On Christmas Eve the coffee can was on the kitchen table again. Lourdes kept meaning to put it back on the shelf.
The freezer was full of tamales in bags of a dozen, dated in Abuela's handwriting, and the kitchen still smelled of yesterday's masa. Inside the can, the bag with Tencha's page had a new line on it in her own block capitals, written in the truck outside Leyva Tires so it would be true at the time.
12/23. WENT TO LEYVA TIRES WITH RECORDS FORM. DID NOT READ. CALLS TO HIM BEFORE ME: 3 (TENCHA, PABLITO, SISTER). BOOK TO RECORDS BY HIM. UNREAD BY 118.
It was true. It was also on a zip bag in her kitchen, and a line on a zip bag in a kitchen wasn't a disclosure. It was a note she could bring out later, and later it would look like something she had kept back until she needed it.
So the question wasn't whether to tell. Rubén had said he'd take the ledger in after Christmas, Records reopened on the 28th, and four days was plenty of time for somebody at 226 to hear from somebody in Tolleson that the notary had been at the tire shop.
The harder question was her name. The holder-side overlap rule was sponsored by license 118. She had drafted it on Tuesday night with Tencha's night in her head, and on Wednesday she had carried the form to the witness. Whatever was in the green book, 226 could attack a rule sponsored by the person who'd carried the form without ever reading it. If it went down that way, it went down for every family shop that would ever need it.
She wrote it out the way she wrote anything she might talk herself out of overnight.
WITHDRAW AS SPONSOR. ASK RECORDS TO HOLD IT OPEN OR LET IT LAPSE. DO NOT ASK DELFINA. DO NOT ASK ANYONE.
The line about Delfina hurt more than it should have. Delfina would sign, and then Delfina's terms would sit on the same file as the visit, and they'd have two witnesses compromised instead of one.
She photographed the bag on the oilcloth with the can beside it for scale. 07:12. To R. Aguirre at Records, copied to 226's counsel, copied to Delfina so she wouldn't hear it from someone else: the note, the three calls, I did not read the ledger, I am withdrawing as sponsor. Her thumb stayed over SEND long enough for the screen to dim. Then she pressed it and put the can back above the rice, behind the good platter, where it had been before December.
Rosaura Aguirre called on Monday at five past ten. Lourdes was in the Food City lot on 35th Avenue with a bag of limes on the passenger seat that Abuela would say were too small.
"I've had your note since Thursday," Aguirre said. Her voice was dry and older than Lourdes had pictured, with a keyboard going underneath it. "I've had 226's counsel since Saturday, which shouldn't be possible on a holiday weekend, and yet. He wants the method out, Amendment 1 out, and you referred to the State Board for contacting a witness. He quotes Ms. Ochoa's disclaimer twice."
Then she said what Records would do, and told Lourdes not to argue with it.
Behind her a delivery robot waited at the crosswalk with its flag up, patient as a dog. Rubén had walked the ledger to intake at 08:30 with the form inside and Aguirre written on the back of his hand. Intake had sealed it and given it to a clerk who wasn't Aguirre. Nobody would open it for the base-tie run. It belonged to the overlap proposal, and since the proposal had no sponsor, for now it belonged to a shelf. The disclosure was timely because it came before the record and before any complaint. Counsel's motion leaned on the visit and left out who had reported it. Aguirre would point that out. The run would go ahead with Lourdes's method, her amendment and Delfina's three terms, and the results would come out Thursday as noticed.
"One more thing," Aguirre said, "and it isn't a favor. Your amendment says the ties are tide- and pressure-corrected. Did the 2038 to 2044 heads log pressure?"
The first-generation heads had a barometer on the board. In fourteen years she had never checked whether the firmware wrote it to the drift file, because in fourteen years no base she trusted had moved by a hectopascal's worth.
"I don't know."
"Find out by Wednesday. If they didn't, tell me what you'd use instead. I'd rather hear it from you than from him."
On Tuesday she took a paid job, the first since the 18th. It was a corner lot on Roeser Road in Laveen where a builder wanted the fill notarized before slab: $640, rank seven, with WEATHER ADVISORY in yellow on the dispatch card. She needed the money, and she needed the front.
