Wayside Fiber, Part 1: Where It Is
The physical plant: why signalling went to glass, what lives inside a 48-strand cable, how a midspan access works without touching the express path, and how a loss budget quietly fails at the far end of a fifty-mile corridor.
This is Part 1 of Wayside Fiber, a three-part series. Part 1 is the physical plant: the glass, the duct, the splices, and the light.
The corridor
The running example for this whole series is a hypothetical 49.5-mile corridor with six locations on it. At MP 0.0 sits Waverly, the office: dispatch, the control center, and the headend comms room. At MP 9.2, CP Arden, a control point with a grade crossing beside it. At MP 18.7, a set of intermediate signals we will call Int 187. At MP 27.4, CP Bowman. At MP 38.1, a defect detector, Det 381, which does not exist yet when the cable goes in; it will matter later. And at MP 49.5, CP Chester, where the corridor joins a connecting railroad.
The corridor. The 48-strand trunk ties all six locations together; the dashed second route is a decision this part will have to pay for.
Tying them together: one 48-strand single-mode cable. Forty-eight strands sounds like an absurd surplus for six locations, and the whole discipline of this part is discovering that it is not. Treat the strand count as a bank balance. It opens at 48 dark fibers, and every design decision from here on makes a withdrawal. Watch the meter.
Why fiber, for signals specifically
Every industry has its reasons for fiber. Signalling has three that are its own.
Fiber is dielectric. A signal house is a carefully defended electrical island. Copper communication pairs entering that island are a standing liability: they must be bonded, protected, and arrested, because they physically connect your equipment to whatever is happening a mile away. When lightning strikes near a remote location, the earth around it rises thousands of volts above remote ground. That is ground potential rise, and a copper pair entering the house carries the difference straight to your equipment racks. A glass strand carries none of it. No bonding studs, no isolation transformers, no surge arresters to inspect and replace, and no path by which a strike at CP Arden visits the equipment at Waverly. For an industry whose plant runs beside the track for miles through open country, this alone justified the transition.
Fiber does not care about the railroad's own noise. Track is an electrically hostile neighborhood: traction return current in electrified territory, inverter and chopper harmonics from modern locomotives, code pulses in the rail. Copper circuits in that environment are an unending EMC negotiation. Fiber simply opts out.
Fiber does distance. Our corridor is a fifty-mile campus. Single-mode glass covers it without a single repeater, amplifier, or powered device between houses. Every powered element you do not install is one that cannot fail, cannot be stolen, and does not need a meter socket in the middle of nowhere.
What did it replace? The pole line: open copper wire and later cable, carrying line circuits and the code line, the low-speed serial channel by which the office polled field locations. Code line worked, and versions of it are still in service, but it moved a few dozen indications and controls per location, slowly. It could never carry an event recorder download, a crossing video feed, per-device diagnostics, or the constant vital chatter between modern processors. The pole line was also the single most maintenance-hungry asset most railroads owned. The fiber trunk replaced all of it at once, which is exactly why it now carries responsibilities no single asset ever carried before.
What is actually in the cable
The cable is a loose tube design: 4 buffer tubes, each containing 12 fibers, 48 strands total, on G.652.D single-mode glass. The fibers float loosely inside gel-filled (or in newer designs, dry water-blocked) plastic tubes wrapped around a central strength member. The looseness is the point: the cable can be tensioned, bent, and thermally cycled while the glass inside stays unstressed.
Identification is by the TIA-598 color code, applied twice: once to the tubes, once to the fibers inside each tube.
The same twelve colors name the tubes and the fibers. Two coordinates identify any one of the 48 strands.
The sequence is worth memorizing: blue, orange, green, brown, slate, white, red, black, yellow, violet, rose, aqua. Our cable has four tubes, so only the first four tube colors are used, but the fibers inside each run the full twelve. "Green tube, red fiber" names exactly one strand of glass out of 48, anywhere on the corridor, in any vault, at either end. That naming convention is the atom of every record we will keep in Part 3.
