The surface layer broke on the nineteenth delivery.
Not completely — the surface layer's key became visible on the nineteenth delivery, which was different from the code being broken. The key was the door. What was behind the door required separate work. But the door had been there for nineteen deliveries and finding it produced a specific quality of satisfaction that he noted and set aside because satisfaction was not useful and the work was not finished.
The frequency baselines had taken eleven deliveries to establish with sufficient reliability. Characters appearing at rates that deviated from the natural frequency distribution of the language produced a pattern once the sample was large enough, and the pattern pointed to the substitution key the way a shadow pointed to the object casting it — not directly, but with enough information to reconstruct the object if you understood the geometry of shadow and light.
He had understood the geometry.
The surface key was a positional substitution — each character's identity determined by its position in the message relative to a counter that reset at irregular intervals. The irregularity of the reset was itself the secondary layer, the thing that had prevented the surface key from being visible until the nineteenth delivery provided enough material to establish the reset pattern.
He broke the surface layer.
What emerged from beneath it was not plain text.
It was a second layer of coding, different in structure from the first — not substitution but transposition, the characters now the correct characters but arranged in an order that was not the order of the message's meaning. The transposition key was separate from the substitution key and did not emerge from frequency analysis because frequency analysis was insensitive to transposition by design.
He needed a different approach for the second layer.
The different approach took six more deliveries to identify and four more to apply with sufficient confidence that he trusted the results he was producing. During those ten deliveries he continued the route without change, continued the opening and resealing, continued the evening reconstruction work, and said nothing to anyone about what he was doing in the hours between the facility's ambient sounds.
On the twenty-ninth delivery he read the first fully decoded message.
He read it three times.
The message was not about him. Was not about the facility. Was not about anything that touched his immediate situation in any way he could identify. It was operational communication about officials connected to Vijayavarman's faction — specific officials, specific actions, the movements of people and resources through the court's administrative structure in ways that were clearly significant to whoever was receiving the information but whose significance he could only partially understand from the message alone.
Partially.
He held the partial understanding and thought about what it meant.
He understood enough to know that the messages were intelligence. The specific kind of intelligence that described a minister's factional relationships and the movements of people within those relationships — who was moving toward Vijayavarman and who was moving away, who was being compensated through which channels, who was being pressured through which leverage. The architecture of a factional operation described in the compressed professional language of people who conducted this kind of work and communicated about it routinely.
He understood enough to know that the intelligence was operational — that someone was using it, acting on it, making decisions based on what these messages described.
He understood enough to know that the source of the intelligence was inside Vijayavarman's operation.
What he did not understand was the full shape of what was being built from it. The messages were components, each one describing a piece of a larger picture, and the larger picture required more components than he currently possessed and a context he had not fully assembled.
Partial understanding. Sufficient to know something large was happening. Insufficient to know what.
He continued the deliveries.
On a morning three weeks after the first decoded message, during the water distribution hour when the parallel corridor's population moved through the shared space in Manickam's established sequence, Manickam passed close enough for an exchange that was not an exchange — two men moving through a shared space, neither looking at the other, the words spoken in the register of ambient conversation that the facility's sounds absorbed before they could carry.
"The new delivery," Manickam said. Not a question.
"Continuing," he said.
A pause in which Manickam's movement through the shared space continued without altering.
"The sealed messages," Manickam said. "You open them."
It was not a question either.
He said nothing.
"Four deliveries," Manickam said. "You started opening them on the fourth delivery."
The accuracy of this was itself information — Manickam had been watching the deliveries with sufficient attention to identify the specific delivery on which the opening had begun, which meant Manickam had access to observation of the route that he had not identified. A gap in his understanding of who was watching what.
He noted the gap.
"The code," Manickam said. His voice had a quality that Chandragupta had not heard in it before — not the authority of a man conducting his operation, not the careful neutral register of someone managing a professional exchange. Something that was assessment and something that was the response to an assessment that had produced an unexpected result. "You broke it."
"Partially," he said.
Manickam was silent for a moment in which they continued moving through the shared space in the ordinary way of men going about their morning.
"Very smart," Manickam said, "or very stupid."
"The distinction matters," he said.
"Yes," Manickam said. "It does."
They completed the water distribution hour without further exchange. He returned to the south corridor and Manickam returned to the parallel corridor and the facility continued its morning in the ordinary way, and he sat against the back wall of the last cell and thought about the gap in his understanding of who was watching what and what the gap implied about the completeness of the picture he had been building.
The picture was not complete.
He had known this — had always known it, had been building toward completeness rather than assuming it. But Manickam's knowledge of the fourth delivery's significance suggested a layer of observation he had not accounted for, a watching that ran beneath the watching he had identified, operating in the specific way that things operated when they were designed to be invisible to the person they were watching.
He thought about the messages and what they described and what the partial understanding he had assembled implied about what he was not yet seeing.
The messages described Vijayavarman's faction from the inside.
The inside implied a source.
The source was inside Vijayavarman's operation and was feeding intelligence through the courier route that ran through the south corridor's last cell to a drop point and from the drop point to wherever it ultimately arrived.
He was the courier for an intelligence operation directed against a minister.
The explanation on the first cloth had said the messages were for the people protecting him. He had held this belief lightly. He held it more lightly now — not because it was false but because it was incomplete, which was a different kind of wrong from false and required a different kind of holding.
He was being used for something larger than protection.
He did not know yet how much larger.
The picture was not complete.
He would continue the deliveries.