Chapter 105: Teach the Water to Lie

I reached the compound at nine that night and wrote down every observation before I ate. Memory simplifies details. I needed the details.

First response: four minutes and eleven seconds from the initial drag to movement, from approximately sixty metres.

Approach speed: slow until the final ten metres, then fast.

Persistence after a signal stopped: eleven minutes over the pipe.

Divided attention: confirmed. Three points split two animals for approximately one minute.

Priority: strongest recent signal, not the nearest.

Guardian refusal threshold: it travelled roughly fifteen metres to drive away another Lurker and returned immediately.

I sat with the last entry for a while.

Fifteen metres for a rival. Nothing for uncertain prey.

Making a louder disturbance would not solve that. The guardian would never value a suspicious meal more than its clutch. I needed something close enough to seize quickly, convincing enough to justify the risk, and apparently moving away.

The dragged pipe had created too clean a line. Liora had called it a boat before either of us had seen a real Lurker investigate one.

Real prey kicked, stopped, weakened, and tried to escape.

I slept before touching the tools. The following morning, I began building the machine. It broke four times before it worked.

The final version used three sealed bladders, each mounted inside a hinged cage made from toolbox frames and reinforced with metal strapping. A tension line ran from every cage to the dry apron. Pulling a line folded the cage and compressed its bladder, displacing close to two litres of water. Releasing the line allowed a bent leaf spring to open the cage so the bladder expanded again.

One pull produced one pressure pulse.

The three lines connected to a camshaft mounted on a frame. I made the shaft from a length of salvaged driveshaft and fitted three eccentric lobes at different angles. They were deliberately uneven. As the shaft turned, one lobe produced a hard pull, another overlapped it, and the third created a weaker pulse after a four-second dead sector.

The result was not a repeating beat. It was a struggle: two hard kicks, a pause, a weak movement from somewhere else, then another attempt.

The remaining problem was the one Liora and I had written down. A man turning a crank on dry ground could not harvest roots beneath a greenhouse.

I solved it with a wound drum and a counterweight.

The hand crank lifted a scrap axle housing filled with wet concrete—about forty kilograms and unpleasantly interested in my toes. Releasing a pawl allowed the weight to descend. The drum turned with it, which rotated the camshaft and operated all three lines without me.

The mechanism still needed braking. Forty kilograms wanted to reach the floor much faster than three wet cords wanted to move, and letting the drum spin freely would turn the first pulse into a single violent collapse. I built resistance into the axle with leather washers, a spring-loaded friction arm, and more adjustment than the arrangement deserved. Tightening it by half a turn changed the runtime by almost a minute.

That was useful in the yard. In the field, silt and seventy metres of line would add resistance I could not measure. The nine-minute limit was not pessimism. It was the only number that remained safe after I accounted for everything I did not know.

The first bladder ruptured because the cage hinge crushed it against a sharp frame edge. A split length of rubber hose fixed that.

The second failure was mine. I cut all three cam lobes from one template, producing a perfect metronome. I recut two by eye and made one deliberately crooked.

On the third test, the counterweight ran away. With no load, it reached the ground in fourteen seconds. With the lines attached, it lasted less than six minutes. I reduced the drum diameter, increased the weight’s travel, and brought the full runtime to eleven minutes and forty seconds.

I wrote NINE MINUTES on the frame in grease pencil. A machine working across four metres of clean tank was not the same machine working across seventy metres of silt and wet guide eyes.

The fourth failure was a line tangle. Three cords under intermittent tension will eventually find each other. Guide eyes at every anchor solved it.

I arranged the finished cam sequence so the strongest pulses came early. As the counterweight descended, the struggle weakened and its apparent centre shifted away from the greenhouse.

Something had been injured near the nest. It was trying to escape. It was getting weaker.

I tested the pattern from above, from beside the tank, and with one hand pressed against the metal wall. Watching the surface made it look mechanical because I knew where every line ran. Feeling the pressure without seeing the device was different. One pulse struck high against my palm, another arrived through the base, and the pause between them was just long enough to suggest exhaustion rather than failure.