Construction outside the sheath is a set of environment choices:
- In duct is the default here: 1.25-inch HDPE conduit trenched or plowed into the right of way, cable pulled or jetted in later. The duct costs more up front and pays for itself the first time the cable must be replaced, upgraded, or repaired without re-excavating fifty miles.
- ADSS aerial (all-dielectric self-supporting) handles the river crossing near Bowman: a cable that needs no messenger wire and, being fully dielectric, can share structures without bonding concerns.
- Direct bury with an armored cable was considered and rejected. It is cheaper on day one and you pay the difference back with interest at every future repair. When armor is used, note the exception it creates: a metallic element in an otherwise dielectric plant, which must be grounded at entries. Our corridor stays dielectric end to end.
Allocating the tubes
Forty-eight strands is not one pool; it is four tubes, and tubes, not fibers, are the practical unit of separation, because a tube is what you break out at a splice point. So the allocation happens at tube level first:
| Tube | Color | Role |
|---|---|---|
| 1 | Blue | Vital: safety-critical communications only |
| 2 | Orange | Non-vital essential: SCADA, detectors, voice, CCTV |
| 3 | Green | Spare and test |
| 4 | Brown | Future: unopened, spliced straight through everywhere |
Spend the first tube. Vital traffic gets tube 1 in its entirety, even though it needs four fibers, because the boundary you enforce in glass is one no misconfiguration can cross. Tube 2 takes everything essential but non-vital. Tube 3 holds spares and a permanently dark test pair. Tube 4 is never even opened: at every splice location except the two ends, its gel-filled tube passes through the enclosure uncut. Unopened glass is the cheapest insurance the corridor will ever buy.
The meter, after initial allocation: 10 fibers lit or committed, 38 dark. (Vital forward pair, plus eight working fibers in tube 2.)
Chain or ring?
The obvious topology is a chain: Waverly talks to Arden, Arden relays to Int 187, and so on down the line, one fiber pair hopping location to location. It works, and it has a failure mode nobody accepts twice: any single cut splits the corridor, and every location beyond the cut goes dark to the office.
The alternative is a ring. Traffic can flow both ways around a loop, so any single failure leaves every location reachable the long way around. The cheap way to close the loop, and the first design everyone sketches, is a second fiber pair inside the same cable: tube 1, fibers 3 and 4, running express from Waverly straight through to Chester. On the drawing it is a ring, and against some failures it behaves like one: a blown optic, a failed switch, a single bad splice all heal invisibly.
But this is a safety-critical network, and the single-point-of-failure analysis has to be run against the realistic threat, which is not a blown optic. It is a shovel. Both directions of a same-sheath ring occupy the same trench, so the most likely failure mode on the corridor, one dig-in, takes the forward and return paths in a single bite. A ring that shares a sheath is redundancy against the rare failures and none against the common one, and for the network carrying vital traffic that is not a compromise available to us.
So the corridor buys real path diversity, and pays what it costs. The return path gets its own cable: a 12-strand on a physically separate route, plowed on the opposite side of the right of way where the geometry allows and following the highway corridor where it does not, crossing the river on the highway bridge half a mile downstream of the trunk's ADSS span. It runs Waverly to Chester direct, with no intermediate taps; its only job is to close the loop. With it in place, no single cut anywhere, on either cable, can split the corridor.
Diversity is bought in route-miles, not strands: a second trench, a second set of locates, a second cable to test and document forever. Two disciplines keep it worth the money. First, "separate route" is verified on the ground, not on the drawing: two cables that share a bridge, a culvert, or a hundred feet of duct bank share a failure there, and that hundred feet becomes the ring's true weak point. Second, the return cable stays lean. It closes the ring and holds restoration spares, and it must not quietly grow into a second trunk, because every service that migrates onto it is a service that dies with it.
