Wednesday, February 2
At 04:30 nothing moved. The QSE line had been quiet since midnight. The training block on feeder 6 ran flat at thirty-one megawatts, and the coffee had gone past burnt to something closer to tar. Marisol couldn't move a cable. Kessler had said so twice, once in writing. She couldn't load a setting either, because the April change window was the April change window. That left reading.
She had Abel's drawings from the 0412 job open on the second screen, and she went through them the slow way, one breaker at a time. Every feeder breaker had a 52a auxiliary contact, a little switch that closed when the breaker closed. It ran to that breaker's own relay on port 2. On every sheet a second line ran off the same terminal to T1, the transformer differential, which used breaker status in its logic and stamped it on port 1.
So there was one piece of copper and two relays watching it. When BKR4 opened, two clocks wrote down the same instant.
She pulled the settings export Kessler had sent in January, read-only, and found the input debounce for each. BKR4 was 4 milliseconds and T1 was 8. If both relays were on IRIG-B, T1's stamp should trail BKR4's by about four milliseconds every time, give or take contact bounce. If BKR4's port 2 had dropped back to the RTU's clock, the gap would be whatever the RTU happened to be off by that day.
She had one known-good moment and one known-bad one. Monday 07:24, right after Abel retightened TB3-7, BKR4 closed under a good clock. T1 minus BKR4 came to 4.3 ms. The January 19 trip had happened inside the thirty-nine days BKR4 spent on DNP time, and there the gap was 43.9 ms.
The two cases were a factor of ten apart. She wrote it on an index card in block capitals, because block capitals were what people read at 6 a.m.
PAIR CHECK. AFTER ANY BKR2/4/6 OPERATION: SER FROM THAT RELAY AND T1 (READ-ONLY, ENG WS). T1 MINUS BKRn = 4 MS ±2 (BKR2, 6: CHECK DEBOUNCE FIRST). OVER 10 = SUSPECT PORT-2 FALLBACK. BASELINE 1/31 07:24 = 4.3. KNOWN BAD 1/19 = 43.9.
The trouble was that it only worked when a breaker moved, and in a quiet week that might never happen. She checked the switching board. BKR2 had its quarterly maintenance cycle Thursday at 06:00, so she added it to the card as the first planned sample.
Trey came in at 05:45 with his lunch in a grocery bag and read the card over her shoulder while she taped a copy under 7f.
"So it tells you after."
"Everything about time tells you after."
"Then what's it for?"
"It tells me before I sign something."
He put the lunch in the fridge before he answered. "That's a different after."
"That's the only after I get."
Thursday, February 3
She didn't tape the card up this time. She stood at the engineering workstation at 05:52 with it in her hand, the way she'd hold a switching order.
The card said to check debounce first on BKR2 and BKR6, so she did it before anything moved. BKR2 was the old relay, swapped in 2019 and never re-templated. Its 52a input was set to 8 milliseconds, the same as T1's.
That made the expected gap zero, not four. A check centered on zero could come out negative, and "over 10" had to mean more than ten in either direction. She penciled it in with the time beside it, 05:55, so nobody, including her, could say later that she'd written the rule after she saw the answer.
Abel's voice came over the radio from the yard. "Opening two."
06:01:14. On the one-line the breaker symbol went green, then a cycle later the SER files had something in them. She pulled both.
The open came out at +0.6 ms.
"Closing two."
06:03:40. The close came out at −1.2 ms.
The close was noisier, and that fit. A 52a contact chatters while the mechanism latches. An open is a clean break and a close is a door banging shut. Both numbers were inside ±2, so BKR2's port 2 had been on IRIG-B at 06:01 and at 06:03. That was all the method could say, and she wrote down only that.
2/3 BKR2 MAINT. DEBOUNCE 8/8, EXPECTED 0. OPEN +0.6, CLOSE −1.2. GOOD 06:01–06:03 ONLY.
