Correct the saturated-nucleus claim and cancel the booking it justified

The previous commit asserted that the 21 Jul Centaurus A data came back
with a saturated nucleus, and booked a short-exposure core set on T32 to
fix it. The assertion was wrong.

It came from taking the maximum pixel inside a 300 x 300 px box centred
on the frame. That measures the brightest thing near the middle of the
image, which is not the same as measuring the galaxy. The 65313 ADU
belonged to a foreground star 128 px (69 arcsec) from the nucleus, well
inside the box. Median filtering the inner 800 x 800 px to remove stars
leaves the galaxy peaking at 1944 ADU, roughly a thirtieth of the clip
level, with not one star-free pixel above 30000 ADU anywhere in that box.
The brightest pixel in the whole frame is at the right-hand edge, 2370 px
from centre.

The core looked blown in the processed image because of the stretch, not
the sensor: the white point sat at the 99.995th percentile, a level set
by field stars, so the galaxy's entire range from a 7 ADU sky to a 1944
ADU peak was compressed into the top few percent of the tone curve.
Re-rendering the same frames with a second curve scaled to the galaxy
recovered the core completely, with no new data.

The T32 reservation has been cancelled without running, so it cost
nothing. The other three reservations were verified intact afterwards
and quota is back to 6:05 of 8:00. plans/hdrcore-t32.txt stays in the
repo and on the scope: 47 Tuc and the SMC may genuinely saturate, a
dense cluster core being the fastest thing there is to clip, and the
ladder is ready if they do.

The corrected habit is recorded in CAMPAIGN.md: measure whether a core is
saturated before buying time to fix it, filtering stars out of the
measurement first. The check takes a minute and would have saved this
whole detour.

TODO.md also records a second result that affects future planning: the
faint outer halo of that stack is limited by the sky-plane subtraction,
which absorbed about 17.9 ADU/px of real halo light, rather than by
exposure time. More integration alone would not have gone deeper.

CAMPAIGN.md

@@ -80,7 +80,7 @@ actual bill, and treat any not-yet-observed scope's card rate as +/-60%.
 | 749230 | T8 | 26 Jul | 21:35-23:55 | catspaw-t8.txt | ~146 (25% moon disc.) | still booked |
 | 749302 | T32 | 21 Jul | 18:50-21:50 | cena-t32.txt | ~350 (x1.6 rate, 25% moon disc.) | COMPLETED, billed 160. EVENING slot: T32 has a 35 deg elevation floor and Cen A culminates ~17:10 before dark, sinking to 11 deg by the old 00:35 slot (which could never have run). Ran 18:50-21:16, finishing 34 min early (plan complete, not a fault); images landed under /images/qisback/T32/NGC5128 |
 | (id not captured) | T8 | 22 Jul | 21:45-00:45 | cra-t8.txt | ~250 | booked + MySummary-verified 2026-07-21 11:45 UTC. This is queue #2 finally placed: see the resolution note below. Runs 11:45-14:45 UTC 22 Jul |
-| (id not captured) | T32 | 22 Jul | 18:50-19:50 | hdrcore-t32.txt | ~25 | booked + MySummary-verified 2026-07-21 13:30 UTC. HDR core set for the 21 Jul Cen A data, whose nucleus is saturated: see the note below. Runs 08:50-09:50 UTC 22 Jul, immediately before the T8 run that night |
+| (id not captured) | T32 | 22 Jul | 18:50-19:50 | hdrcore-t32.txt | ~25 | **CANCELLED 2026-07-21 ~15:00 UTC, never ran, 0 points spent.** Booked to fix a saturated Cen A nucleus; the nucleus is not saturated (see the correction below). Deleted from MySummary with the other three reservations verified intact afterwards; quota back to 6:05/8:00 |
 | 749314 | T33 | 21 Jul | 00:05-00:50 | ngc104-test.txt | 0 (free) | booked + grid-verified 2026-07-21 08:55 UTC. 47 Tuc first-light retry, third attempt. Runs 14:05-14:50 UTC 21 Jul. Reschedule flag again No (free scopes never rebook) |
 