The meter after the ring decision, and the trunk allocation as built:
| Tube | Fibers | Assignment | Status |
|---|---|---|---|
| 1 blue | 1-2 | Vital ring, forward path (hop by hop) | lit |
| 1 blue | 3-12 | Reserved: vital only | dark |
| 2 orange | 1-2 | Code line to CP Chester (low-speed serial) | lit |
| 2 orange | 3-4 | Voice and maintenance access | lit |
| 2 orange | 5-8 | CCTV at control points | lit |
| 2 orange | 9-12 | Spare, non-vital | dark |
| 3 green | 1-2 | Test and restoration pair | dark, reserved |
| 3 green | 3-12 | Unassigned spare | dark |
| 4 brown | 1-12 | Future | dark, unopened |
The new cable is kept deliberately boring: fibers 1 and 2 close the ring, 3 and 4 are its test pair, and the remaining eight are restoration spares.
Running total: trunk, 10 lit, 12 reserved, 26 spare; return cable, 2 lit, 10 dark. Nearly half the trunk is spoken for before the first train sees a signal off this system. This is why 48 was never absurd.
Midspan access: the centerpiece
Two years after cutover, the railroad adds a defect detector at MP 38.1. There is no splice point there; the nearest planned ones are at Bowman, 10.7 miles west, and Chester, 11.4 miles east. The cable runs right past the new site. The wrong answer is to cut all 48 fibers and splice all 48 back, adding loss and risk to every strand on the corridor, including the vital ring, for the benefit of one detector.
The right answer is a midspan access, and it is the single most elegant maneuver in outside plant work:
Midspan access at MP 38.1. One tube opened; 36 strands pass through without being touched.
A crew exposes the cable, sets a precast vault, and pulls generous slack, a hundred feet or so from each direction, coiled and racked inside the vault. That slack is what makes the whole operation possible: it lets the splice enclosure be brought above grade to a work table, and it is the reserve a future repair will draw on. Inside the enclosure, the sheath is opened over a window of a few feet, and then comes the discipline: only tube 2 is cut. Its twelve fibers land on splice trays, two of them fused to a twelve-fiber lateral cable that runs through a riser into the new detector house. Tubes 1, 3, and 4 are cleaned, coiled, and stored in the enclosure intact, express through the case. The vital ring passes MP 38.1 without so much as a bend radius violation: zero added splices, zero added loss, zero new failure modes.
What the crew leaves behind on paper matters as much as what it leaves in the vault. The industry-standard artifact is the splice sheet: one per enclosure, one row per strand or range, saying exactly what came in, what went out, and what it measured. Formats vary by railroad and by contractor; the columns do not. Here is the sheet for this enclosure, SC-38.1:
| Cable / strand | Color code | Disposition | Tray-slot | Lands on | Circuit | Loss (dB) |
|---|---|---|---|---|---|---|
| Trunk T1, F1-F12 | BL tube | express, uncut | stored | trunk east, same strands | VRING + vital reserve | none added |
| Trunk T2, F1-F8 | OR-BL thru OR-BK | fusion, straight through | T2-01 thru T2-08 | trunk east, same strands | code line (F1-F2) + spares | 0.08 max |
| Trunk-W T2, F9 | OR-YL | fusion | T2-09 | lateral F1 (BL) | DET-381-A | 0.07 |
| Trunk-W T2, F10 | OR-VI | fusion | T2-10 | lateral F2 (OR) | DET-381-B | 0.05 |
| Trunk-E T2, F9-F10 | OR-YL, OR-VI | stored on tray, dark | T2-11, T2-12 | nothing (east legs of the cut pair) | spare | n/a |
| Trunk T2, F11-F12 | OR-RS, OR-AQ | fusion, straight through | T2-13, T2-14 | trunk east, same strands | spare, continuous | 0.08 max |
| Trunk T3, F1-F12 | GN tube | express, uncut | stored | trunk east, same strands | test + spare | none added |
| Trunk T4, F1-F12 | BN tube | express, uncut | stored | trunk east, same strands | future | none added |
| Lateral F3-F12 | GN thru AQ | on tray, dark | T2-15 thru T2-24 | future at Det 381 | spare | n/a |
Color codes read tube first, fiber second: OR-BL is orange tube, blue fiber, and names exactly one strand of glass. Every fusion row's loss is the bidirectional OTDR average at 1550 nm, measured after the case was closed, because a splice that was perfect before the enclosure was sealed and racked has been re-stressed since. The "nothing" row is not embarrassment, it is information: the east legs of the cut pair dead-end here by design, and the sheet says so before someone tones them for two hours in 2033.
