This meant: whatever was happening in the calls was either not audible to the recording system, or the recording system was being modified, or — and this was the explanation that sat most uneasily but fit the evidence most precisely — the calls were real in some sense that the recording system could not capture. They were happening at a frequency or on a channel that the equipment was not designed to detect. They were using the infrastructure of the call system but not entirely within it.
She had found one more thing. On night seven, reviewing the call queue statistics for the previous month — she had access to these as part of her documentation duties — she had found a pattern. After midnight, the call volume on certain account numbers was statistically inconsistent with normal customer behavior. Specific accounts called at very regular intervals — not the randomness of genuine customer calls, but the periodicity of a system: every third night, between midnight and 3 AM, the same accounts produced calls. The accounts were real — she had cross-referenced them, they were active customer accounts with normal histories. But the pattern of their calls was not human.
She brought this to Rajan on night eight.
He read through her notes on her phone. He did not express surprise — he expressed something that was not surprise but was attention, the quality of attention of someone who has been adding pieces to a puzzle for years and has just been handed a piece that connects two sections he had not previously linked.
"The account patterns," he said. "I had suspected something like this but I didn't have the data access you have. How did you get the full queue statistics?"
"Divya gave me the QA system password. The QA system has full queue access."
"Divya." He was quiet for a moment. "She gave you the password without knowing what you would do with it."
