Tracing a result
Where each number comes from, how it was checked, and what has changed
Every result in Situ should lead back to the files and settings that made it, and to the checks it passed. This page follows that path for one value. It then sets out which kinds of checking Situ has done so far and what has been revised, and ends with what can be shared.
Four kinds of statement
A sentence on these pages is one of four kinds. The kind says where a statement comes from. The status words (verified, partial, planned, assumption) say how far it is supported.
| Kind | What it is | Example |
|---|---|---|
| source observation | what a published file or a photo records | the city’s Géobase draws Rue Wellington as a secondary arterial |
| computed result | what my code or a tool calculates from sources | S101’s walking choice at 400 m is 1.053 |
| researcher decision | a choice I made and wrote down, usually after looking at photos | the footway behind Brennan and Ann is walkable (C04) |
| assumption | a reading, or a rule chosen because no source settles it | a street with high choice may carry more walkers |
The decision files (S1, S3) keep every decision with its evidence. The rules S9 runs on are listed in its ledger.
One result, traced
Segment S101 is a 76 m stretch of Rue Wellington. Its walking choice at 400 m is 1.053. These records follow that value back to the city’s file. The full records, with every field, are in evidence_records.csv.
| Step | Kind | What happened | Checked by | Status |
|---|---|---|---|---|
| T1 | source observation | the city’s Géobase street lines, saved 20 September 2026 (files dated 17 September) | photo checks in Griffintown, C01 to C19 | verified |
| T2 | researcher decision | my merge script adds paths and connectors and removes highways: 47,197 lines | the merge report’s checks C1 to C6 | verified |
| T3 | computed result | PlanX computes integration and choice on that network, 10 October 2026 | line count and shapes; PlanX’s formula and radius checked against my own counts (Reading closely, Checking the sources) | partial: PlanX 4.19.0; its settings beyond the radii are not kept in a file |
| T4 | computed result | S2 lays line 9140 under S101 (76.41 m shared) and copies its values | S2’s six passing tests and the Lab’s check L2 | verified |
| T5 | assumption | choice may relate to how many people walk a street | S9: the imagined morning agrees with it only weakly (rank correlation 0.24 to 0.29) | assumption |
Which file, which version
The data I download is not kept in a public archive, so a file name alone cannot prove which copy a system read. Each system therefore writes a run record next to its results: S1, S2, S3, S4, S5, S6, S7, S8 and S9. A run record lists every input file with its size, date and SHA-256 fingerprint, every setting the code reads, the fingerprint of each code file, the systems it builds on, the fingerprint of each output, and the test results. Two copies of a file with the same fingerprint are the same file.
Since 21 October 2026 the text results are also kept in the project’s private history, so each published version of a result can be found again with the code that made it.
Three kinds of checking
A passing test says the code does what I meant. It says nothing about whether Griffintown works that way. So I keep three kinds of checking apart.
- Implementation: does the code do what I meant? Each system’s tests and stress runs, checked again by the Lab. A test either passes or fails; a few only report a number, and the counts below leave those out.
- Correspondence: do the results match something observed independently? Photos, or a second source such as the city’s counts.
- Interpretation: does the result mean, to the people who live and move here, what I take it to mean? Only time with people, on site and in play, can answer this.
| System | Implementation | Correspondence | Interpretation |
|---|---|---|---|
| Walking network | merge checks C1 to C6 | photo checks in Griffintown only | planned |
| S1 Streets | 6 tests passing, Lab L1 and L5 | 19 places checked on photos | planned |
| S2 Scores | 6 tests passing, stress run, Lab L2 | none yet beyond S9’s comparison | planned |
| S3 Rules | 8 tests passing, stress runs, Lab L3 and L4 | signs checked on photos at chosen junctions | planned |
| S4 Buildings | 9 tests passing, stress run, Lab L6 | homes within about 1 % of a provincial land-use file (2026 roll), checked in aggregate | planned |
| S5 Rhythm | 12 tests passing, stress run, Lab L7 | the counts’ bicycle shape against the bicycle detectors (correlation 0.89) | planned |
| S6 Streetscape | 7 tests passing, stress run, Lab L8 | photo detections against the city’s hydrants: 68 of 104 found | planned |
| S7 Commuters | 6 tests passing, stress run, Lab L9 | none yet | planned |
| S8 Views | 6 tests passing, stress runs, Lab L10 and L12 | none yet | planned |
| S9 Observer | 9 tests passing, stress runs, Lab L11 | the city’s counts at seven junctions and at 23 sites around the area | planned |
The interpretation column is empty on purpose. It needs people, and Situ has not yet asked anyone.
What has changed
Nothing is overwritten without a note. When a result changes, the older one stays where it was described, and the page that uses it says which one it uses.
| When | What changed | What it affected | Where it is described |
|---|---|---|---|
| September 2026 | Sample 4: the first space syntax values, on the city’s streets alone | kept for reference, never mixed with the new values | S2 |
| 10 October 2026 | the merged walking network and a new PlanX run; the first run that day used a faulty link (C06) and was redone | every walking value in S2 | The walking network |
| 10 October 2026 | the driving network rebuilt without pedestrian ways | 31 segments closed to cars lost a vehicle value they should never have had | S2, The driving network |
| October 2026 | S1 checked on photos: 190 segments and 12 dead ends became 264 segments and 2 dead ends | every later system | S1 Streets, Checks and decisions |
| October 2026 | test failures fixed: S2 stored shared metres too coarsely, S4 counted 89 properties twice, S6 rounded outlines before measuring them | the files named; each fix is described on its page | S2, S4 build, S6 build |
| 21 October 2026 | S2 writes its trace table with a fixed column order; values unchanged | none | this page |
| 21 October 2026 | run records added, and text results kept in the private history | none; it records what exists | this page |