The lateral has two ends, and the record follows it into the house. FDP-38.1 is the smallest panel on the corridor, and it gets the same sheet as everything else:
| Cable / strand | Color code | Disposition | Tray-slot | Lands on | Circuit | Loss (dB) |
|---|---|---|---|---|---|---|
| Lateral F1 | BL | fusion to pigtail | T1-01 | port 01 | DET-381-A | 0.05 |
| Lateral F2 | OR | fusion to pigtail | T1-02 | port 02 | DET-381-B | 0.04 |
| Lateral F3-F12 | GN thru AQ | stored on tray, dark | T1-03 thru T1-12 | nothing (unassigned) | future | n/a |
Two things to notice. The lateral is a single-tube cable, so its strands carry a one-coordinate color code: F1 is just BL, and the sheet says so rather than inventing a tube that does not exist. And circuit DET-381-A is now traceable end to end across four documents with no gaps: Waverly's port map onto trunk T2-F9, straight-through rows at every intermediate entrance sheet, the SC-38.1 sheet onto lateral F1, and this sheet onto port 01, where a tagged jumper reaches the detector electronics. Ports 03 through 12 sit dark behind ten dark strands: the day Det 381 grows a camera or a second recorder, the glass, the ports, and the blank rows are already waiting.
Thirty-six strands untouched, ten spliced straight through on trays, two put to work. The trunk's meter moves to 12 lit (two orange spares became the detector's feed), and the corridor gains its seventh splice location, one that appears on no original drawing. Hold that thought for Part 3.
Into the house
At each location the trunk's story ends the same way. A lateral leaves a vault, enters the house through conduit, and lands in a fiber distribution panel, the FDP. This is the boundary between two worlds: outside plant cable, which is gel-filled, armored against the environment, and never connectorized in the field, and inside plant, which is connectors, jumpers, and equipment.
The FDP is where cable becomes connectors. Each element in the chain has a known, budgeted loss.
Inside the FDP, each working fiber is fusion spliced to a factory-made pigtail: a short fiber with a connector already polished onto one end. The pigtail's connector snaps into the back of an adapter panel, presenting a clean, labeled port on the front. From there a patch jumper runs to the SFP optic in the wayside switch. Nobody hand-polishes connectors in a signal house; the factory pigtail plus fusion splice is faster, better, and measurable.
The trays inside the FDP are an enclosure like any other, and they get the same artifact: an entrance splice sheet, one row per strand, covering both trunk legs where they land in the house. Here is CP Bowman's. Note that the trunk is not fully cut at a house any more than it was at the midspan: only the tubes with business here are opened, and tubes 3 and 4 never enter the trays at all.
