Context. Technology is being developed for the characterization and detection of small, Earth-size exoplanets by nulling interferometry in the mid-infrared waveband. While high-performance nulling experiments have shown the possibility of using the technique, to achieve these goals, nulling has to be done on multiple beams, with high stability over periods of hours. To address the issues of the perceived complexity and difficulty of the method, a testbed was developed for the Terrestrial Planet Finder Interferometer (TPF-I) project which would demonstrate four beam nulling and faint exoplanet signal extraction at levels traceable to flight requirements. Containing star and planet sources, the testbed would demonstrate the principal functional processes of the TPF-I beam-combiner by generating four input beams of star and planet light, and recovering the planet signature at the output. Aims. Here we report on experiments designed with traceability to a flight system, showing faint exoplanet signal detection in the presence of strong starlight. The experiments were designed to show nulling at the flight level of ≈10 −5 , starlight suppression of 10 −7 or better, and detection of an exoplanet at a contrast of 10 −6 compared to the star. This performance level meets the flight requirements for the parts of the detection process that can be demonstrated using a monochromatic source. To achieve these results, the testbed would have to operate stably for several hours, showing control of disturbances at levels equivalent to the flight requirements. Methods. A test process was designed which would show that the necessary performance could be achieved. To show reproducibility, the tests were run on three separate occasions, separated by several days. The tests were divided into three main parts which would show first, starlight suppression, second, a realistic faint exoplanet signal production, and finally, exoplanet signal detection in the presence of the starlight. Results. A number of data sets were acquired showing the achievement of the required performance. The data reported here show nulling at levels between about 5.5 and 8.5 × 10 −6 , starlight suppression between 8.4 × 10 −9 and 1.4 × 10 −8 , and detection of planet signals with contrast to the star between 3.8 × 10 −7 and 4.4 × 10 −7 . The signal to noise ratios for the detections were between 14.0 and 26.9. These data met all the criteria of the demonstration, showing reproducible stable performance over several hours of operation. Conclusions. These data show the successful execution, at flight-like performance levels, of almost the whole exoplanet detection process using a four beam, nulling beam-combiner.