Four Ports That Passed at Install and Fail in Production

The key fact: 100BASE-TX uses two pairs; 1000BASE-T uses all four.

So a link with a fault on the 4–5 or 7–8 pair will happily run at 100 Mb and will never negotiate 1 Gb. The auto-negotiation fallback is the cable telling you a pair is impaired - it is a symptom, not a switch problem.

Likely causes, in order of probability given the history:

1. The keystone coupler. An RJ-45 coupler in a ceiling is an unapproved splice in a horizontal link. It introduces an impedance discontinuity and additional insertion loss, and it is a mechanical joint in an inaccessible location. It also breaks the permanent-link validation entirely - nothing tested at install covers it.
2. Damage during the re-route. Pulling a previously installed and dressed cable around a new duct can exceed the bend radius, crush the cable against a joist, or nick conductors. One damaged pair = 100 Mb fallback.
3. Length. At 94 m the permanent link already exceeds the 90 m limit, and adding patch cords pushes the channel past 100 m. Insertion loss alone can force the fallback.
4. A termination disturbed during the re-route - a conductor pulled partially out of an IDC.

Diagnostic sequence: run a full certification autotest on the four links with the correct adapters; read the wire map first (an open or split on 4–5/7–8 confirms the pair-level fault), then use the tester's TDR to get the distance to the fault and map it to the coupler location or the duct detour.

Do not simply lock the switch port to 1 Gb - that removes the symptom and leaves a marginal link carrying production traffic.

They are not the same links that were tested. That is the whole answer, and it is a change-control failure, not a testing failure.

1. The cable was modified after acceptance. Two links were re-routed, lengthened by ~12 m, and spliced with a coupler after the acceptance test was complete. Any post-test change invalidates the test result for those links. The pass report describes a cable that no longer exists.
2. Length compliance was lost. At 88–94 m, at least one link is over the 90 m permanent link limit, and with patch cords the channel exceeds 100 m. It may have been 82 m and compliant at test time.
3. A coupler is not part of a permanent link. There is no compliant way to splice a horizontal cable in a ceiling. Even if it tests marginal today, it will drift with temperature and vibration.
4. "Pass" quality is unknown. If that report came from a Basic autotest, a channel-adapter setup, or an out-of-calibration tester, the original pass may never have been meaningful.

Process fixes to state plainly:

• The HVAC re-route should have triggered a change order and a re-test, not an ad-hoc fix by the crew.
• Any pathway change to installed horizontal cable requires inspection for bend radius and jacket damage, and re-certification of the affected links.
• Re-testing is scoped to the four reported links plus every link touched by the re-route, because the failures users report are only the ones someone noticed.

PoE phone rebooting when the lights switch on - two mechanisms, both plausible here:

1. Excessive loop resistance / voltage drop. PoE pushes current down the same pairs. Drop = I × R over the full loop (out and back). A coupler, a partially seated IDC, a nicked conductor, or an over-length run raises R. Under inrush or ring current the powered device sees less than its minimum input voltage and resets. A link that passes a wire map can still have a joint with far too much resistance.
2. Common-mode noise coupling from the LED driver. Switching power supplies on lighting circuits are a strong noise source. If the re-route put the cable within inches of the fixture circuit, the separation that existed at install (305 mm / 12 in from power, or a grounded barrier) is gone. The phone's PHY errors out and re-negotiates or resets.

Floor polisher drop is the same story with a different source: a motor with brushes injecting broadband noise into a cable that is now routed near it or has lost its separation. The correlation with the polisher's location is the diagnostic clue.

Permanent fix (in order):

1. Remove the coupler and replace the spliced run with a single continuous length of the correct cable, re-terminated at both ends. If the new route cannot be made under 90 m permanent / 100 m channel, re-plan the pathway - do not leave an over-length link in service.
2. Restore separation from lighting and motor circuits: minimum 305 mm from power, or a grounded metal barrier where separation is impossible. Route off the ceiling tile; do not lay cable on tiles or on the fixture.
3. Re-terminate any disturbed ends with the correct tool and blade, maintaining twist to within 13 mm of the termination.
4. Check for heat. If PoE cables are tightly bundled in a warm ceiling void, bundle temperature rise increases conductor resistance; keep bundles loose and follow the cable's temperature derating guidance.

Verification:

• Full Cat 6 permanent link certification on the affected links, plus the wire map read for pair-level faults.
• A DC loop resistance / PoE test with a tester that reports loop resistance and loaded voltage, confirming the PD receives voltage above its minimum under load.
• Physical inspection of the re-routed path for bend radius, jacket damage, and separation distance - with photographs, because this will be the second time this route has caused a callout.