Your Permanent Link Passed. The Patch Cord Did Not.

Technical library · Updated 19 September 2026 · 22 min read

Why a certified permanent link still fails at the switch: the five meters nobody tests, and the six checks that settle it

An RJ45 patch cord is not a short cable. It is two mated connections plus a short cable, and it is normally the only component in the channel that nobody measured. The permanent-link test exists precisely to exclude it. That single fact explains most of the “the wall passed, the port is broken” tickets: the object carrying the signal at both ends is thinner than the cable it joins, made of stranded wire, bent a hundred times a year, and bought on a price per unit rather than a spec sheet. This page runs the arithmetic on it — resistance, heat, return loss, wear — and lists the checks that need nothing beyond a four-wire meter, a caliper, a scale and the tester already in the van.

1. An RJ45 patch cord is two connections plus a cable

Every patch cord adds two mated interfaces to a link: its plug against the panel jack at one end, its plug against the equipment jack at the other. The cable between them is usually the least troublesome part of the assembly. A 3 m cord is not 3 % of a 100 m link that happens to be floppy; it is four connections plus 3 m, in a channel whose total connection budget is counted in the standards and is small.

Three consequences follow, and they explain most of what goes wrong in the field:

• The test configuration decides what gets blamed. A permanent-link test measures the fixed plant between two outlets. It exists so installers can be graded on work they cannot control — not because the excluded parts do not matter. When a link passes permanent link and then misbehaves in service, the difference between those two measurements is made of cords and equipment ports, and the cords are the part anyone can still change.
• The cord is the only component assembled twice per end. Horizontal cable is terminated once, tested once, then left alone for twenty years. A cord is terminated or molded, plugged in, unplugged for a change, plugged into a different port, coiled, dressed, and unplugged again. Each of those actions is a mating cycle on contacts and a bending cycle on the boot — see section 9, where the same arithmetic a customer service desk needs is done in one table.
• Its performance is a claim about an assembly, not a property of a length. Component limits for cords are specified with a stated length, shield construction and category. Cut the length in half and you have changed the object the limits were written for: the loss falls, the reflections do not, and at some lengths the reflection is all that is left. That is not pedantry — it is why a 0.3 m cord can fail a spec a 3 m cord of the same build passes.

2. What is inside the jacket, and why it is not reel stock

Patch-cord construction differs from horizontal cable in five ways, and each one has a failure mode attached. The vocabulary is worth having straight, because a cord quoted as “Cat6, 28 AWG, molded, PVC” tells you about three of these five and nothing about the other two.

2.1 The question that comes up on every job: can cords be made from the reel?

Yes, and the result is a permanent cord rather than a patch cord. Reel stock is solid conductor, sized for the fixed plant (usually 22–24 AWG), with a jacket and insulation geometry chosen for a termination that is never moved. Terminate it in a plug and you get three properties at once: lower resistance than any stranded cord of the same length, because solid is the shortest path through the copper; a jacket too stiff for a small boot to control, so the bending force lands on the termination instead of the cable; and a conductor that fatigues where it flexes, because solid copper tolerates far fewer bending cycles before it cracks than a bundle of fine strands does.

The practical reading: reel-stock cords in an equipment room that is repatched twice a decade are common, cheap and electrically excellent. The same assemblies taped to a door frame or routed under a raised floor that gets walked on are the ones that come back as “intermittent”. The plug’s side of that argument — the window that has to grip the insulation and the jacket that has to sit inside the boot — is covered in the retention article; the point here is that a cord’s construction has to satisfy both sides at once, and a cord made from horizontal cable is optimized for neither.

2.2 What molding does and does not buy

An overmolded boot bonds the plug to a meter of jacket, so a pull loads the jacket rather than the contacts. That is a genuine advantage, and it is the whole advantage. What molding cannot do is make an undersized conductor, a bad untwist or a mismatched insulation diameter acceptable — it can only make them invisible, which is why a molded assembly with no published per-pair resistance figure is a weaker purchase than a crimped one that publishes three numbers. If you cannot see the build, buy the measurement instead.

