Carry #54-slipped rolls a logbook reads at /10 in the roll table (#71)

[NewGraphEnvironment]

Summary

• FLYING_HEIGHT is 10.764x too large on 1,589 catalogue frames, so fly_footprint(dem =) sized them at 110 km #54's ÷10.764 is ÷10 wherever a logbook reads the slipped frames. The logbook pages for the five slipped rolls besides the control were read blind, with the catalogue values withheld. The crew's height is the catalogue's divided by 10 to within whole-metre rounding on bc78078, bc79103 and bcc00085 (all three heights), and 0.9% off on bc79027. On each of those, and on the control bc78065, the logbook sits 6.7–7.6% above ÷10.764. FLYING_HEIGHT is 10.764x too large on 1,589 catalogue frames, so fly_footprint(dem =) sized them at 110 km #54 was drawing these frames 7.6–9.9% narrow (15–19% in area), because width follows height above ground.
• Six roll-heights (299 frames, 1978–2000) are now in flying_height_rolls.csv at factor 0.1 (cause height_decimal_dropped), and both roll tables gain a tail column. The roll-table gate in fly_footprint() is now tab_factor != 1. The old > 1 silently handed a 0.1 row back to FLYING_HEIGHT is 10.764x too large on 1,589 catalogue frames, so fly_footprint(dem =) sized them at 110 km #54's repair, and a mocked test goes red under it.
• fly_height_slip_factor() is unchanged. The 1,290 slipped frames the table does not reach are each listed in flying_height_rolls_excluded.csv with a reason ending "FLYING_HEIGHT is 10.764x too large on 1,589 catalogue frames, so fly_footprint(dem =) sized them at 110 km #54's 10.764 still applies". The 2003 rolls have no logbook and fit 10.764 better. bcc05001 has no logbook either and is undecided: round feet favour ÷10, its ratio favours ÷10.764. bc79027 reads ÷10, but spacing rejects it; its own page logs 80% overlap against the ~60% window. bcb98013 frame 52 is a leading-digit typo and bc5596 has no page; both are filed as Two #54-slipped roll-heights a sibling frame settles and no logbook does #74.

Related Issues

• Closes fly#54's ÷10.764 looks like ÷10 on the pre-2000 slipped rolls #71
• Follow-up: Two #54-slipped roll-heights a sibling frame settles and no logbook does #74

Test plan

• devtools::test(): FAIL 0 | PASS 2348
• The mutation tab_factor > 1 turns the new gate test red (observed before the fix, and reproduced independently in code-check round 1)
• The calibration script reruns clean. Every control passes, the corrections reach exactly the frames they were measured on (1,300), and the lower-tail rows are identical to main on every original column
• End to end against the shipped table: a bc78065 1:2000 frame over 50 m ground gives corrected_roll_table, height_agl 361.5. A bcc03004 frame stays corrected_unit_slip
• /code-check, three rounds. None found a code bug. All findings were false set-level claims in prose ("pre-2000" where bcc00085 is 2000, ASL ratios quoted as width, "exactly" where one roll is 0.9% off). Rounds 2 and 3 enumerated every claim against the CSVs (27, then 39), and all now hold

Notes

• NEWS and the version bump are left to /gh-pr-merge. Suggested minor bump: a shipped-table schema change (tail), and 299 frames change width
• The reasoning is in the new fly#71 section of inst/notes/terrain-correction.md. The measurement record is in planning/archive/2026-09-issue-71-slip-factor-ten/README.md

🤖 Generated with Claude Code

https://claude.ai/code/session_0115csb82i78r86GWFzwPzis