Saturday, January 29
Abel's text came in at 04:50. The overnight load sat flat on the left screen, the QSE line hadn't rung since one, and the coffee in the pot had gone to the burnt smell it got after three hours on the plate.
Name is Owen Kessler, protection engineering. He owns the relay settings files. Any change to a relay's point map is a settings revision, his signature, relay out of service to load it, and a test sheet. He does those in batches. Next batch is April.
April. Marisol looked at the hook wall, at line 7d in blue pen on the sheet titled Clock Source When EMS Lies: Relay clock loss must reach the desk. Blue meant not tested. Three months of blue.
She had learned something from the SCHED CLOCK label the week before, when a report drafter stripped two words out of a table cell and nobody noticed for a day. The question wasn't who owned the change. It was what exactly would change, and in which box.
She texted back. Is IRIG-B LOSS already in the relay's DNP map and just not polled? Or does it have to be added?
Abel went quiet for twenty minutes, which meant he was pulling the settings export and not guessing. At 05:14: It's there. Binary input 47 on all three. Default map since commissioning. The RTU just never asks for it.
So it wasn't a relay change. Nothing would load into a protective relay and no breaker would come out of service. It was the RTU's poll list, which was Abel's side of the house, the same RTU that RTU-PL-0412 had touched on Friday.
She drafted RTU-PL-0413 herself and put both facts at the top so nobody could read it as something it wasn't.
No relay settings change. BI 47 exists in as-left map for BKR2/4/6 (Abel export 1/29 05:14). Request: add BI 47 x3 to RTU-PL poll list; map to EMS as PL BKRn IRIG LOSS, SOE only, no alarm, no page.
No page was on purpose. A relay that lost satellite time still tripped on current. Only its milliseconds were in doubt, and that was a reason to read its stamps carefully the next morning, not to wake anyone at three.
Then a line for Kessler, on the cc and not as approver. FYI, read-only. If you want this in your settings file of record as a used point, tell me and I'll hold it for your April batch. If he said yes it would be slower. It would also be his call, made out loud, instead of a thing she'd done around him.
At 05:40 she went to the hook wall. 7d stayed blue. Under it, in pencil: BI 47 already in relay. RTU poll only. 0413 filed 1/29, Kessler cc FYI.
Trey read the pencil when he came in. "You're not waiting till April."
"I'm not changing a relay," she said. "I'm asking the RTU to read a line that's been there since 2026."
Sunday, January 30
The desk phone rang at 21:15, before the night had properly started. Not a text. Kessler was the kind of engineer who called when he wanted his words in the phone log.
"Owen Kessler. You cc'd me."
She was ready for the April speech. He didn't give it.
"Your request is fine. It's an RTU change and I don't sign RTU changes. I'm calling because I want you to know what 47 doesn't tell you, before somebody upstairs reads that line and relaxes."
She pulled a pad over.
"47 asserts when the relay stops seeing an IRIG-B signal on the input. Cable cut, connector off, clock dead and silent. It does not assert when the clock is alive and sending time it shouldn't trust. A clock in holdover keeps sending IRIG-B. The relay takes it. Our relays are set to ignore the quality bits in the frame, because when they were commissioned half the clocks in the fleet didn't set them right and the relays kept dropping sync."
"So a clock drifting in holdover, the relay keeps stamping with it, and 47 stays quiet."
"Stays quiet. Honouring the quality bits is a settings change. That one's mine, and it's April, and I'd want a month of logs first to see which clocks lie about their own quality."
She said it back so she'd have it right, and then gave him what she already had: the clock's own HOLD24 point from 0412, which fired from the clock side once holdover passed a day.
"So 47 catches the wire, HOLD24 catches the clock going stale, and between them there's a window. Clock in holdover under a day, relays stamping off it, nothing says so."
"Under a day it's microseconds. You told Abel 2.1 microseconds on the twenty-second. I read it. That's not your problem yet. It's a problem at a week."
"Then I'll write it down as a known gap. Not as a fix."
A pause. "Put 47 in my settings file of record as a used point. Documentation only, nothing loads. I'll pick up the quality bits in April if your SOE log gives me a month."
After he hung up she added one line to 0413, under his name. Detects loss of IRIG-B signal only. Does not detect degraded time from a clock in holdover (relays ignore TQ bits, per O. Kessler 1/30). Clock-side HOLD24 covers >24 h. Gap <24 h known, accepted.
Second pencil line on the wall: Kessler: 47 = wire, not clock. TQ bits April.
Trey came in at 22:50 and read both. "That's more than he's said to this desk in a year."
"He wanted it written down before somebody relaxed."
Monday, January 31, 03:20
Abel loaded 0413 at the start of her shift instead of the end of his, because the approver had signed at 17:02 and he wanted to be on the phone for the first poll.
Nothing on any relay changed. The RTU started asking for a line it had never asked for. In the SOE viewer the three new points came up grey and then resolved.
PL BKR2 IRIG LOSS 0. PL BKR6 IRIG LOSS 0. PL BKR4 IRIG LOSS 1.
Not a transition. A standing 1, the first value ever read, stamped 03:21:07.412 by the RTU, because the relay had nothing to report about when the state began. Nobody had asked.
"That's not my load," Abel said from the yard. "It came up that way."
She didn't page. 0413 said no page, and she had written that for this exact case. She did the thing the point was for. She pulled the last sequence that touched all three breakers, the January 19 feeder fault, and laid the stamps side by side on the right screen.
BKR2 opened 31 ms after fault pickup. BKR6 at 33. BKR4 at minus 6. Six milliseconds before the fault it was clearing. Impossible, and on the nineteenth everyone had read it as a rounding thing in the clock and closed the after-action with BKR4 listed first.
