Why did the Agena-D leak oxidiser when starting its engine?

Question

The Agena-D upper stage, although extremely reliable, had a tendency to leak some of its RFNA oxidiser when starting its Bell 8096 engine; although not a problem for a stock Agena-D, it did cause issues when developing the Agena-D-derived GATV:

The first Gemini-Agena Target Vehicle (GATV) was launched on October 25, 1965, while the Gemini 6 astronauts were waiting on the pad. While the Atlas performed normally, the Agena's engine exploded during orbital injection. Since the rendezvous and docking was the primary objective, the Gemini 6 mission was scrubbed, and replaced with the alternate mission Gemini 6A, which rendezvoused (but could not dock) with Gemini 7 in December.

An investigation into the failure concluded that it was most likely caused by design modifications to the GATV versus a standard Agena D stage. The Agena D was designed to have its engine restarted just once while the GATV would need to be restarted five times. While a standard Agena D pumped oxidizer into the combustion chamber first and then followed with the fuel, the GATV was modified to do the reverse because the normal start method had a tendency to leak oxidizer. While this would not be a problem for the Agena D with its single restart, the multi-restart GATV would eventually lose all of its oxidizer before the stage's operating life (which would last weeks instead of hours) could be completed. Unfortunately, pumping the fuel into the combustion chamber first caused the engine to backfire and rupture from mechanical shock. It was found out that Lockheed engineers did not properly test the GATV to root out this problem (it had been tested at a simulated altitude of 21 miles up when actual Agena engine start would occur at around 75 miles up). The solution to the problem was switching back to the normal oxidizer-first engine start and also testing the GATV in appropriate conditions. Bell Aerosystems, the manufacturer of the Agena's engine, were also instructed to perform further ground-level tests. [Source.]

What caused the Agena-D to leak oxidiser during engine startup?

Answer 1

"On The Shoulders of Titans" suggests that the oxidizer leakage was simply the natural outcome of allowing the oxidizer to flow into the (open-ended) combustion chamber prior to the fuel:

The problem was rooted in NASA's original specification for a Gemini target vehicle able to start and stop its main engine five times during a mission, in contrast to the Standard Agena's two-start engine. This 150 percent increase in demands on the engine at once raised the problem of fuel and oxidizer economy. In the two-start engine, the oxidizer began flowing first, while a pressure switch restricted fuel flow until a given amount of oxidizer had reached the firing chamber. This was known to enhance the engine's starting characteristics, but it was also wasteful. Oxidizer leaked through before engine firing, and some continued to flow after shutdown; the oxidizer would be gone long before the fuel ran out. So Lockheed accepted a proposal by the engine subcontractor, Bell Aerosystems Company, to remove the pressure switch and thus allow fuel to enter the chamber first.

Which leads to the question of why allowing fuel into the chamber first -- and thereby leaking fuel -- was considered a preferable alternative.

According to Solution of the High-Vacuum Hard-Start Problem Yadda Yadda, the "preflow" (propellant entering the chamber before ignition) for the original (model 8096) Agena engine was 6-8 lbs of oxidizer; for the hard-starting Agena 6 engine (model 8247) it was 1.9 lbs of fuel, and for the "fixed" engine ("modified 8247") it was back to 6-8 lbs of oxidizer. (The Agena mixture ratio is about 2.5:1 oxidizer, so the loss of 1.9 lbs of fuel would seem to be about as significant as the loss of 5 lbs of ox.)

However, the postflow is much larger: 32 lbs in the 8096 and modified 8247, 20 lbs in the hard-starting 8247 -- and unfortunately, the paper doesn't appear to explain the change.