Assess Denver, and freeze the Stage 1 inventories for replication (#96, #59) - #97

Open

jonfroehlich wants to merge 33 commits into

mainfrom

feat/location-precision-96

Commits on Jul 31, 2026

Assess Denver, and freeze the inventories so Stage 1 can be replicated (#96, #59)

jonfroehlich
and
claude
committed
87d903e
Sweep the geometry gate across all nine cities -- and correct the Denver claim (#96)

jonfroehlich
and
claude
committed
7279959
Make the review sheet usable: toggleable overlay, 4x imagery, click-to-measure (#96)

jonfroehlich
and
claude
committed
259407c
Verify the review sheet is registered and scaled, rather than asserting it (#96)

jonfroehlich
and
claude
committed
d482fef
Put the review rubric inside the instrument, not in the reviewer's head (#96)

jonfroehlich
and
claude
committed
9251977
Prove the review sheet's page logic runs, not just that it contains strings (#96)

jonfroehlich
and
claude
committed
6058bf1
Say what the 2016 imagery choice costs, not just what it buys (#96)

jonfroehlich
and
claude
committed
1331e60
Stop the per-corner comparison crying wolf on large corners (#96)

jonfroehlich
and
claude
committed
3f95381
Show the published records instead of judging them (#96)

jonfroehlich
and
claude
committed
f2b0599
Stamp the sheet with a build id, and make the markers findable (#96)

jonfroehlich
and
claude
committed
fb7128e
Put the build id where it can be seen (#96)

jonfroehlich
and
claude
committed
186a76b
Hold the published data until BOTH of the chip's own numbers are recorded (#96)

jonfroehlich
and
claude
committed
000a7a8
Record what the review sheet structurally cannot measure (#96)

jonfroehlich
and
claude
committed
7ddd831
Jon's Denver review, 58/59, and the reduction that makes it a number (#96)

jonfroehlich
and
claude
committed
53239c2
Denver is Good: 92.3% of records within 1 m of the ramp (#96, #59)

jonfroehlich
and
claude
committed
022bc6d
Stage 1 uses only the BEARING, so the tolerance is angular (#96, #59)

jonfroehlich
and
claude
committed
e53d225
Probe the basemaps for the two calibration cities (#96)

jonfroehlich
and
claude
committed
63cc7ca
Freeze Seattle, and catch its basemap problem before booking a reviewer (#96)

jonfroehlich
and
claude
committed
b01a09c
Seattle's offset is a SHIFT, not imprecision -- caught by Jon, now automatic (#96)

jonfroehlich
and
claude
committed
dea603c
Seattle's 2.06 m is not a registration error -- both frames clear at high n (#96)

jonfroehlich
and
claude
committed
7a482b8
Don't track the chips run's duplicate registration PNGs (#96)

jonfroehlich
and
claude
committed
38bb0a3

Commits on Aug 2, 2026

Stage 1 records its own registration error -- no gov files needed (#96)

jonfroehlich
and
claude
committed
56db3e8
Rebuild Charlotte's sheet: 1 of 60 chips becomes 60 of 60 (#96)

jonfroehlich
and
claude
committed
d17c5d6
Record the Charlotte 404 trap in the doc, not just the commit log (#96)

jonfroehlich
and
claude
committed
e554c5e

Commits on Aug 3, 2026

Charlotte is Good, and its bad records carry a null date (#96)

jonfroehlich
and
claude
committed
c06da5c
Charlotte's review is HIGH confidence, and the value travels with the data (#96)

jonfroehlich
and
claude
committed
17f7f4c
Grade a basemap where it will be USED, not at one downtown point (#96)

jonfroehlich
and
claude
committed
3c9dc33
Sample the review sheet by install-date stratum, not by whatever a city publishes (#96)

jonfroehlich
and
claude
committed
78f0279
Seattle's stratified 2025 sheet: 60 chips, 20 per date stratum (#96)

jonfroehlich
and
claude
committed
af6a97d
Seattle is imprecise, not confounded -- and Charlotte's rule does not generalise (#96)

jonfroehlich
and
claude
committed
3fe720c
Seattle's 2025 review is MEDIUM confidence (#96)

jonfroehlich
and
claude
committed
c9ce711
Supply is not the constraint, and never was: ~2M candidate records (#96)

jonfroehlich
and
claude
committed
479a436
Write the 5n I promised, and file the tool as #103 (#96)

jonfroehlich
and
claude
committed
d03c3e5