It was not proof that a Lurker would believe it. It was the first time I understood why one might.

The secondary decoy took an hour and did not connect to the main machine. One bladder, one cage, one loaded spring, and a pin trigger produced a single violent burst. It belonged beside my withdrawal route. Its only purpose was to steal several seconds from anything already chasing me.

Then there was the walkway.

Collapsing the greenhouse itself would crush the eggs and destroy the root bed. Instead, I planned to drop the middle section of the service catwalk into the pursuit channel after I crossed it.

Both middle stanchions were badly corroded at the waterline. A bar through each collar and a doubled line leading to the dry path should pull them down together. Eleven metres of steel grating would fall into water one metre eighty deep, leaving the guardian an obstruction without touching its nest.

I checked the fall direction on paper three times. The span had to land between the guardian and me, with me already on the apron side.

The release line also needed to remain reachable after I came out of the water. I marked a point twenty metres up the path where the ground rose enough to keep my footing and rehearsed reaching it with one arm held against my ribs. If the guardian struck me or the walkway during the exit, I could not assume both hands would still work.

The harvesting rehearsal was less dramatic and more important.

I filled a shallow tray with ordinary soil and threaded pale cord through it to imitate the Marrowlace network. Pell’s two mineral-solution cases sat on either side, weighted with their one-way valves facing up.

The procedure was simple: define a stable block of substrate, undercut it without tearing the roots, lift it with both hands, guide it through the valve, and seal the case within one minute.

My first attempt took two minutes and ten seconds. The block fell apart before reaching the valve.

I had arranged the tools badly. I started with the knife, switched to the trowel, then needed the knife again. I reversed the order: trowel on three sides, knife once for the undercut, both hands to lift.

The eighth attempt took fifty-one seconds. The twelfth took forty-four.

Then I closed my eyes. In the greenhouse I would have about a metre of visibility, and most of the work would happen by touch.

Ninety seconds. Seventy. Fifty-eight.

The cases would rest on the bed before I cut, one on each side. A sealed case could go into the spatial ring, but the ring could not harvest the root or protect it from contaminated water. Every block still had to pass through Pell’s valve first.

I wrote the abort conditions in capitals.

GUARDIAN HAS NOT PASSED THE FORTY-METRE MARKER.

BOTH WITHDRAWAL LANES OCCUPIED.

LESS THAN FOUR MINUTES OF DRIVE AT THE FIRST CUT.

ANY VALVE FAILS TO SEAL—NO SECOND BLOCK.

I wrote them before the operation because I knew what I became when a margin narrowed after I was already committed.

[Blind Call Recovery Remaining: 19:41:07]

The charge would not be ready in time for the operation, and I did not need it to be. Even if it had recovered, the Call might produce someone who knew nothing about flooded fields. I planned to do this alone.

The full rehearsal began at four that afternoon.

I wound the drum and released the pawl. The concrete counterweight descended while the camshaft turned, making the three bladders pulse into the tank in a sequence I could not predict by watching the surface.

For the rehearsal, I laid every component at the measured distances from my map. The harvesting tray stood where the root bed would be. The secondary trigger sat along the return path. A rope tied to two scrap uprights replaced the corroded walkway supports. Nothing reproduced the water depth, the guardian, or the fact that I would be holding my breath, but the order had to survive without thought.

I crossed the yard, completed two simulated cuts, sealed both cases, triggered the secondary decoy on my return, and pulled the walkway-release line.

The counterweight remained forty centimetres above the ground when I stopped the clock.

I ran it again after resetting everything. On the second attempt, the left field case snagged against my belt and cost six seconds. On the third, I reached automatically for the knife before the trowel and stopped myself. I repeated the entire sequence until the correct hand found each tool without searching.

The machine could make a tank lie convincingly about a dying animal. I dismantled it, ate, and slept.

Before dawn, I would find out whether the guardian believed it.