Outcome (PR #75)
Confirmed, and settled per roll rather than by changing the factor. Logbook pages exist for
the slipped frames of only six of the 13 rolls, not the "other 12" this issue assumed. The
2003/2005 rolls have no flight_log_url, and bc5596 frames 204–211 have none either. The five
besides the control were read blind. On every one of them except bcb98013 the crew's height is the catalogue's ÷10
(0.9% off on bc79027, within 0.01% elsewhere), and it sits 6.7–7.6% above ÷10.764. Six roll-heights (299
frames, 1978–2000) ship in flying_height_rolls.csv at factor 0.1. #54 had drawn them 7.6–9.9%
narrow in width, not the flat 7.6% below, because width follows height above ground.
fly_height_slip_factor() is unchanged, and the 1,290 slipped frames the table does not reach are
listed in flying_height_rolls_excluded.csv. The table below also omits a 13th roll: bcb98013
frame 52 (97,924 m), which is a leading-digit typo that no factor fits. That frame and bc5596
are filed as #74. bc79027 reads ÷10, but spacing rejects it, since its page logs 80% overlap.
bcc05001 is undecided: round feet favour ÷10 and its ratio favours ÷10.764.
Problem
fly#54 repairs 1,589 film frames by dividing FLYING_HEIGHT by 3.28084² = 10.764. Measured
while settling fly#60, that factor looks right for the 2003 rolls and about 7.6% wrong for
the pre-2000 ones, where the defect appears to be a dropped decimal point, ×10.
Two independent signs, both from public data:
- A flight logbook. Roll
bc78065 is catalogued at 4,115 m (13,500 ft) on its 1:2000
frames. Its scanned logbook page (flight_log_url), read with the catalogue values withheld,
gives TRUE HEIGHT 1.35, i.e. 1,350 ft. ÷10 gives 411 m; ÷10.764 gives 382 m.
- Round feet. Planned altitudes are round numbers of feet. Under ÷10 the pre-2000 slipped
heights land exactly on them: bc5596 8,600 ft, bc78065 1,350, bc78078 3,100, bc79103 7,200,
bcc00085 18,000 / 18,500 / 19,000. Under ÷10.764 they don't.
The ratio to scale × focal_length points the same way. Ordinary frames centre on r = 1.03:
| roll |
year |
frames |
r ÷10 |
r ÷10.764 |
| bc5596 |
1974 |
8 |
0.957 |
0.856 |
| bc78065 |
1978 |
24 |
0.975 |
0.880 |
| bc78078 |
1978 |
15 |
0.957 |
0.884 |
| bc79027 |
1979 |
10 |
0.970 |
0.898 |
| bc79103 |
1979 |
24 |
1.031 |
0.929 |
| bcc00085 |
2000 |
236 |
1.030 |
0.944 |
| bcc03004 / 06 / 07 / 08 / 46 |
2003 |
1,054 |
1.15–1.22 |
1.06–1.11 |
| bcc05001 |
2005 |
217 |
1.078 |
0.979 |
The 2003 rolls carry measured per-frame heights, not planned ones, so round feet cannot speak
for them, and there ÷10.764 is the better fit.
Why it matters
A pre-2000 slipped frame is drawn about 7.6% narrow (15% in area). Every one of the 1,589 lands
in the band under either factor, which is why #54's band test could not separate them.
What would settle it
The logbook pages for the other slipped rolls (only five have pages — see Outcome), read the same way as fly#60's
(data-raw/height_calibrate-lower_tail_rolls.R, stage 4). If they agree, the fix is to carry
the pre-2000 rolls in flying_height_rolls.csv with factor 1/10, not to change
fly_height_slip_factor().
Evidence
data-raw/flying_height_logbooks.csv (the CONTROL_bc78065 rows),
inst/extdata/flying_height_sweep.csv (set == "upper_tail"), and the fly#60 section of
inst/notes/terrain-correction.md.
Outcome (PR #75)
Confirmed, and settled per roll rather than by changing the factor. Logbook pages exist for
the slipped frames of only six of the 13 rolls, not the "other 12" this issue assumed. The
2003/2005 rolls have no
flight_log_url, andbc5596frames 204–211 have none either. The fivebesides the control were read blind. On every one of them except
bcb98013the crew's height is the catalogue's ÷10(0.9% off on
bc79027, within 0.01% elsewhere), and it sits 6.7–7.6% above ÷10.764. Six roll-heights (299frames, 1978–2000) ship in
flying_height_rolls.csvat factor 0.1. #54 had drawn them 7.6–9.9%narrow in width, not the flat 7.6% below, because width follows height above ground.
fly_height_slip_factor()is unchanged, and the 1,290 slipped frames the table does not reach arelisted in
flying_height_rolls_excluded.csv. The table below also omits a 13th roll:bcb98013frame 52 (97,924 m), which is a leading-digit typo that no factor fits. That frame and
bc5596are filed as #74.
bc79027reads ÷10, but spacing rejects it, since its page logs 80% overlap.bcc05001is undecided: round feet favour ÷10 and its ratio favours ÷10.764.Problem
fly#54 repairs 1,589 film frames by dividing
FLYING_HEIGHTby 3.28084² = 10.764. Measuredwhile settling fly#60, that factor looks right for the 2003 rolls and about 7.6% wrong for
the pre-2000 ones, where the defect appears to be a dropped decimal point, ×10.
Two independent signs, both from public data:
bc78065is catalogued at 4,115 m (13,500 ft) on its 1:2000frames. Its scanned logbook page (
flight_log_url), read with the catalogue values withheld,gives TRUE HEIGHT 1.35, i.e. 1,350 ft. ÷10 gives 411 m; ÷10.764 gives 382 m.
heights land exactly on them: bc5596 8,600 ft, bc78065 1,350, bc78078 3,100, bc79103 7,200,
bcc00085 18,000 / 18,500 / 19,000. Under ÷10.764 they don't.
The ratio to
scale × focal_lengthpoints the same way. Ordinary frames centre on r = 1.03:The 2003 rolls carry measured per-frame heights, not planned ones, so round feet cannot speak
for them, and there ÷10.764 is the better fit.
Why it matters
A pre-2000 slipped frame is drawn about 7.6% narrow (15% in area). Every one of the 1,589 lands
in the band under either factor, which is why #54's band test could not separate them.
What would settle it
The logbook pages for the other slipped rolls (only five have pages — see Outcome), read the same way as fly#60's
(
data-raw/height_calibrate-lower_tail_rolls.R, stage 4). If they agree, the fix is to carrythe pre-2000 rolls in
flying_height_rolls.csvwith factor 1/10, not to changefly_height_slip_factor().Evidence
data-raw/flying_height_logbooks.csv(theCONTROL_bc78065rows),inst/extdata/flying_height_sweep.csv(set == "upper_tail"), and the fly#60 section ofinst/notes/terrain-correction.md.