She set one head on the lot's control nail and one on the county base in the mast arm across the road, so she had two instruments under one sky. Across the road the mast-arm light had a municipal head in its base like every pole in the county now, the inversion office's grey collar listening down for voids and pipe breaks on its own nightly schedule. It had nothing to do with her job, and it would be weighing the same front. Each of her heads gave her three columns: raw, tide-corrected, and tide-and-pressure-corrected. That made six. She had never needed the last one.
At 20:40 the wind came first, then the dust, then the cold, all in about four minutes, and the truck rocked once on its springs. The barometer pane had read 1014.8 since half past six, and it began to climb. It took her a second to remember that pressure rises behind a cold front, not ahead of it. By 21:30 it read 1020.9.
Both raw columns dropped about 1.8 microGal. More air overhead pulled up on the proof mass a little harder than its weight pressed the ground down. The tide-corrected columns dropped with them. For the twelve minutes the gust front was on the lot, from 20:40 to 20:52, every column on both heads jumped around by a microGal and more, the heads shaking on their tripods as much as the truck had. She threw that window out and marked it on the sticker roll: GUST 20:40–20:52, DISCARDED. Before it and after it, the fully corrected columns sat within four-tenths of each other on both heads. So 0.3 per hectopascal wasn't something she had watched happen. It was a fit across a gap, a before and an after with the noisy part cut out, and that was the only way anybody in the field ever got it.
The builder's fill tolerance was three microGal. Without the sixth column the lot would have read 1.8 lighter after the front than before it, and she would have signed for a slab on a number that was true for the wrong reason. She signed at 22:52 and added a line he wouldn't read: FRONT 20:40, +6.1 HPA, GUST WINDOW DISCARDED, PRESSURE-CORRECTED, BOTH HEADS.
Then, with the heater running, she opened the archive from her own retired 0219, the head she had run in 2042, and looked at the drift file for 15 May. Time, raw, tide, temperature, tilt, battery.
No pressure column. February had none, and neither did October 2043. The board had a barometer, but it was there for the sensor chamber, to correct the proof mass for buoyancy and to catch a leaking seal. It was the instrument looking after itself, so nobody had thought to write it down as a reading of the sky. Six years of West Valley drift files, hers among them, and not one of them recorded what the air weighed.
The second source was obvious, and she was embarrassed it had taken a front in Laveen to make her think of it. Sky Harbor had logged sea-level pressure every hour since before she was born, and the archive was public. Nobody had ever lined it up against a drift file.
The airport was 25 kilometres from Tolleson. A front crossed the valley in under an hour. Sky Harbor could tell her when the air changed and roughly by how much. It couldn't tell her what the air did at a particular base, to a tenth. So it could give her a direction and a rough size for a night, but it couldn't give her a number for any particular base.
At six on Wednesday, with Abuela asleep, she made the table.
On 15 May 2042, Tencha's night, the pressure read 1004.2, 1004.4, 1004.5, 1004.5. It held steady, so OVERLAP stayed OVERLAP.
On 9 October 2043, 150's night, it rose 1.2 hectopascals over the evening, about four-tenths of a microGal. That night was already grey and stayed grey.
On 21 February 2042, Tencha's other night, the one she had called CORROBORATED, it read:
18:00 1012.3. 19:00 1012.6. 20:00 1015.9. 21:00 1018.4. 22:00 1018.9.
A front had come through between the evening tie and the late one, six and a half hectopascals of it. That was about two microGal the drift file had never known about.
She did the arithmetic three ways, though she didn't want to. The corroboration rested on the tie offset matching the 0402's January value, +4.6, and not the 0517's −1.2. The front had come between the ties, so it was the late tie that took the step. The pressure had gone up, so the late tie had read low. That wasn't noise. It was a bias, and she knew which way it went: about two microGal, six and a half hectopascals times 0.3. Shift the late tie back by that and the offset still sat on the 0402. The 0517 was more than four sigma off. It was still one head. But the number she'd printed had been two microGal off on one tie and she'd never known it. And the sigma beside it, the one Delfina had made her print the same size as the result, was too small, because the airport could say which hour the front came through but not which minute, and Sky Harbor was 25 kilometres from the base.
The result stood, but the error bar beside it had been false.