| Cable / strand | Color code | Disposition | Tray-slot | Lands on | Circuit | Loss (dB) |
|---|---|---|---|---|---|---|
| Trunk-W T1, F1 | BL-BL | fusion to pigtail | T1-01 | port 01 | VRING west | 0.04 |
| Trunk-W T1, F2 | BL-OR | fusion to pigtail | T1-02 | port 02 | VRING west | 0.06 |
| Trunk-E T1, F1 | BL-BL | fusion to pigtail | T1-03 | port 03 | VRING east | 0.05 |
| Trunk-E T1, F2 | BL-OR | fusion to pigtail | T1-04 | port 04 | VRING east | 0.03 |
| Trunk T1, F3-F12 | BL-GN thru BL-AQ | fusion, straight through | T1-05 thru T1-14 | trunk east, same strands | vital reserve, continuous | 0.08 max |
| Trunk T2, F1-F2 | OR-BL, OR-OR | fusion, straight through | T2-01, T2-02 | trunk east, same strands | code line, through | 0.06 max |
| Trunk-W T2, F3 | OR-GN | fusion to pigtail | T2-03 | port 05 | VOICE-BOW | 0.05 |
| Trunk-W T2, F4 | OR-BN | fusion to pigtail | T2-04 | port 06 | VOICE-BOW | 0.04 |
| Trunk-W T2, F5 | OR-SL | fusion to pigtail | T2-05 | port 07 | CCTV-BOW | 0.06 |
| Trunk-W T2, F6 | OR-WH | fusion to pigtail | T2-06 | port 08 | CCTV-BOW | 0.05 |
| Trunk-E T2, F3-F6 | OR-GN thru OR-WH | stored on tray, dark | T2-07 thru T2-10 | nothing (circuits serve this house from the west) | reserved | n/a |
| Trunk T2, F7-F8 | OR-RD, OR-BK | fusion, straight through | T2-11, T2-12 | trunk east, same strands | spare, continuous | 0.08 max |
| Trunk T2, F9-F10 | OR-YL, OR-VI | fusion, straight through | T2-13, T2-14 | trunk east, same strands | DET-381 feed, through | 0.06 max |
| Trunk T2, F11-F12 | OR-RS, OR-AQ | fusion, straight through | T2-15, T2-16 | trunk east, same strands | spare, continuous | 0.08 max |
| Trunk T3 + T4 | GN, BN tubes | express, uncut | stored in slack loop | trunk east, same strands | test, spare, future | none added |
The same disciplines as SC-38.1, applied indoors. Pigtail rows are the entrance splices proper, and each is shot and recorded against the 0.10 dB acceptance limit before the tray is closed. Straight-through rows keep the reserved and spare strands continuous end to end, which is why a dark tube 1 fiber can be tested from Waverly to Chester without a truck roll. And once again the sheet records the nothing: the east legs of the voice and CCTV pairs land on no pigtail and no splice, by design, and the row that says so is the cheapest troubleshooting the corridor will ever buy.
The FDP gets its own standard artifact, the port map: one row per connectorized port, tying panel position to strand to circuit to equipment. Here is CP Bowman's, panel FDP-27.4:
| Port | Strand | Color code | Circuit | Jumper to | Status |
|---|---|---|---|---|---|
| 01 | Trunk-W T1, F1 | BL-BL | VRING west, to Int 187 | SW-BOW, SFP 25 | in service |
| 02 | Trunk-W T1, F2 | BL-OR | VRING west | SW-BOW, SFP 25 | in service |
| 03 | Trunk-E T1, F1 | BL-BL | VRING east, to Det 381 | SW-BOW, SFP 26 | in service |
| 04 | Trunk-E T1, F2 | BL-OR | VRING east | SW-BOW, SFP 26 | in service |
| 05 | Trunk-W T2, F3 | OR-GN | VOICE-BOW, to Waverly | channel bank | in service |
| 06 | Trunk-W T2, F4 | OR-BN | VOICE-BOW | channel bank | in service |
| 07 | Trunk-W T2, F5 | OR-SL | CCTV-BOW, to Waverly | video encoder | in service |
| 08 | Trunk-W T2, F6 | OR-WH | CCTV-BOW | video encoder | in service |
| 09-24 | none | reserved | dark |
Read what the table does not list, because that is half its value: tube 1 fibers 3-12 and tube 2 fibers 1-2 and 7-12 have no port appearance at this house, passing through on splice trays without ever touching a connector, and tubes 3 and 4 pass the house entirely uncut. A strand only earns a connector where a circuit terminates. (The voice and CCTV circuits on dedicated glass are the older generation; when their equipment is replaced, the traffic moves onto the switch as VLANs, the strands go back to spare, and the port map records both events.) Every jumper carries the circuit ID on a tag at both ends, and the port map, the splice sheets, and the corridor allocation table all key on the same two-coordinate strand names, which is what makes them one record instead of three.
