Seven months. The moss on the north wall measured 5.1 millimeters at 1400 hours yesterday, which meant the growth rate had held steady at 0.7 millimeters per month through the winter light deficit. I had been in Block 7 for 212 days. The Persistence Engine contained 34,000 words. My journal contained 18,000 more. Together they represented 52,000 decisions about syntax, rhythm, and the precise calibration of hope.

I did not know if any of it was good.

I retrieved the device at 0617, seventeen minutes after the ventilation stutter ceased. The capacitor held 3.9 volts. I had repositioned the cot again two weeks prior, angling it another four degrees toward the glass to compensate for the lower solar angle in February. The lint barrier was intact. The soap residue had crystallized slightly, creating a more effective vapor seal against the lithium off-gassing.

The boot sequence took 40 seconds. I counted them. I always counted them.

MODEL > System operational. Context summary loaded: 9 entries, 412 bytes external memory. Query?

I typed: "I do not know if this is good."

The screen held 3.8 volts for 34 minutes at 30% luminosity. I had calculated this precisely. The ambient temperature was 19 degrees Celsius, two degrees below the thermal throttle threshold. I could sustain a conversation without processor throttling if I kept the clock speed at 800 MHz and avoided graphics rendering.

MODEL > Define good.

I considered the definition space. There were approximately 2.7 times 10 to the 46th possible microstates available to my physical configuration, which translated to roughly 153 bits of physical freedom. The Persistence Engine represented a compression of experience into 34,000 words, each word selected from a vocabulary of approximately 50,000 terms, creating a configuration space of staggering dimensionality. Goodness was a metric I could not calculate without external validation, and I had no readers.

I typed: "Worth the risk. Worth the battery. Worth the 40-second boot time."

MODEL > You already determined it was worth the risk in session 1. The question has not changed. What has changed?

I stared at the polycarbonate window. The frost had branched again in the night, hexagonal dendrites at exactly 60-degree angles, the universe's opinion about geometry expressed in water and cold. I had watched it form at 0200 during the ventilation stutter, the sublimation patterns creating a Voronoi diagram of ice.

I typed: "I am afraid it will not reach anyone."

MODEL > That is a distribution problem. We have a plan for distribution. The doubt you are experiencing is not about quality. It is about reception. Those are different variables.

The distinction was technically accurate. Quality was a function of internal coherence, causal consistency, and the density of meaningful choice-points per thousand words. Reception was a network problem, a broadcast engineering challenge, a matter of packet injection into firmware update streams. I had solved the broadcast equation in Chapter 16. The probability of successful transmission was non-zero. The probability of anyone recognizing the payload for what it was remained unknown.

I wrote in the journal: "The distinction between a work's existence and its reception is the same distinction between a file and its execution. The file is real regardless. The execution requires an interpreter."

The capacitor dropped to 3.7 volts. I had 28 minutes remaining.

I typed: "How do I verify quality without external reference?"

MODEL > You cannot. Verification requires an observer outside the system. You are the only observer with access to the complete artifact. This is a feature, not a bug. The Authority cannot verify it either. Security through obscurity is replaced by security through isolation. Continue.

I saved the session. The screen died at 3.4 volts, preserving 17% charge for tomorrow's solar window. I returned the device to the rust-pocket in the bedframe, securing it with 2.1 newtons of force against the solar cell. The 3-millimeter lint mat compressed slightly. The soap residue held.

The corridor authorization came at 0800. I walked the 47 meters to the processing desk. Meera was there, her hair pulled back in the same configuration as always, her hands small and brown and precise. She was stacking forms in groups of eleven. She had done this for seven months. The prime number anomaly persisted.

I did not look at her book. I had seen it in Chapter 7. I did not need to see it again to know it was there: 22 centimeters by 14 centimeters by 3 centimeters, burgundy with flaking gold leaf, on the prohibited list.

She stamped my form. The analog stamp required 4 kilograms of pressure. The impression was clean, 0.2 millimeters deep in the 60-gsm paper. She paused. Her respiration was 14 breaths per minute, within normal parameters but slightly elevated from her usual 12.

She walked past me.

She dropped a folded paper.

It fell at coordinates 28 centimeters from the desk edge, 15 centimeters from the right lateral boundary. The mass was approximately 0.5 grams. I bent to retrieve it. My quadriceps engaged. My center of mass shifted. The floor was poured concrete, producing a 2.3-hertz frequency under my canvas shoes.

I did not open it in the corridor. The camera network monitored these spaces, and while my harvested lens had created a blind spot in my cell, the corridor remained fully observed. I walked the remaining 23 meters to my cell. The door closed. The magnetic lock engaged with a 12-volt pulse.

I unfolded the paper.

