The evolution of a dilute electron-positron fireball is calculated in the regime of strong magnetization and very high compactness ( \ell \sim 10 ^ { 3 } -10 ^ { 8 } ) .
Heating is applied at a low effective temperature ( < 25 keV ) , and the fireball is allowed to expand , so that the formation of a black-body spectral distribution is inhibited by pair annihilation .
The diffusion equation for Compton scattering is coupled to a single-temperature pair gas and an exact ( trans-relativistic ) cyclo-synchrotron photon source .
We find that the photon spectrum develops a quasi-thermal peak , with a power-law slope below it that is characteristic of gamma-ray bursts ( F _ { \omega } \sim const ) .
The formation of a thermal high-frequency spectrum is checked using the full kinetic equations at \ell \sim 10 ^ { 3 } .
These results have several implications for the central engine of GRBs , and the mechanism of energy transport .
1 .
Baryon rest mass carries less than \sim 10 ^ { -5 } of the energy flux at jet breakout at \lesssim 10 ^ { 12 } cm from the engine , with most carried by the magnetic field .
2 .
This degree of baryon purity points to the presence of an event horizon in the engine , and neutrons play a negligible role in the prompt emission mechanism .
3 .
X-ray flashes are emitted by outflows carrying enough baryons that the photosphere is pair-depleted , which we show results in faster thermalization .
4 .
The relation between observed peak frequency and burst luminosity is bounded below by the observed Amati et al .
relation if jet Lorentz factor \sim ( { opening~ { } angle } ) ^ { -1 } at breakout .
5 .
Stellar models are used to demonstrate an inconsistency between the highest observed GRB energies , and a hydrodynamic nozzle : magnetic collimation is required .
6 .
The magnetized pair gas is dilute enough that high-frequency Alfvén waves may become charge starved .
Finally , we suggest that limitations on magnetic reconnection from plasma collisionality have been overestimated .