Settle the r≈2 mass per roll: half is a wrong scale, not a mislabelled lens (#72)

[NewGraphEnvironment]

Summary

• Half the r ≈ 2 mass was a wrong scale, not a mislabelled lens. The logbooks and adjacent-frame spacing were read per roll-height over all 252 sampled frames beyond the band. 24 roll-heights (120 frames) carry the height the crew flew beside a scale recorded too small. They now ship in flying_height_rolls.csv at factor 1 (scale_wrong) under a new tail = "near_upper". Their keys reach at most 3,227 catalogue frames. The nominal fallback drew these frames at 1/r of their width (half at r = 2); they are now corrected_roll_table and sized from their height.
• On 21 roll-heights (82 frames) the logbook writes a 12" lens. That confirms the lens reading there with a witness spacing never saw (bc78051, bc79072 and bc80122 are among them), and those frames stay on nominal scale. Every roll-height left out is listed in flying_height_rolls_excluded.csv with its reason.
• fly_footprint() needs no logic change: the table lookup already works on both sides of the band. Only comments and roxygen change, along with the note, NEWS and CLAUDE.md.

Related Issues

• Closes Half the r≈2 mass is a wrong scale, not a mislabelled lens, and the fallback draws it at half width #72

The rule, which is the load-bearing part

The plan review found that the logbook height cannot separate the two defects. A wrong lens and a wrong scale both mean the crew flew the catalogued height, so factor 1 "agrees" on lens rolls too. For this tail the rule therefore also requires spacing to reject nominal scale, not merely fit the reported height. On a lens roll the reported height implies about 0.80 overlap, against a window top of 0.78, so fitting alone would have accepted lens rolls. This was fixed before the run.

A legible logbook scale equal to the catalogue's would veto a row. No sampled page writes a scale, so the veto never fired.

Logbooks

• 144 pages were fetched for 53 rolls. One page (bc80048_4.jpg) returns 404 upstream.
• Three unnamed readers transcribed them blind, given only the images and one control sheet each. All three controls read back as catalogued, and I spot-checked two sheets against the images myself.
• The logbook CSV gains 185 rows. Two kinds of row were added at merge, and each says so in its note:
  • 18 whole-feet heights on 1984–86 sheets;
  • 2 interior-strip rows.
• focal_mm on these pages is filled only where a focal length is written, never inferred from a camera model, because on this set the lens is the question.

Review

• Plan review: 15 findings, recorded in the archive. Three of them (A1–A3) were folded into the rule before the run.
• Code-check, 3 rounds:
  • Round 1: the reach claim had terrain's direction backwards; the scale parser read scales out of remarks.
  • Round 2: the parser fix still swallowed trailing digits, a defect inside a fix; the bc80048 wording was wrong; bc7223 shipped as "no page" when its page covers it but its height is unread.
  • Round 3: named the mechanism (claims written from a picture of the data rather than recomputed) and enumerated all 43 claims in the new prose. Two were false, and both were fixed and verified directly:
    • NEWS row counts went stale after the round-2 fix.
    • The bound paragraph missed 1,534 frames on nine 1985 rolls that sit just under the sampled stratum's floor. They are likely the same class and are unmeasured; this is stated in the note.

Test plan

• devtools::test(): FAIL 0 | SKIP 0 | PASS 2537
• The generator reruns byte-identically. Lower and upper rows of both tables are unchanged; the diff adds near_upper rows only.
• Restore-the-bug: with the near_upper rows removed from the table (mocked), the bc5509 frame falls back to implausible at nominal width.
• pkgdown::check_pkgdown() clean

Version bump and NEWS heading are left for /gh-pr-merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_0115csb82i78r86GWFzwPzis