3. The arithmetic nobody runs on five meters

Resistance is the cord’s most under-discussed property because a link works across a wide range of it, and because the numbers look trivial. They are trivial for data. They are not trivial for power, for budget, or for a channel whose remaining headroom is smaller than people remember.

The calculation is the same one used in the copper-clad aluminum article, and it needs no instruments — only a length, a gauge and the annealed-copper constant 0.017241 Ω·mm²/m at 20 °C. Conductor area comes from the AWG definition, d = 0.127 × 92^((36−n)/39) millimeters, area from π d²/4; loop resistance is then 2 · L · ρ / A, the factor of two being there and back.

Now put that against the fixed plant, because a cord does not have to be bad to break a budget. A 90 m run of 23 AWG Cat6A horizontal measures about 12.0 Ω loop by the same arithmetic. So:

• two 5 m 24 AWG cords add about 1.7 Ω — 14 % more loop resistance than the cable alone;
• two 5 m 26 AWG cords add 2.7 Ω — 22 %;
• two 5 m 28 AWG cords add 4.3 Ω — 36 %.

The percentages are what make the difference matter. A channel limit is not usually expressed in ohms, but DC resistance unbalance and loop resistance show up in the PoE discussion and in component specifications, and the permanent-link figure commonly quoted for Cat6A is 21 Ω. If a 90 m run used 12.0 Ω of that, the remaining 9.0 Ω is what the cords and their terminations have to live inside — and one 5 m 28 AWG cord consumes 24 % of that remainder before the far end has been given anything. Verify the limit in the revision of ANSI/TIA-568.2-D you are designing against; the number is quoted here because it is commonly specified, not because any single cord or panel was measured for this page.

4. Heat and voltage drop in the patch field

Type 4 PoE pushes 960 mA per pair, or 480 mA through each of eight contacts, which is what the PoE++ article is about. Do the same arithmetic one meter along the chain and the cord turns out to be the hottest convenient place in the room:

Seven watts in a hand is a warm object. Seven watts in a cord with 4 mm of jacket, standing in a bundle of forty, at the top of a rack where the air is already the exhaust of something else, is a different thermal problem — and it is self-reinforcing, because copper’s resistance rises about 0.393 % per kelvin. At 60 °C the same cord is 15.7 % resistive than at the 20 °C the table assumes, which makes more heat, which is where the equilibrium sits. It does not run away, but a figure measured cold on a bench is not the figure in the rack at four in the afternoon in August.

The voltage side of the same number: at 960 mA, a 5 m 28 AWG cord drops 2.04 V per pair, against 0.81 V for the same length in 24 AWG. Add the horizontal run, the far cord and the contact resistances and the powered device sees less of a range it does not control — which is the mechanism behind the specific complaint pattern that gets the most misdiagnosis: a camera that trains up at 06:00 and reboots when the roof gets hot, with clean data counters throughout. The gauge rules for cords under PoE, including the 15 m cap commonly applied to the 28 AWG portion of a channel, are set out in section 6 of the PoE++ article; what this page adds is that the drop is per meter of cord, not per meter of link, so shortening a cord is free and upgrading its gauge is not.

4.1 Cable dressing is a thermal and an electrical decision

Cords are dressed for looks, then evaluated for heat. Two habits do most of the damage:

• Nylon cable ties on small-gauge cords. A tie tightened around a 4.2 mm unscreened cord flattens the jacket and the geometry underneath it. Deformed pair spacing is a return-loss and skew defect that no permanent-link test will ever see, because it is in the cord. Hook-and-loop straps at a tension you can set with two fingers are not a cosmetic preference.
• Looms of forty-plus cords. The same derating logic that applies to bundled horizontal cable applies harder to cords, because cords have less copper per ampere and no chance of shedding heat into a tray. Where the bundle is unavoidable, the honest options are bigger gauge, fewer cords per duct, or a lower PoE type — in that order of cost.