Then one more line. The lesson of January 19 was that a good clock at one minute said nothing about the thirty-nine days around it.
NEXT SAMPLE WHENEVER A BREAKER MOVES. GAP SINCE LAST SAMPLE GOES ON THE LOG, IN DAYS.
The gap since Monday was three days. She wrote 3.
Abel came in with the switching order to sign back, his cheeks red from the wind off the bay. "Get your number?"
"Two numbers," she said. "And a minus sign I didn't know I needed."
Friday, February 4
Nothing was scheduled to move BKR6 until the fifteenth, so at 13:30, between the afternoon market print and the evening ramp, she did the reading now instead of at the breaker.
She pulled the BKR6 settings file from the engineering share, not from the relay, and diffed it against BKR4's. Debounce was 4 ms, the same as BKR1 through BKR5. Expected offset against T1 was +4, the old number. That part was fine.
The other difference wasn't in the settings at all. It was in the comms notes at the bottom of the bay drawing, in a font smaller than anyone had meant anyone to read. BKR6 had gone in on the new bay in 2021. Its IRIG-B came off output 2 of the distribution amplifier. The other five feeder relays and T1 all hung off output 1.
She sat back.
Every pair check on BKR1 through BKR5 compared two relays fed by the same amplifier output. If a terminal on output 1 worked loose, both relays would lose IRIG together and fall back to the same RTU time together. The check would read GOOD, and it would be right in the narrowest way possible. Only BKR6 against T1 crossed outputs.
She wrote it under the debounce table.
PAIRS ON OUTPUT 1 CATCH A RELAY-SIDE FAULT ONLY. BKR6/T1 CATCHES AN AMP-SIDE FAULT. WEIGHT BKR6 SAMPLES.
Then the line she'd have skipped a year ago.
NEITHER CATCHES THE GPS CLOCK ITSELF. EVERYTHING SHARES THAT.
That was Kessler's question, not hers, and it lived in April. She sent him two lines with the amplifier output map attached: no settings touched, and she asked that the April time-quality work include the GPS clock's own holdover alarm in the RTU map.
He answered in four minutes. Good catch on the amp. Clock alarm's already on the April list, item 9.
She hadn't read item 9. She read it now. It was there, plainly worded, and had been since December. She wrote K. ITEM 9 on the card so she wouldn't ask twice, Being right about the problem and slow on the paperwork was its own kind of embarrassing.
Monday, February 7
The weekend gave her two BKR6 operations she hadn't had to ask for. The dispatcher had switched the capacitor bank Saturday at 06:02 and 18:47, both on schedule and both in the log.
At 07:40 she pulled the event records and lined up BKR6's 52a against T1 by hand. The pair-check summary Trey had built Friday afternoon was written for output-1 pairs, and she wanted to see the numbers the weighting was supposed to protect.
06:02 came out at +4.2 ms and 18:47 at +3.9. Debounce said +4, so both were inside half a millisecond, which was as good as the yard got.
So for the first time she could write that output 2 and output 1 had agreed on Saturday, at two times eleven hours apart. She wrote exactly that and nothing bigger.
Under it she did the arithmetic she made herself do. There were two samples, both from one cold dry day, and both came within an hour of a capacitor switch that loaded station service the same way each time. If output 2 had a fault that came with temperature, a February Saturday was the worst day to find it.
BKR6/T1: 2 GOOD (2/5, CAP SW). SAME CONDITIONS ×2 = 1 TEST, NOT 2. NEXT: WARM NIGHT OR UNSCHEDULED OP.
She checked the one thing that would have made Saturday meaningless, the RTU's own clock-quality flag for both events. It was blank, as it always was, because the TQ bits weren't mapped until April. She'd half known that. She wrote TQ BLANK, EXPECTED (K. ITEM 9) and closed the file.
Trey, reading over her shoulder again: "Two goods and you wrote it as one."
"Two goods on one kind of day. Ask me again on a different kind."
