Monday, January 24
DSP-1236 had said Awaiting approver (Display) for eleven days. Marisol read it again at 07:20, after turnover, with her coat still on and the EMS room settling into its day sounds: the HVAC stepping down, the QSE line quiet, Hall 4's load trace flat and green at the top of the left screen because nobody had called it off since Saturday.
The ticket was small. It built a group alarm called DR-CALLED: when a breaker on a called point read 11, both contacts at once, the display would page somebody. Right now it relied on Trey noticing. Trey had noticed twice in January. A habit that worked twice was still a habit.
She had commented on it three times. She did not comment a fourth time. She scrolled back to November, to DSP-1228, the last display change that had gone through, and found the signature: D. Pruitt, display on-call. The change calendar had a desk number for display on-call in Corpus. She dialed it on the desk phone, not her cell, so the call would show up in the log if anyone ever asked.
"Pruitt."
"This is Marisol Cantu, night lead at the coast campus. I've got a display ticket waiting on approval since the thirteenth. DSP-1236."
Typing. "I've never seen it."
"It's assigned to Display."
"Display's a group mailbox," Pruitt said. "Was. They retired it at the December reorg. Anything routed there goes to an inbox nobody reads." A pause, more typing. "Okay. It's waiting on an inbox nobody reads. I'm putting it on me."
Marisol watched the field change on her screen while they were still talking. Awaiting approver: D. Pruitt. Eleven days, and the whole delay was one dead address.
"Before I sign," Pruitt said, "does the group page on any 11, or on 00 too?"
Marisol didn't need to look, but she looked anyway. The desk note was taped to the hook wall beside the night headset, a sheet titled Clock Source When EMS Lies in her own block capitals, with struck lines and smaller lines under them. She read line 11 off it. "Any 11 on a called point. Or 00 for over one second on a called point."
"The 11 I can do today. The 00 timer needs the SOE tag to read UTC."
"It does since the nineteenth. Saucedo's build."
"Then both." The ticket updated: Approved 07:41 D. Pruitt. "Anything else while I've got the change tool open?"
There was. Marisol read her own Saturday comment off DSP-1244 aloud: page at 24 hours of satellite-clock holdover, not at first loss. A clock that had lost GPS for an hour was still good to the microsecond. A clock that had been alone for a week was not.
"1244 can't page on anything," Pruitt said. "The clock alarm isn't a point. There's nothing for me to map. That's an RTU change. Whoever owns the control-house RTU at your PL site has to bring it in first, and it'll need its own CR."
"Abel," Marisol said.
"Then it's Abel's CR. And check who approves it before you wait on it."
On the turnover sheet, under the CLK row, she wrote: DSP-1236 signed 1/24 07:41 Pruitt. Approver field was a dead mailbox. Check approvers on 1244 before you wait on it. Then she sent Saucedo one line on the conversion CR he owned, the scheduler one: Release note should name who approves, by person, not by group.
He replied with a thumbs-up, which was not a yes.
Tuesday, January 25
Abel was in the control-house lot at 06:05 because that was where his truck was at six. She had him on speaker at the hook wall.
"Pruitt's right," he said. "There's no point. There's a contact."
"Tell me what the contact does."
A cabinet door, the scrape of a flashlight on steel. He read from the back of the clock. The 2019 wiring was one dry contact, normally open. It closed when the clock lost GPS.
She stood there with the pen uncapped. "So if the clock loses power, the contact stays open."
"Yeah."
"And the RTU reads open as locked."
Silence. Then: "Yeah. Dead clock reads healthy."
That went into the CR first, ahead of the thing she had called about. She typed while he read terminal numbers. The clock had four programmable outputs and the site used one. They split it. PL CLK FAIL went on a normally closed output, so a power loss, a cut wire or a pulled card would all open it and alarm at once. PL CLK HOLD24 went on a second output that the clock itself would assert after twenty-four hours in holdover. The page would come off the clock's own timer, not off some EMS logic that nobody would remember to test after the next reorg.
"That's your Saturday line," Abel said.
"It's a better Saturday line. It's in the hardware."