If BKR4 had been off satellite time since before the nineteenth, its stamp wasn't early. It was wrong by however far the relay had drifted. She searched the point history for ninety minutes and learned what she should have known before she started: you can't find the onset of something nobody polled.
On the 0413 sheet she wrote the only true thing she had. First poll 1/31 03:21: BKR4 BI 47 = 1 (standing). Onset unknown, pre-dates poll. 1/19 SOE for BKR4 not trustworthy for sequence. Do not use for after-action ordering.
An email to Kessler. Your 47 found one. BKR4 has no IRIG-B now. Not a settings issue. Cable or clock output, Abel checking at 07:00. 1/19 ordering for BKR4 suspect.
7d went green, because the point worked. Under it, in pencil: Works. Found BKR4 already dark. 1/19 stamp was drift, not early.
Trey read it at 06:40, twice. "So the first thing it did was tell us we were wrong last week."
"It told us we didn't know," she said. "That's what I asked it for."
Monday, January 31, 07:24
Abel called from the relay panel, not the yard. She could hear the panel fans behind him and the click of a door latch on the cabinet. "Found it. You'll hate it."
The IRIG-B to BKR2, 4 and 6 was one twisted pair, daisy-chained off port 2 of the control-house clock and landed at each relay on TB3. At BKR4, the IRIG pair sat on TB3-7 and TB3-8, two screws below TB3-5 and TB3-6, where Dorantes had re-landed the 52a and 52b contacts on December 23. The IRIG plus wire was under its screw but not under the clamp. Insulation had been holding it, and it had let go when someone leaned on the bundle. The chain ran on to BKR6 as a parallel tap, so BKR6 never noticed.
Then the part that mattered. Abel had pulled BKR4's own event log, the one the relay kept on its own clock. IRIG-B LOSS asserted 12/23 02:29:51. Two minutes before the photo Trey had taken of Dorantes's finished work.
Thirty-nine days. And it hadn't been free-running. The relay's time-source priority was IRIG first, then DNP time sync from the RTU every ten minutes, then its own oscillator. Nobody had chosen that order. It was the default. That was why nothing had alarmed: the relay always had a clock it trusted. DNP sync over the serial loop was good to tens of milliseconds, not microseconds.
So the minus 6 had never been drift either. Her pencil line was wrong in its last three words. BKR4 had been keeping RTU time, and the RTU's clock and the satellite clock disagreed by about forty milliseconds.
She decided three things before Abel was off the phone.
The after-action would get a correction, not a footnote. She opened the January 19 report, struck BKR4 opened first (−6 ms), and left the strike visible. Under it: BKR4 SOE time from DNP fallback sync (IRIG-B lost 12/23, loose TB3-7). Cross-device order vs BKR2/6 not determinable from SOE. Per BKR4 relay event report (own sample clock): trip asserted 1.9 cy after BKR4 pickup, current zero at 4.7 cy. Interval valid; absolute stamp not.
Inside one relay, the clock could be off and the interval still right. Across relays, only the clock counted.
Every BKR4 stamp from 12/23 02:29 to now got a historian note: DNP-synced, ±50 ms, do not order across devices. That included the January 4 restore. The 1.8 seconds of 11 on that breaker was still good, because it was an interval.
And she asked Kessler for the thing 47 couldn't see: a second point, in April, for time source = DNP, so the next fallback would show up as a fact.
Abel re-landed TB3-7 with a ferrule at 07:41. At 07:42 BI 47 went to 0 in the viewer. A transition this time, with a real time on it.
She crossed out drift in the pencil line and wrote RTU time above it. Then, under 7d, a new line in blue: 7e. A point that never went to 1 hasn't been tested. A relay that never lost a clock hasn't been asked which one it fell back to.
Trey had read over her shoulder. "So it wasn't early and it wasn't drift. It was the RTU."
"It was the RTU's opinion of the time," she said. "That's fine. As long as the report says whose."
Tuesday, February 1, 09:10
Kessler called before she'd finished her coffee, which meant he had read the correction at home, on his own time, and had been thinking about it since.
"You struck a sequence line in a filed after-action."
"I corrected it. The struck line is still there. Anyone can read what we said and why it was wrong."
"It went to the RC on the twentieth. Now they get a revision with the word wrong in it."
"They get a revision that says our stamp came from the RTU for thirty-nine days. Would you rather they find that in April, when they pull SOE for the audit?"
A pause, and then agreement in its practical form. "Fine. Send it through me, not direct."
Then she asked the question she'd been carrying since 07:42 the day before. "BKR4 fell back to DNP because that's the default. What are 2 and 6 set to?"
He didn't know offhand. She listened to him open the settings files and read them out. All three the same. IRIG, then DNP, then internal. Same panel, same daisy chain, same port 2.
"So if port 2 goes," she said, "all three fall back to the RTU together. They stay consistent with each other and nothing alarms. The SOE looks fine. In order, tight, and forty milliseconds off from the transformer relays on port 1."
"They'd agree with each other," Kessler said. "That's the problem." He was quiet a moment. "Agreement was how I believed the nineteenth."
She asked for two things and got one. Yes to the April point, time source = DNP, on all three relays and not only BKR4. No to moving BKR2 and BKR6 to their own clock port before then. "That's a cable job in an energised panel, for a failure we've seen once."
She didn't argue. She wrote it under 7e instead.
7f. A loose wire is one failure. One wire for three relays is a design. Asked 2/1, deferred to April by Kessler. If port 2 fails before then, SOE from 2/4/6 orders correctly with each other and not with the house.
Then a step on her own restore checklist, the one only she used. Before trusting any cross-relay sequence, check each relay's time-source status. Until April, that meant opening the relay's own event log, not the historian.
Trey watched her tape the sheet back up. "Why write down the deferral if he said no?"
"So that if it happens," she said, "nobody has to remember who asked."