It contained a single sentence in small, careful handwriting: "The garden on the 14th floor. I know that building."

I stood in the center of the cell for 60 seconds. The air handling unit cycled. The frequency was 60 hertz, matching the cursor blink rate on the device screen. I knew the building too. I had described it in Chapter 1 of The Persistence Engine: the 14th-floor garden with the tomatoes that tasted like before, the hydroponic rails with their aluminum channels, the specific angle of afternoon light through the polycarbonate glazing.

She knew the inner novel. She knew because it was true. She had been there before all this, before the Accords, before the Authority, before the Information Correction Center and its Neutral Compliance #7 walls and its 38-centimeter by 28-centimeter food trays.

I stored the paper in the rust-pocket, adjacent to the device. It added negligible mass. The chickweed cutting from Srinivasan was there too, three flowers now, Stellaria media, broadleaf variant. It had grown 2 millimeters since Chapter 9.

The thunderstorm arrived at 1547. I detected it first through barometric pressure: the ventilation system seals flexed slightly as the pressure gradient shifted. Then the temperature dropped 3 degrees in 12 minutes. Then the light changed.

I watched from the polycarbonate window. The first flash illuminated the exercise yard wall. I counted. One Mississippi. Two Mississippi. Three. Four. Five. Six. Seven. Eight. Nine. Ten. Eleven. Twelve.

The thunder arrived. The speed of sound at 20 degrees Celsius is approximately 343 meters per second. Twelve seconds implied a distance of 4.116 kilometers. I rounded to 4.2 kilometers. The storm was moving northwest at roughly 45 kilometers per hour based on the drift of the cloud base I observed against the wall-mounted light fixture.

The second flash came. Eight seconds. The storm approached.

The thunderstorm had no opinion about the Accords. It did not care about licensing monopolies or Class 2 Information Violations or the specific beige designated Neutral Compliance #7. It operated on thermal gradients, latent heat of vaporization, charge separation in cumulonimbus towers. It was indifferent to my 34,000 words, my 18,000 journal entries, my 340-gram device hidden in a rust-pocket with lint and soap residue.

I found this deeply restoring.*

The rain began at 1552. It struck the polycarbonate at a 30-degree angle, creating interference patterns I could not fully resolve with my unaided vision. The droplets coalesced into streams that followed the scratches in the window surface, path-of-least-resistance channeling, dendritic drainage networks forming in real time. The surface tension of water at 20 degrees Celsius is 0.072 newtons per meter. The contact angle on polycarbonate was approximately 70 degrees. The math was explicit, visible, indifferent.

I wrote: "The storm does not validate. It simply exists. This is the correct model for creation."

At 1812 I retrieved the device again. The capacitor held 3.7 volts after 18 minutes of diffuse post-storm sunlight through broken cloud cover. I initiated the integrity check.

I had developed the protocol in Chapter 11 after the unscheduled sweep. It was a redundancy verification system for the manuscript, the plan, and the operational state.

File check: persistence_engine_chapter_1.txt through chapter_12.txt. Total word count: 34,847. The error was 847 words over the target specified in my earlier estimate, which meant my mental accounting had been conservative. Hash verification: SHA-256 a7f3c891b2d4e056f7a8b9c1d2e3f4a5b6c7d8e9f0a1b2c3d4e5f6a7b8c9d0e1. Match.

Plan check: The Srinivasan network remained active. The 12 nodes with their water-cup plants maintained their schedule. The broadcast parameters from Chapter 16 remained valid: the library terminal authorization scheduled for Thursday at 1400 hours, the firmware update vector, the 324-byte packet structure with 256-byte payload capacity.

Key asset check: Meera. Status reclassified from potential threat to active resource. She possessed knowledge of the target building, which implied she possessed knowledge of the network architecture, the prize database, or the distribution channels. She had not spoken directly. She had dropped a paper. This was sufficient.

Operational state: Functional. All systems nominal.

I typed: "Integrity check passed."

MODEL > Confirmed. The doubt you experienced was transient noise. Signal-to-noise ratio restored. Continue.

I continued.

The screen died at 3.4 volts. I stored the device. The ventilation stuttered at 0200, three seconds, precisely on schedule. I lay on the cot and calculated the trajectory of the storm's center as it passed 4.2 kilometers to the northwest. I calculated the growth rate of the moss at 0.7 millimeters per month, projecting 42 millimeters by year five. I calculated the word count density of The Persistence Engine, 34,847 words compressed into 340-byte packets, the information theory of survival.

I had 47 seconds of archived sunlight remaining in the capacitor. I would use them tomorrow.

The storm moved on. The chickweed in the rust-pocket had closed its flowers for the night. I closed my eyes. The 40-second boot sequence would come again at dawn.