As a reference for bending, the guidance used in the retention article carries over: about one jacket diameter static for unscreened cable, four diameters while under tension, more for screened. For a 4.2 mm cord that is a curve no tighter than 17 mm across while you are pulling it — and a cord that has been coiled to a smaller diameter inside a box for two years arrives with that shape already in it.

5. Return loss, untwist and the mated interface

If the copper is honest and the gauge is right, cords rarely fail on loss. They fail where the cord stops being cable and becomes a joint. Three mechanisms account for most of what shows up in a data report.

5.1 Reflections, and why short is harder

Return loss measures how much of the signal comes back out of an impedance discontinuity. Every mated interface has one, and a cord contributes two per end: plug-to-jack at the panel and plug-to-jack at the device. Over a long run, attenuation is a two-way street for reflections — a discontinuity buried at the far end of 90 m has its echo eaten on the way back, which is one reason a long cable can look cleaner than a short one at the same nominal category. A cord gets no such help: its discontinuities are at the near end, where there is almost nothing between them and the measurement port to absorb anything. This is the mechanical truth behind a claim that is otherwise easy to get wrong: a shorter cord is not automatically an easier pass.

5.2 The untwisted tail

Between the last twist lay and the IDC, the pair is open to the world and running parallel to its neighbours, and the distance it is allowed to stay that way is short — a maximum of about 13 mm at the termination is the figure commonly published for higher-category terminations, and it is worth checking against the standard and the manufacturer’s instruction sheet for the part in hand. The tail is the reason the same cord design can pass in one build and fail in another: it is not a component property at all, it is a workmanship property that the finished assembly hides. On a molded cord the tail length is fixed on a machine; on a hand-built cord it is fixed by whoever was there that day, which is the entire argument in favour of molding, and the entire argument against buying molding from someone who publishes no measurements.

5.3 The plug window and the insulation that has to fit it

Stranded cord conductors usually sit under thicker insulation than a given AWG of solid horizontal wire does, and cord jackets come in a wider spread of diameters than installers assume. A plug that is published for 0.90–1.10 mm insulation and jackets up to 6.5 mm — the sort of window a Cat6 pass-through body states — is not automatically correct for a 28 AWG slim cord, and it is certainly not correct for a round 30 AWG flat cord whose opposing corners exceed the housing diagonal. When the load bar cannot seat, the conductors stop short of the IDC’s bite, the boot bridges air, and the failure looks like a plug problem because it is measured at the plug. The matching rules belong to the connector, and they are in the wire-gauge guide and the termination walkthrough.

5.4 Contacts: the row that gets read backwards

Cords are mated far more often than panels, so they are the place plating thickness pays back — and the place a thin finish shows up first. The relationship between micro-inches of gold, cycles and socket life is set out in the gold-plating guide; what belongs here is only the cord-side conclusion: 3 µ” of gold on an assembly that is unplugged three times a month is a consumable, and a 50 µ” assembly on a port that is patched once in 2019 and once in 2027 is a waste of the same budget. Read the wear claim against the port’s actual duty, not against the catalogue price.

6. Length budgets, category mixing and the connector count

Three separate limits get conflated in conversation about cords. Keeping them apart is the difference between an argument you can win and one that ends in “the standard says what the standard says”.

6.1 The 28 AWG rule, and what it is really about

Small-gauge cords carry a length restriction in the standards — the figure applied to the 28 AWG portion of a channel is a cap in the low tens of meters, with a derating allowance in the annexes rather than a fixed 100 m assumption. That sounds like paperwork for a 2 m patch lead, and mostly it is; the reason the rule exists is the same reason this section exists. 28 AWG is roughly a third of the copper of the horizontal it is joined to, so it is where resistance, heat and loss concentrate, and the standard restricts how much of it a model may tolerate. The applied version, including what it does to PoE classes, is in section 6 of the PoE++ article.