She typed the test before he could tell her it was obvious. Pull the antenna lead; confirm HOLD in the clock's event log and no page. Pull clock power; confirm CLK FAIL at the EMS inside one scan. Restore; confirm both clear.
Then she did what she had written on the turnover sheet the day before, and opened the approver field before she hit submit. Substation Eng (group).
"Who reads that?"
"I don't know," Abel said, honestly.
She searched the change tool for the last RTU CR on the PL site. RTU-PL-0311, 2026, approved by R. Benavides. She put Benavides in the field by name, the group in CC, and wrote in the description: Approver named per person; group mailbox retired at Dec reorg, see DSP-1236. Submitted 06:38 as RTU-PL-0412, linked to DSP-1244.
On the desk note, under the line marked 7b, she wrote smaller: 7c. CLK FAIL = no clock, don't trust ms at all. HOLD24 = clock alone a day, still good, check the panel.
Trey read it at turnover with his coffee. "You're up to c."
"It keeps having letters," she said. "That's how you know it was never one line."
Wednesday, January 26
Tuesday evening the QSE had called Hall 4 off for the 01:15 UTC interval and released it at 02:40 UTC, a little after twenty to nine Central. The evening desk had handled it. It was the first called restore since Saucedo's December scheduler release, which meant it was the first time line 5 on the desk note got tested by a real night instead of a staging replay.
Line 5 said: Restore time is the scheduler's resume stamp, labeled SCHED CLOCK. Drafter must not convert it to EMS time or round it to a bin. She had written that on December 17 and initialed it, and it had sat there in blue pen for six weeks, unproven.
At 06:10 she opened the 06:00 morning-report draft before Trey could sign it. The drafter was the campus tool that wrote the report's narrative paragraph from the logs. It wrote well. That was the trouble with it.
The shed half was right. Breaker open 01:14:58.31 UTC (relay record). Keep bin 01:15 (meter). Each time had its instrument next to it. She had fought for that in December.
The restore half was not. The EMS log pane said JOBS RESUME START 02:41:07 SCHEDULER (SCHED CLOCK). The drafter had carried it into the report as 20:41:07 EMS. The label was gone, the clock was renamed and the hour was converted to Central. Two of her three rules were broken in one line.
Nothing was wrong by more than a second. She sat with that for a moment, because it was the part that would be hard to explain. In March, a settlement analyst would put the scheduler's second next to the EMS's second, see a restore that started before the release, and either flag it or, worse, fix it.
She didn't fix the report by hand. A hand fix taught the drafter nothing and taught the next person that the report was right. She opened Saucedo's CR thread and pasted both lines, the log pane and the draft, one above the other, and wrote one sentence: Drafter strips SCHED CLOCK and converts to local; spec 12/17 says carry the label and the clock as written. First live restore, 1/26 02:41. She tagged it as a defect against the release, not a request.
In the report, she let the drafter's line stand and added a comment under it: Time is SCHED CLOCK 02:41:07 UTC as logged; drafter conversion is a defect, see CR. The wrong line and the correction would both be on the record.
Turnover sheet, CLK row, her fourth entry: LOCK 10.
Trey signed the report with her note in it. "You could have just typed it right."
"Then the next one comes out wrong and nobody's here at six."
Beside line 5, in pen, small: Tested live 1/26. Drafter fails. CR open. MC 1/26. Under that, she left the space she kept for green ink empty.
Thursday, January 27
Saucedo called at 05:50 instead of writing, which meant he had read the defect and didn't like it.
"The drafter converts everything to Central," he said. "Ops asked for that in 2026. Every time in the morning report is local. If I carry SCHED CLOCK through raw, the report has two time zones in one table."
The desk note was in front of her and she didn't need it. "It already has three clocks in one table. Relay, meter, scheduler. Ops just can't see them, because the drafter makes them all look like EMS."
"I can add a footnote."
"A footnote is how the label got lost the first time. Nobody copies a footnote into a spreadsheet." She turned her chair so she was facing the hook wall and not the screen. "Keep the conversion if ops wants it. Put the label in the cell. 20:41:07 CT (SCHED CLOCK). Same cell. If somebody copies the cell, the clock goes with it."
He was quiet. She could hear him scrolling something. "That's a format string," he said finally. "Not a redesign. I can get it in the February build."
