A network cable is four separate twisted pairs pretending to be one cable. 100 Mbps uses two of them. Gigabit uses all four - so one tired pair creates a fault that half works, and half working is the hardest kind to catch.
Here is the mechanic behind a whole family of strange faults. Fast Ethernet at 100 Mbps sends and receives on two pairs - the wires on pins 1, 2, 3 and 6. Gigabit Ethernet sends on all four pairs at the same time, in both directions. So a cable with one damaged pair can be perfect at 100 Mbps and broken at gigabit - and the ports know it, which is why they negotiate down or keep retrying.
The two pairs on pins 4-5 and 7-8 are only used by gigabit. Damage there is invisible at 100 Mbps and fatal at 1 Gbps - which is why the fault hides until the network is upgraded.
Clean fallback: the ports sense the bad pair during negotiation and settle at 100 Mbps. The user reports only that things feel slow; ping is clean. Flapping: the pair is marginal - connected, but barely. The ports try gigabit, the link errors climb, the link drops and renegotiates. Every renegotiation eats a few packets, giving you the video-case symptom: 3 to 4 pings dropping every few seconds while everything else looks healthy. Marginal is worse than broken, because broken gets fixed once and marginal gets argued about for weeks.
In real offices the guilty pair is rarely crushed mid-cable. It is a conductor seated a millimetre short inside a plug, one punch-down on the wall jack or patch panel that bit the insulation instead of the copper, a jack that has been plugged and unplugged for five years, or a cable bent hard behind a desk. That is why the video team fixed the IO jack and re-patched as part of their hunt - terminations are where pairs die. Start at the ends; the middle of the run is the last suspect.
See the video - the team worked the same chain, wire and IO jack included, before the final culprit showed itself in the adapter settings.
Watch Full Video100 Mbps Ethernet dedicates one pair to sending and one to receiving. Gigabit Ethernet sends and receives on all four pairs simultaneously to reach 1000 Mbps. Remove any pair and gigabit is physically impossible, so the ports must fall back or keep failing and retrying.
Yes. A tester proves each wire conducts at rest. A marginal crimp or punch-down can conduct for the tester yet corrupt gigabit signalling, and split pairs pass continuity while failing from crosstalk. Confirm with the live link speed and a long ping test, not the tester alone.
Pins 4, 5, 7 and 8 - the two pairs that 100 Mbps never uses. Faults there produce the classic symptom: link works, speed is stuck at 100 Mbps, or gigabit drops out intermittently. Re-terminate those ends first.