The Return Pipe

At 04:10 in 2035, Nerea found the missing water in a place the ledger did not know how to look. The feed meter had counted 2,000 litres entering the process. The return meter had counted 1,940 leaving it. The difference sat in red on the wall screen, beside a button labelled WATER BALANCE COMPLETE. A thin alarm chimed every thirty seconds, just long enough to make the operators glance at the number and away from their gauges.

She turned the button off. Jules taped a strip of white paper across the green label until the numbers earned it back. The demonstration plant was six months into its first continuous run, and every line still wore fresh paint. A hundred metres of pipe crossed the floor between the geothermal feed and the last tank, bending around the test columns and the filters that pulled lithium from the brine. At this scale, the process fit inside a building. The numbers on the dashboard still wanted to grow into a mine.

Nerea's sister Paloma worked the morning water-sampling route along the canal. She had promised to bring the first jar from the upstream gate before noon. Nerea had promised her not to use a guess as a community number. The plant's product could be measured in a beaker; the water used to make it passed through pumps, tanks, wash lines and a return pipe that bent behind a wall. If she signed the green label, the estimate would travel farther than the sample. Paloma had spent six months asking the plant to post the numbers in plain units, not in a phrase like low-impact. Nerea used to tell her that engineering ran on measured uncertainty. Paloma answered that a family reading a water notice had to decide what to do before anyone finished measuring. Their argument had followed them from childhood, across kitchen tables and pump stations, and Nerea had not yet found a way to make both sentences true.

The obvious explanation was a leak. Nerea walked the line with a flashlight, checked every flange with a dry cloth and tapped the steel where the pipe crossed the slab. Nothing wet marked the concrete. A second model blamed the meter. She compared its log to the calibration record; the two counters agreed within their ordinary drift. The missing sixty litres were not on the floor and not in the instruments. Jules brought a wrench and checked the pump's vibration with two fingers against the housing. The line hummed evenly. Nerea opened every drain pan and sight glass, then compared the tanks' level marks with the automatic log. One holding tank sat a little higher than it had at shift change, but not enough to explain the whole gap. The water had moved somewhere inside the process, or the meters were comparing different moments.

She opened the process diagram. In the lab, the pre-treatment unit, extraction column and rinse stage had each been tested on their own. Here the whole line ran at once. Brine moved through the column, the column held a little between cycles, and a later rinse pushed that volume into another tank. The feed and return meters were reading the same minute, not the same parcel of water.

Nerea could explain that to a reviewer, but not prove it. She set a timer beside the glass level marks and put a harmless tracer into the closed test loop. The tracer would not change the lithium product. It would show how long the liquid took to cross each stage. She watched the clear bead of dye disappear into the brown flow, then checked the first outlet every ten minutes. The loop was built to keep the brine moving, not to give every litre a straight path. A column could hold liquid while the rest of the line continued, and the rinse moved it onward in pulses. Nerea marked each valve position with chalk so Jules would not unknowingly change the route while she waited.

The first sample came back blank. So did the second. Her shift partner Jules offered to shut the line down and declare the discrepancy a sensor fault. Nerea shook her head. A shutdown would empty the columns and destroy the timing she needed. She turned the plant's wall clock toward them and wrote the start time in block letters. She kept the pumps running, set a second collection jar under the return spout, and called Paloma to say she would be late.

On the fourth sample, a blue thread appeared in the clear liquid. It widened through the next two jars and then faded. Nerea wrote the time on each label. The pulse had taken almost five hours to make its way through the loop. During that time, the plant had been counting fresh feed against yesterday's return. The water had not vanished; it was still moving through the equipment.

They shifted the dashboard window by the measured residence time. The two totals came together. Nerea felt the knot under her ribs loosen. She had answered the immediate question: the plant did not have a sixty-litre leak. She had not answered how much water a commercial line would use, or whether the demonstration's pipes and tanks could predict it.

A tiny pilot can make a battery salt before a full water balance is known. Nerea pressed SAVE, then watched the screen replace her red warning with a green check. The commercial dashboard, connected to the same account, had already copied it into the proposal marked READY FOR SCALE.