6.2 Short channels and 25G/40GBASE-T: where a 1 m cord is the whole design

Cat8 links invert the usual intuition. The channel is short — commonly 30 m total, for example a 24 m permanent link and two 3 m cords — specified to 2 000 MHz, and always screened. At those numbers, the cords are a tenth of the length and about half of the connections, and the loss budget is so loose that a marginal cord is not caught by attenuation at all. What catches it is the mated-interface quality described in section 5 and the alien-crosstalk geometry of a rack, both covered in section 6 of the shielded-plug article. If you are buying cords for a 25 G or 40 G port, the row to ask for is the cord’s own performance at the top of its stated frequency band, in the length you intend to install — not a “40G ready” adjective.

6.3 The rule that avoids all three arguments

Design the fixed run to 88 m or less, keep cords at 2 m or less in 24 AWG (or 26 AWG where a slim profile is genuinely required), and record the installed cord lengths on the as-built drawing. That is not conservatism for its own sake: it is the difference between a link that is provably inside a model and one that is inside only if the cord you did not measure behaves.

7. Six assumptions about cords that do not survive arithmetic

8. Flat and slim cords: what the geometry buys and costs

Flat cords are popular for three reasons and unpopular for two, and the lists overlap in a way that makes them easy to buy wrongly.

The corner that catches installers is geometric rather than electrical: a flat cord is not a smaller circle. Where a plug body has a rectangular window and a boot molded around a round jacket, a flat cord presents two corners that want to sit outside the housing diagonal. If the boot is then forced flat, it holds the conductors instead of the jacket and the strain path goes to the wrong place. That is a housing question, discussed in section 4 of the retention article; the cord-side rule is that a flat cord needs a flat-specific boot and plug, and not merely a smaller one.

Bend direction matters more than bend size. Bending across the wide face keeps the pair-to-pair distance; bending edge-on compresses the outer pairs and stretches the inner ones, and does it at the one place where the jacket is thinnest. Where a flat cord has to turn tightly — out of a wall plate, around a corner in a ceiling — the flat face should follow the turn. Under a carpet or a door, a flat cord with no tensile member takes every kilogram of furniture in the jacket, and jacket stretch is transmitted to the termination as a slow pull.

9. Boots, snagless tabs, dressing and wear

Mechanical failure in cords is a different fault from mechanical failure in a fixed installation, because a cord is the only part of the channel that is handled. The retention mechanics — latch geometry, climb-out, what a boot does to pull-out behavior, why the jacket must be inside the housing — are laid out in the retention article. What is specific to cords is the duty cycle.

The ratings in the last two columns are the figures commonly published for RJ45 modular plugs and jacks, and they describe the mechanism, not your link: a latch is a cantilever spring being deflected to its design limit, and the number of times it can do that before it stops holding is a material and geometry question. Two things follow. On the busy ports, the cord is a consumable and should be bought and replaced as one; a 50 µ” gold finish does not rescue a latch that has taken a set. On the dead ports, a heavy finish buys nothing, and the failure you are actually protecting against is dust and oxidation in a connector that has not been mated for six years — which is why spare cords for rare ports should live in a bag, not on a shelf.

9.1 Snagless is not a retention feature

A snagless boot — wings either side of the tab so the latch cannot catch on a neighbour while the cord is pulled — trades one failure for another. It stops the tab being levered off by a yank, and it adds a molded part that snaps on a door edge. It also reduces the clearance some panel shutters and fingers need to reach the latch, which is a real compatibility problem rather than a quality one: read the boot as part of the panel, and if the shutter has to be removed to seat the cord, the cord is wrong for that panel. The matching detail, including the boot-to-jacket argument, is in section 4 of the retention article.

9.2 What a coiled spare cord is doing to itself

Cords in storage are usually wound small and left that way. Two consequences: the jacket takes a set, so the assembly never lies flat and keeps a steady load on the plug while mated; and if the tail is wound tighter than the bend allowance, the deformation reaches the pair geometry and shows up as a return-loss or skew difference between new stock and the loose cord on the bench. If spares matter, coil them loosely, or store them in the figure-eight a cable installer uses. It costs no money and it is the cheapest way to keep a shelf of spares honest.

