Three clear pear-shaped glass bulbs of 1904 on a wooden table in a laboratory in London, each seven centimetres across with a glass tip at its top, standing side by side on rings of orange-brown spirally wound material, each ring carried on three bent wire legs in black insulation, inside the middle bulb a dark metal plate beside a carbon filament glowing faint orange, inside the left bulb a zigzag carbon filament, thin wires outside the glass, beside them a mirror galvanometer of brass and mahogany throwing a thin line of light onto frosted glass, glass battery jars, a hand in a dark tweed sleeve holding a wire clamp, gaslight from above.
What it was used for
Fleming, scientific adviser to the Marconi company, built incandescent lamps in 1904 with a metal cylinder around the filament and put them in the receiving circuit of wireless telegraphy. By his own account the filament was heated from a battery of 6 cells; the valve passed current in one direction only, so the wave trains of a spark transmitter appeared as a tone in the telephone or as a deflection on the galvanometer. The wave trains held 20 to 100 oscillations; the oscillations themselves were too fast to move a galvanometer needle directly. The operator heard Morse as short and long stretches of this tone.
What was new
The valve rectified high frequency without any part having to respond mechanically to each wave train, as the filings coherer did; Fleming later called the detector simple and easily managed. Operation paid the price: the detector needed a heating battery of its own. The barb is in the patent. On 21 June 1943 the US Supreme Court held the Fleming patent invalid by reason of an improper disclaimer: the specification had also contemplated low-frequency currents, the overclaiming was not inadvertent, and the delay of 10 years in disclaiming was unreasonable.
What the sources leave open: The IEEE milestone page gives no patent number. The Justia text of the judgment contradicts itself: the syllabus says 803,864, the opinion heading 803,684 - so no US number is given here. Fleming's own text states the spark rate differently in two places (50 to 500 and 50 to 100 per second), so it is omitted.