison that shows it is blunt.

Guessing no limit in every one of the two hundred and six places would have been right a hundred and sixty-eight times. Eighty-two per cent, against the actual forty-two.

So a person who ignored every impression and assumed they could always stay would have outperformed a person reading each room by forty points.

The guesses carry information and the bias swamps it. Reading the room was worse than refusing to.

That is the most uncomfortable sentence this record produces about its own instrument, and it is the fairest summary of what the assumption field found.

The information that is in there

It is not nothing, and the way to see it is to look within each decision.

assumed a limit (128)    23 had one      18%
assumed none (78)        63 had none     81%

When the guess was no limit it was right four times in five. When it was a limit it was right fewer than one time in five.

So the guessing is not symmetrically bad. One half of it is reliable and the other half is almost noise, and the trouble is that the unreliable half was used a hundred and twenty-eight times.

The asymmetry, stated as an instrument

The guesses are not uniformly bad. They are bad in one direction.

when the guess was "no limit" (78)     right 63    81%
when the guess was "a limit" (128)     right 23    18%

One half of the instrument is reliable and the other is close to noise, and the unreliable half was used in sixty-two per cent of places.

That is a specific, correctable fault rather than a general unreliability: a walker who simply never predicted a limit would have scored eighty-two per cent, and one who kept their no limit calls and dropped the rest would have scored better still.

What a better guesser would have needed

Not local knowledge. Nothing about the twenty-three correct limit-readings required knowing the city: fifteen of them were places where a notice was expected and found, which is a reading of a room rather than a memory of one.

What they would have needed is a rule about silence. A hundred and twenty places in the record state nothing about time, and every wrong guess in the walk was made into one of those silences.

Treat an absence of any statement as an absence of any rule, and the error rate falls from fifty-eight per cent to eighteen.

That is a rule somebody could apply in a city and it is not advice, because it is derived from one walk and could be wrong in a different one. The page on what this record refuses puts it among the five things that do not transfer.

The one group the guesses handled well

Places that ask nothing at the door.

free places (71)        right 48    68%
paying places (120)     right 31    26%
fee places (15)         right 7     47%

Where no money changes hands, the guesses were right more than twice as often. The counter degrades the instrument exactly as it degrades the belief, and by about the same amount.

What that means for anybody else

Nothing demonstrated, and that bears repeating at the end of a page full of percentages.

The assumption field belongs to one person. A record of what one walker expected, written down honestly under a rule that could not be bent afterwards, in two hundred and six places. It is a good record of that and evidence of nothing else.

Whether the same asymmetry — confident and wrong about restriction, quietly right about freedom — holds for anybody else is the first question on the list of what a second walk should ask, and this record cannot answer it.