10. How to read a cord data sheet

A cord is a component assembly, so it can be specified. The test of a data sheet is whether the rows would let someone else reject a bad batch. Rows worth demanding, and rows that mean nothing, are below.

11. Six checks at the dock

Ordered by cost. The first three need a bench meter, a scale and a caliper; the last three need a cable certifier and thirty minutes. None of them require a laboratory, and none of them require anyone’s permission.

12. Triage: symptom → is it the cord? → the one measurement

What to do with this

1. Pick the two cord types you install most — usually one round and one slim or flat — and run checks 1 and 3 from section 11 on each. Write the measured loop resistance next to the computed figure from Table 2 and keep it as the reference for future batches.
2. Measure one link twice, channel with cords and MPTL without them. The delta goes in the handover file, where it answers the next “the wall passed” argument without an instrument.
3. Take Table 7 to the next cord specification you are asked to approve and delete the rows that are not there. If resistance, OD and a cycle count with a method are all present, you are buying a component; if only a category word is present, you are buying a hope at a unit price.

Frequently asked questions

Sources and standards referenced

• ANSI/TIA-568.2-D — balanced twisted-pair cabling components, channel and permanent-link models, connector counts, and the length and category rules that patch cords are specified against. Read the tables in the edition you design to: this page states the structure of those limits (fixed run plus cords, counted connections, small-gauge length caps) rather than quoting clause numbers from memory.
• ISO/IEC 11801-1 and -2 — generic cabling, channel and equipment-cord models, and the screened-channel shield-continuity figure commonly given as 2.5 Ω or less. Confirm in your copy before repeating it in a specification.
• IEC 60603-7 — the modular plug and jack interface documents, including the family of performance classes for 8-position connectors and the mating-cycle and durability framework the wear figures in section 9 are expressed against.
• IEEE Std 802.3bt-2018 and IEEE 802.3bq (25GBASE-T and 40GBASE-T) — per-pair currents for Type 1 to Type 4 power over Ethernet, and the short Cat8 channel with two connections that section 6.2 discusses.
• TIA TSB-184-A and TSB-184-A-1 — power distribution over balanced twisted pair, bundle heating and the derating context in which cord resistance stops being academic.
• ASTM B3 — annealed copper wire: the resistivity constant 0.017241 Ω·mm²/m at 20 °C behind every figure in sections 3, 4 and the FAQ. The AWG diameter relation 0.127 × 92^((36−n)/39) mm is the standard definition and is given so the tables can be rerun for any gauge and length.
• IEC 61156-5 / IEC 61076 series — multidimensional cord assemblies and connector mechanical requirements, the documents under which an assembled cord is a component rather than a length of cable.
• NFPA 70 (NEC) Article 800 and UL 444 — cable listing and jacket/flammability classes (CM, CMR, CMP, LSZH) that determine where a cord may legally be installed, which is a different question from where it performs.
• UL 94 — flammability classification of housing polymers, relevant to boot and housing material claims on assembled cords.

On the numbers in this article. Resistances, voltage drops and heat figures are computed in the open from the AWG geometry of the conductor and the published resistivity of annealed copper, so they can be rerun for any gauge, length and current; they are arithmetic, not measurements, and they are not a compliance verdict on any particular cord. Temperature coefficients (0.393 % per kelvin for copper) and the strand-lay penalty are handbook-level and used for order of magnitude only. The 21 Ω permanent-link resistance figure, the 2.5 Ω shield-continuity figure, the 13 mm untwist allowance and the 750/1,000-cycle and 3–50 µ” plating ranges are values commonly specified or commonly published for components in this class; they are reproduced here as row labels worth asking for and arguing about, and each should be confirmed against the standard revision or data sheet that governs the part in hand. No “typical” performance for any supplier’s cord appears anywhere on this page, and nothing here reports a test that was run for this article.