Since Friday, one line on the card had gone from untested to tested once. She'd rather the card said once than said fine.
Thursday, February 10
At 02:00 the fog came in off the bay so thick the yard lights wore halos, and you could hear the 345 kV lines from the parking lot, a wet crackle like bacon on low. She knew that sound. Salt was settling on the insulators faster than the dew could wash it off.
At 02:14 BKR6 tripped.
The alarm went. The first reclose failed and the second shot held. It was a phase-to-ground fault, and the target lights on the relay faceplate said what she'd already guessed: one flashover on a post insulator on the new bay, the coastal winter thing. Feeder 6 was back by 02:14:09. The training block on it wasn't. It had gone down with the breaker and was taking its time coming back.
At 02:14:29 the EMS logged it: ARMED TRAINING BLOCK F6 DROPPED, 31 MW.
At 02:40 the account desk emailed. They worked nights now too, since the PCLR contract made nights worth money. Could the 31 MW drop go into February's deployment report as delivered interruptible load?
She understood the question better than whoever had typed it. Delivered interruptible load was what the campus got paid for being willing to give up. If the training load had come off first and the breaker had opened after, the campus had curtailed, and the drop counted. If the breaker had opened first, a fault had done it, and nobody gets paid for being tripped. The whole answer was in the order, and the order was only as good as the clocks.
She didn't open the sequence yet. First she pulled the pair check, because this was the sample the card had been waiting for. It was BKR6's first unscheduled operation, on a wet night, with nobody's schedule behind it.
BKR6 trip against T1 came out at +4.1 ms, and the reclose at +3.8. Expected was +4.
She wrote it on the card before anything else: 2/10 02:14 BKR6 FAULT TRIP, FOG. +4.1 / +3.8. GOOD 3/5. DIFFERENT CONDITIONS. COUNTS AS TEST 2.
Then she let herself read the sequence. BKR6 opened at 02:14:07.312, relay time, pair-checked. T1's own metering showed feeder 6 current falling in the same cycle, because the load was on the feeder that opened, and there wasn't anywhere else for it to go. The EMS entry at 02:14:29 was the RTU noticing, twenty-two seconds late and on its own clock.
So the breaker had gone first. It wasn't close: the training load hadn't started to fall until the breaker opened.
She answered the account desk in two lines.
Training MW on F6 fell with BKR6 at 02:14:07.312 (relay time, pair-checked same night). Not a curtailment. Do not report as PCLR delivery.
The reply came at 02:51. Understood. Can you send the relay file for the record? She attached it, read-only export, with the pair check numbers in the body so whoever opened it in March wouldn't have to take her word for which clock had stamped it.
Then she did the thing she'd been circling since January. She took the card down from under 7f and wrote across the top, above PAIR CHECK, in letters a size bigger than the rest:
ANY SEQUENCE THAT CROSSES THE EMS AND A RELAY IS SIGNED IN RELAY TIME ONLY WITH A PAIR CHECK FROM THE SAME NIGHT. WITHOUT ONE, IT'S RTU TIME, AND IT SAYS SO.
She read it back and changed nothing. It wasn't a policy. Policy was what Kessler's April list was for. It was what she already did, written where she'd have to look at it before she picked up a pen.
At 05:50 Trey came in shaking water off his jacket and read the top of the card before he said good morning.
"Did you just cost us a month's deployment credit?"
"I cost us a fault," she said. "The fog did the rest."
He looked at the account desk email on her screen, then at the card. "And if the breaker hadn't moved tonight?"
"Then I'd have had the EMS's twenty-two seconds and nothing to check them against. I'd have written RTU time and let them argue."
"They'd have argued."
"They'd have argued with a label. That's easier than arguing with a number that's wrong."
She put Abel's name and F6 POST INSULATOR, NEW BAY on the washing list for the weekend. The next flashover might come on a night with no breaker moving to check the clocks against. Then she taped the card back under 7f, at four corners, and the gap on the log went back to zero days.