AMENDMENT 2. 2038–44 FIRST-GEN DRIFT FILES DO NOT LOG PRESSURE. USE SKY HARBOR HOURLY BETWEEN TIES. OVER 2 HPA: SHIFT THE LATER TIE 0.3 µGAL PER HPA, IN THE DIRECTION THE AIRPORT GIVES. THEN WIDEN BY HALF THE SHIFT FOR TIMING AND DISTANCE, IN QUADRATURE. FRONT IN THE SAME HOUR AS A TIE: DIRECTION UNKNOWN, NOT RESOLVING. THE AIRPORT GIVES A DIRECTION AND A SIZE, NOT A NUMBER FOR THE BASE. CORRECTION, 63 / 21 FEB 2042: CORROBORATED. LATE TIE +2.0. SIGMA 0.9 → ~1.3 (√(0.9² + 1.0²)). FRONT 19:00–21:00.
At 09:40 it went to Aguirre, 63, 150, Delfina and 226's counsel, all in the same minute. She attached the airport table and her Laveen columns, because Laveen was the only proof she had that 0.3 per hectopascal was the right number here. Under the table she wrote what it actually showed: ONE FRONT, ONE VALLEY FLOOR. 0.3 is the number for a weather system the size of a continent, and a narrow front could give a different one at any particular base.
Pablito called at 09:52 for his aunt, who didn't read her phone on Wednesdays. For a while he didn't say anything.
"It still says one head."
"It still says one head. It just says it less loudly."
"You couldn't have left that alone."
"No," she said. "That's the whole method."
On Thursday at three the light came flat through the window onto McDowell. Pablito was there for 63 in a pressed shirt, with Tencha's licence card in his pocket. 226's counsel was on the wall screen, a young man in a quarter-zip holding a law-firm mug. Delfina didn't come. She sent a sticky note through the clerk: BOILER OFF. TABLES UNCHANGED. D.O.
Aguirre put the results on the screen without commentary. There were eleven nights and twenty-two records, two per night, 50 to 94 kilometres apart.
ONE HEAD: 6. TWO HEADS: 2. NOT RESOLVING: 3.
She read out the grey ones before anyone could ask. Two were on TB-22, Lateral 8: under Amendment 1 the coefficient hadn't been measured, so they stayed grey until the gate change on January eleventh. The third was 3 March 2041, when Sky Harbor had a seven-hectopascal front arriving in the same hour as the late tie. The airport couldn't say whether it came before the tie or after, so under Amendment 2 there was no direction to shift in. All that was left was the width, and with that much width two heads couldn't be told apart.
Then the six. On six nights, 226's two records sat more than fifty kilometres apart and carried the same base-tie fingerprint to within a microGal. That meant one instrument, two sites and one name on the licence.
Counsel said the method had been amended twice in ten days by the person who proposed it, and that this was the opposite of a stable method.
Aguirre let him finish. "Both amendments made results less certain, including results that helped the complainant. One widened the sigma on 63's own night. You can call that unstable. Each one is dated, and I've written down for the board which way each one cut."
In September Lourdes had guessed seven. It was six.
Pablito leaned over. "Two heads on two nights. They weren't lying those two."
"Or they had two crews those two nights. It doesn't say who."
"It never says who."
While counsel was gathering himself off the wall, Aguirre slid one sheet across the table. It wasn't the results. It was an intake slip.
LEYVA, R. / LEDGER (INSURANCE SIGN-IN), 2039– / RECEIVED 12/28/48 08:30 FROM HOLDER IN PERSON / SEALED. ASSIGNED: OVERLAP HOLDER-SIDE PROPOSAL (NO SPONSOR). NOT OPENED FOR BASE-TIE RUN. 118 DISCLOSURE ON FILE 12/24, PRIOR TO RECEIPT.
"That's all I'm allowed to write," Aguirre said. "It's also all you need. If anyone on the board picks the proposal up, it gets opened in front of them, not you."
"Has anyone picked it up?"
Aguirre was already putting her glasses away. "Ask me in January."
Outside, McDowell had started the long weekend early. On the light pole at the corner, the municipal head's collar had gone from amber to green for the year's last night tap, three hours ahead of schedule, because someone in the inversion office wanted to be home by midnight too. It would listen down through the street for water and weigh the air above it as it went, and it would write both down. Lourdes stood under it with the intake slip folded in her jacket until the light changed, and then she drove home to put it in the can.