"And the rounding?"
Line 5 had three rules and Wednesday had broken two. The third was the one she worried about, because rounding was the kind of error that looked like tidiness. He checked while she waited. The drafter hadn't rounded this one; the stamp had just landed on a whole second. He'd add a test that fed it 02:41:07.4.
"Write label in cell into the CR," she said. "In your words, not mine."
"Why mine?"
"So the next developer reads it as the owner's decision and not the night lead's complaint."
He did it while they were still on the phone. The thread updated at 05:58.
She didn't touch the desk note. The green-ink space under line 5 would stay blank until the February build ran a real restore. On the turnover sheet, CLK row: LOCK 11. Beside it: CR fix agreed, Feb build.
Trey read it at six. "So you won."
"He agreed to put two words in a cell," she said. "Ask me in February."
Friday, January 28
Benavides signed RTU-PL-0412 on Thursday afternoon, by name, two days after it was submitted. Marisol checked: that was nine days faster than anything on the PL site that had gone to a group.
Hall 4 wasn't called and the QSE had nothing forecast before 07:00, so 05:30 Friday was the window. Abel was in the PL control house with the cabinet open. She had him on speaker and the CR on the left screen, and she read him the test one line at a time.
"Antenna lead."
"Off. 05:34."
She leaned toward the screen. "Panel?" "HOLD," Abel said. On the EMS, nothing. No page, no alarm. That was the result she wanted. A clock alone for four minutes was still a good clock, and a system that woke somebody for it would get muted by March. She typed: 05:34 HOLD, NO PAGE, PASS.
"Back on." LOCK came back in ninety seconds.
"Power."
"Pulling it. 05:41:20 by my watch."
PL CLK FAIL came up on her screen at 05:41:22, inside one scan, red. Until Tuesday that point hadn't existed. Trey, who had come in early and was standing behind her with his coat on, made a small sound.
Then the SOE list moved again, two lines under the red one, and she stopped typing.
Three relays, PL BKR2, BKR4 and BKR6, each logged IRIG-B LOSS within the same second. She hadn't written a test for that. She hadn't known it was there.
"Abel. The relays are logging their own time input."
"Yeah, they always have. It's in their SOE. Nobody ever mapped it to anything."
"What do they do when they lose it?"
"Fall back to their own crystal. Good for a few hours. Not a few days."
She looked across at the hook wall, at line 7c. CLK FAIL = no clock, don't trust ms at all. It was still true. But it was blunt. It said stop trusting the milliseconds when the EMS told you the clock had failed, and the EMS was the thing the whole desk note was about not trusting. The relays had been carrying a better answer all along: a stamp in their own record that said exactly when each one had lost satellite time, per relay, without asking the EMS anything.
"Power back," Abel said. "05:44."
All three relays logged IRIG-B OK. CLK FAIL cleared at 05:44:31. She typed PASS on all three test lines in the CR, and then added a fourth that nobody had asked for: Relay IRIG-B LOSS is self-logged in SOE, not mapped to EMS. Use it to bound relay ms during clock failure.
Then she reached for the green pen, the one in the cup that nobody else used.
Under 7c, she wrote the first green line on the whole note: Tested 1/28 05:41, Abel + MC. FAIL pages in 2 s. HOLD doesn't page. Passed.
Beside it, in blue, because it hadn't been tested in the field yet and so wasn't allowed to be green: 7d. Relay says IRIG-B LOSS? Its ms after that are its own clock. Check the stamp, not the EMS.
Trey came around the chair and read it. The note had been on the hook wall since November. It had struck lines, smaller lines, three different pens, a strip of blue tape where a corner had gone soft. This was the first thing on it in green.
"That's new," he said.
"That's the first thing on this note that's actually been proved." She capped the pen. "And it found the next thing."
He looked at 7d. "So now there's a d."
"There's going to be an e," she said. "Line 5 goes green in February or it doesn't, and either way somebody writes it down." She turned back to the screen. The CR was still open, and Abel was still on speaker, closing the cabinet. "Abel. Who approves mapping a relay SOE point to the EMS?"
A pause. The cabinet latched.
"I'll find you a name," he said.