 The T33 first-light test is now on its THIRD booking (749314); the two prior free
@@ -103,24 +103,34 @@ Verify every booking via Edit.aspx?id= probing (sequential ids), NOT the grid.
 | 5 | GRAS008 | 2026-08-08T02:35:00 | 2026-08-08T06:05:00 | 2:35 AM | 6:05 AM | smc47tuc-t8.txt | ~390 |
 | 6 | GRAS032 | 2026-08-09T03:05:00 | 2026-08-09T06:05:00 | 3:05 AM | 6:05 AM | tarantula-t32.txt | ~365 |
 | 7 | (T70, blocked: server unreachable) | night TBD before 5 Aug | ~2.5h | | | scorpius-t70.txt (in scratch, re-upload needed) | ~238 |
-| 8 | BOOKED 2026-07-21 for 22 Jul 18:50-19:50 (see table above) | | | | | hdrcore-t32.txt | ~25 |
+| 8 | WITHDRAWN: booked then cancelled 2026-07-21, premise was wrong (see below) | | | | | hdrcore-t32.txt | 0 |
 
-### Short-exposure core sets are now part of the campaign
+### CORRECTION (2026-07-21): the Cen A nucleus was never saturated
 
-The 21 Jul Cen A data came back with a saturated nucleus: a single 300 s
-luminance sub reads 65313 ADU in the core, so the deep stack is clipped there
-and no amount of processing recovers it. plans/hdrcore-t32.txt is the fix, an
-exposure ladder (L 8x60s, 8x15s, 8x5s, RGB 5x30s, all BIN2 to match) that
-blends in under the clipped core. It costs about 25 points, roughly a sixth of
-what the deep run cost, and the same trick applies to any bright target.
+An earlier entry here claimed the 21 Jul Cen A data came back with a saturated
+nucleus, and booked a short-exposure core set on that basis. **The claim was
+wrong and the booking has been cancelled without running, at a cost of zero
+points.**
 
-**Apply this to the rest of the queue.** NGC 6744 (already shot, 19 Jul) has a
-bright nucleus and probably needs the same treatment; 47 Tuc and the SMC
-(queue #5) certainly will, since a globular cluster core saturates fast. Before
-each remaining bright-target run, budget an extra ~25 points and an hour of
-reservation quota for a matching short set, and cut the ladder from the deep
-plan's exposure rather than guessing: shortest sub = deep sub / 60 is what was
-used here.
+The 65313 ADU figure came from taking the maximum pixel inside a 300 x 300 px
+box centred on the frame. That measures the brightest thing near the middle of
+the image, which is not the same as measuring the galaxy: it was a foreground
+star 128 px (69 arcsec) from the nucleus, sitting inside the box. Median
+filtering the inner 800 x 800 px to remove stars leaves the galaxy peaking at
+**1944 ADU**, about a thirtieth of the clip level, with **not one star-free
+pixel above 30000 ADU**. The core looked blown in the first processed image
+because of the display stretch, whose white point was set by field stars, and
+re-rendering the same data with a second tone curve scaled to the galaxy
+recovered it fully.
+
+**The habit that IS worth keeping, in corrected form:** before booking extra
+time to fix a saturated core, *measure whether it is saturated* - median-filter
+the inner region to remove stars first, and check where the frame's bright
+pixels actually are. The check costs a minute. 47 Tuc and the SMC (queue #5)
+may still genuinely saturate, a dense cluster core being the fastest thing there
+is to clip, so run the same measurement on their data before assuming either
+way. plans/hdrcore-t32.txt stays in the repo (and on T32) as a ready-made
+ladder if one of them turns out to need it.
 
 ### Session 2026-07-21 ~08:55 UTC (booking round run early, before Cen A ran)