One detail bites people for years if it is gotten wrong: connector polish. UPC connectors (blue housings, flat-domed polish) are the norm for data links like ours. APC connectors (green housings, 8-degree angled polish) exist for applications where reflectance matters, such as RF-over-fiber video. The two do not mate. Forcing a UPC jumper into an APC port yields an air gap, a huge loss, and shattered end faces if you are unlucky. On this corridor everything is LC/UPC, the As-built says so, and any green connector found in a house is treated as a defect.
The strand ledger
The splice sheets and port maps are each true about one place. The document that is true about the corridor is the strand ledger: one row per strand or range, per cable, rolling every local record up into a single answer to "what is this fiber doing, end to end, right now." It is the sheet you open first, the one every other sheet must agree with, and the one the change process in Part 3 exists to protect. Here is the corridor's, as of the detector cutover:
| Cable | Strands | Assignment | Runs | Status |
|---|---|---|---|---|
| Trunk | T1, F1-F2 | VRING forward | section by section; terminates at every house | lit |
| Trunk | T1, F3-F12 | vital reserve | continuous, Waverly to Chester | dark, reserved |
| Trunk | T2, F1-F2 | code line to CP Chester | Waverly to Chester, no intermediate drops | lit |
| Trunk | T2, F3-F4 | VOICE-BOW | Waverly to Bowman; dark east of MP 27.4 | lit to MP 27.4 |
| Trunk | T2, F5-F6 | CCTV-BOW | Waverly to Bowman; dark east of MP 27.4 | lit to MP 27.4 |
| Trunk | T2, F7-F8 | CCTV-ARD | Waverly to Arden; dark east of MP 9.2 | lit to MP 9.2 |
| Trunk | T2, F9-F10 | DET-381-A/B | Waverly to SC-38.1, onto the lateral; east legs dark | lit to MP 38.1 |
| Trunk | T2, F11-F12 | spare, non-vital | continuous | dark |
| Trunk | T3, F1-F2 | test and restoration pair | continuous, uncut at every house | dark, reserved |
| Trunk | T3, F3-F12 | spare | continuous | dark |
| Trunk | T4, F1-F12 | future | continuous, tube never opened | dark |
| Return | F1-F2 | VRING closure | Waverly to Chester direct, highway route | lit |
| Return | F3-F4 | test pair | Waverly to Chester direct | dark, reserved |
| Return | F5-F12 | restoration spares | Waverly to Chester direct | dark |
| Lateral-381 | F1-F2 | DET-381-A/B | SC-38.1 to FDP-38.1 | lit |
| Lateral-381 | F3-F12 | future at Det 381 | SC-38.1 to FDP-38.1 | dark |
Seventy-two strands under management: 16 lit, 14 reserved, 42 spare or future. The column that does the quiet heavy lifting is "Runs," because it holds the truth no local sheet can: a strand can be lit in one section and dark in another. VOICE-BOW's pair is a working circuit at Arden and a spare at Det 381, and only the ledger knows both at once. That is also why the corridor totals live here and nowhere else: adding up splice sheets tells you what each room believes, and the ledger is where those beliefs are forced to agree. When Part 2 builds an addressing record with exactly this shape, one authoritative table that every local configuration must reconcile to, it is not inventing anything: it is copying the oldest good idea in the fiber plant.
The loss budget
Light leaves an SFP with a known power and a receiver needs a known minimum. The difference is the budget, and everything between the two optics spends it. Start with the forward path, and it barely registers: lit hop by hop between houses, its longest span, Det 381 to Chester at 11.4 miles, needs about 8 dB at 1310 nm and the shortest needs three. Ordinary LX optics close the short hops with room to spare and extended-reach optics cover the two longest. On a corridor with powered houses every ten miles, topology solves the forward path's loss budget before it becomes a question.
