What does 25°C mean on a transistor data sheet?

The background is, I'm trying to understand how well a typical transistor obeys the Ebers-Moll equation at constant (junction) temperature.

See this snapshot taken from the Fairchild datasheet of the 2N3904 transistor.

My question is what is meant by 25°C? Is it

1. ambient (room) temperature?
2. The temperature of the epoxy casing?
3. The actual temperature of the silicon (junction) itself?

Also, how is constant temperature maintained? Will a manufacturer publish information on how it took its measurements for datasheets?

In general, there is a temperature gradient from ambient to casing to junction. So the junction is hotter than the casing which is hotter than ambient. The temperature in the Ebers-Moll model refers to junction temperature.

My own speculations:

Of course if the answer is 3., we expect all of the lines to be straight since Ebers-Moll gives

$$I_C = I_S \exp(V_{BE}/V_T)$$

where $I_S$ and $V_T$ are constant at constant temperature and $V_{CE}$. Note the logarithmic scale for $I_C$. So there is definite deviation at 125°C near $I_C= 100 \text{mA}$.

Is this deviation really due to increased junction temperature due to heat dissipated by $I_C$, or is there genuine deviation from the model?

I am thinking that it can't be 1. because the junction would get hotter with increasing $I_C$ and the line would deviate strongly from straight. For the 2N3904 with $V_{CE}=5V$, I calculated that junction temperature increases with $I_C$ as 1°C/mA at constant ambient temperature. ($I_S$ and so $I_C$ increase by about 9% per °C.)

Perhaps they can achieve 3. by taking all the measurements automatically in a fraction of a second, before giving it a chance to heat up.

Any ideas?

It's the junction temperature $T_J$, and yes, they would typically take such a measurement in microseconds before the junction temperature changes.