Verify Cen A, book CrA on T8 for 22 Jul evening, rebuild the rate model

Cen A (reservation 749302, T32) completed at 11:16 UTC, 34 minutes early
because the plan finished, not because of a fault: NGC5128 images are on
t32.itelescope.net:8032 and transaction 851344 billed 160 points. Balance
is now 2,097 of the original 2,638.

That second bill settles the billing question the 20 Jul note left open.
T32 billed 160 for a 120 min plan (2.00 h x 80) and T59 billed 381 for a
180 min plan (3.00 h x 127); both land exactly on the plan's exposure
total and neither lands on wall-clock time. So billing is per imaging
minute as originally documented, and the x1.6 wall-clock multiplier is
withdrawn from CAMPAIGN.md, TODO.md and NOTES.md. What is actually wrong
is the rate card's per-scope numbers, so each scope's rate now has to be
derived from its first real bill. The practical consequence is the
opposite of the old warning: the campaign is now heading to UNDER-spend
by roughly 800 points, so later planning sessions should add or lengthen
runs rather than trim them.

Queue #2 (Corona Australis on T8) is booked at last. The original
01:35-04:35 slot was confirmed dead by probing New.aspx, which answered
"Conflicting reservation. Start time must be later than 2:25 AM" (user
hajo's reservation 749207, exactly as the previous session predicted).
Rather than take the leftover 02:30-05:30, where the target sits 3.5 to
6.5 hours past the meridian, the run moved to the next night's evening,
2026-07-22 21:45 to 2026-07-23 00:45 SSO local, which brackets the ~22:30
culmination so all 115 minutes of imaging happen near the zenith. This is
the geometry of the previously rejected option (c) on the night of option
(b): (c) was only rejected because Cen A held the reservation quota, and
that expired when Cen A completed. Verified via Reservation/MySummary.aspx;
quota now 6:50 of 8:00.

The New.aspx probe is documented in plans/README.md as the way to test a
slot for conflicts, since the DayPilot grid renders client-side and shows
nothing to a plain HTTP fetch.

state/NOTES.md

@@ -61,6 +61,19 @@ time, vs the 80/hr the card predicted. So either Time Used (wall clock) is what
 gets billed, or the card's data-default-rate is stale. Consequence: estimate
 sessions at rate x reserved-slot-length (slots run ~1.15-1.5x imaging time), i.e.
 multiply old exposure-based estimates by ~1.6 until more bills confirm the model.
+
+**RATE CORRECTION 2 (2026-07-21, second bill; SUPERSEDES the x1.6 rule above).**
+T32 session 851344: 120 min of exposure, Time Used 2h25m55s, 25% moon discount,
+billed 160. 160 = 2.00 h x 80 exactly, and the T59 bill is 381 = 3.00 h x 127
+exactly. Both land on the plan's IMAGING minutes to the point, and neither lands
+on wall time. So the wall-clock theory is dead and the original "billed per
+minute of exposure" documentation was right; what is wrong is the card's
+per-scope numbers. Observed effective rates: **T59 127/hr** (card said 80),
+**T32 80/hr even after a 25% moon discount**, implying a base near 107/hr (card
+said 146). The pair looks transposed, but two samples cannot prove it. Rules:
+do NOT apply any blanket multiplier; derive each scope's rate from its first
+real bill; until then treat a card rate as +/-60%; and compute estimates from
+the plan's exposure total, never from the reserved slot length.
 Paid imaging needs no reservation when a scope is idle: same walk-in ACP flow
 as the free scopes, points just deduct. Reservations (to guarantee a slot) go
 through lookup.itelescope.online, which is Cloudflare-protected: Playwright