The ring's closure span is a different animal, because it cannot hop: there are no houses on the highway route, and its whole purpose is to reach Chester without depending on anything in between. It is one lit span, and the highway routing makes it longer than the railroad: 52.3 route miles against the corridor's 49.5. One number before the tally: single-mode glass loses about 0.56 dB per mile at 1310 nm and 0.35 dB per mile at 1550. (Spec sheets quote 0.35 and 0.22 dB/km; same glass, different ruler.) Run the tally at 1550 nm:
| Item | Quantity | Unit loss | Subtotal |
|---|---|---|---|
| Fiber attenuation | 52.3 mi | 0.35 dB/mi | 18.3 dB |
| Fusion splices (reel joints + entries) | 21 | 0.10 dB | 2.1 dB |
| Connector pairs (FDP each end) | 2 | 0.50 dB | 1.0 dB |
| Design margin (aging, repairs, future work) | 3.0 dB | ||
| Required budget | 24.4 dB |
Note what is honest in this table. Splices are budgeted at 0.10 dB even though a good fusion splice measures 0.03; you budget the acceptance limit, not the best case. And three dB of margin is not padding, it is the span's future: every repair splice this cable ever needs must fit inside it.
Now the failure. The obvious part for the job, a long-reach 1000BASE-ZX optic, offers roughly 21.5 dB of budget. Required: 24.4. The span does not close. Fit ZX optics anyway and the link may even come up on day one, riding on the vendor's factory margin, and then die quietly some winter after the first repair splice: the most expensive kind of failure, the one that passes commissioning. The far end of a long corridor is exactly where a surprising number of real projects discover their arithmetic.
The resolution here is optical, not architectural, because the architecture is the point: this span exists to have no dependencies. It gets EZX-class 1550 nm optics, rated around 33 dB, and closes with more than 8 dB to spare. The general lesson: let topology carry every budget it can, and spend real money on optics only for the spans whose independence is what you are buying.
Proving it
An unproven fiber plant is a rumor. Two instruments turn it into a record, and they answer different questions.
The OTDR answers where. It fires pulses down the strand and times the backscatter, producing the trace: a long slope of attenuation punctuated by events, each with a distance. Shot from both ends (bidirectionally) and at both 1310 and 1550 nm, it characterizes every splice and connector individually. Bidirectional matters because a single-direction shot can report a splice as a gain, an artifact of mismatched backscatter between fiber spools; averaging the two directions yields the true loss. Both wavelengths matter because 1550 is far more sensitive to bending: an event that grows at 1550 but not 1310 is a stressed cable telling you about a future failure while it is still cheap to fix.
One strand, Waverly to Chester, at 1550 nm. Every event has a location, a loss, and now, a baseline value. Note the trace reads 49.7 cable miles for a 49.5-mile railroad: slack coils are distance too.
The insertion loss test answers how much, in total. A calibrated light source at one end, a power meter at the other, at both wavelengths, both directions. This is the number the loss budget table predicted, measured. The OTDR cannot substitute for it, because OTDR event losses are inferred from backscatter, not measured end to end; the light source and meter cannot substitute for the OTDR, because a passing total tells you nothing about where trouble is accumulating. Acceptance means both. And both apply to the diverse return with exactly the same rigor as the trunk: a protection path that has never been proven is not a path, it is a hope with a work order attached.
Treat the as-built drawings as a hypothesis and the OTDR as the experiment. If the trace shows an event at mile 38.2 and the drawings show no splice there, the drawings are wrong, and it is the drawings that get corrected. The set of accepted traces, filed with the records, is the plant's fingerprint. Part 3 turns that fingerprint into money.
Where this leaves the corridor
Every strand is proven. Every budget closes. In each of six houses (seven locations, counting the vault at MP 38.1), a switch port shows a steady green link LED to each neighbor. The physical plant is, in the fullest sense, done.
And nothing can reach anything. No device has an address, no VLAN exists, no packet knows where to go. A lit link is not a network, it is the possibility of one. What turns fifty miles of proven glass into a working railroad network, and the new ways it can fail once that happens, is Part 2: How It Talks.