I slept dreamlessly that night. The hotel room was quiet and comfortable and had all the comforts of a three star property. I lay on the large double bed and stared at the darkness managing multiple arguments in my head and then, somewhere in that negotiation, I was gone.
I woke at four-thirty by force of habit. No alarm required.
For a few seconds before full consciousness set in there was nothing. Then it came back in order, as it always does. The folder. The photographs. The economy class observation delivered with the pleasantness of someone stating an irrefutable fact. And finally, the part that sat lowest and heaviest: Anjali had looked at everything and agreed.
Meera Ahuja. I had not asked her name on the plane and she had not offered it, but I had known it beforehand. Anjali's mother. The woman who ran a business empire spanning multiple industries with, by all accounts, a mind like a steel trap. I had no reason to doubt either assessment.
I got up, dressed for a run and went out into the Bangalore morning.
The city at five in the morning has a specific quality that I had heard about but not experienced. The air is cooler here than I was used to, carrying the faint eucalyptus sharpness of a place that still has trees growing close to its arteries. The streets were not empty. Bangalore is never truly empty but they were quiet enough for the city's scale. I ran for an hour without thinking about anything, letting the rhythm of the pavement do the work that sleep had not fully finished.
By the time I came back, the thinking part was done.
I would not call Anjali. I would not message her. I would not find a way around the edges of the instruction, of which there were several that I could already see. The reason was not Meera Ahuja's threat to my career, considerable as that threat was. It was the second threat—the one delivered quietly, almost gently, in a voice that had briefly lost its boardroom composure. I understand there are people who helped a frightened sixteen-year-old boy become someone else. People who have built their own lives in the years since. She had not named them. She had not needed to. I knew exactly who they were and exactly what exposure would cost them. They had taken a risk for a boy they did not have to take a risk for. They had built lives on the assumption of certainty of some things remaining buried.
I was not going to disturb those lives for the sake of my own wanting.
I showered, dressed in formals and went to the hotel restaurant for breakfast. The run had made me ravenous and I needed the nutrition.
Then I took a cab to my new office.
The Bangalore branch of the company occupied six floors of a glass tower in the northern tech corridor, announcing its own importance through the quality of its lobby marble and the height of its atrium ceiling. The security at the ground floor was tighter than at the previous office—card access at the lobby, a second scan at the elevator banks and a biometric reader at the entrance to the sixth floor where the Business Intelligence division was housed. My new ID card, issued in Delhi, cleared all three without incident.
I had expected that much.
What I had not fully prepared for was the silence.
Not the silence of an empty office. The floor was occupied, populated with people at desks and moving between meeting rooms and standing at coffee stations with the purposeful looseness of people being busy. It was a different kind of silence. The silence that falls specifically when you walk into a space where everyone already knows who you are and has already decided what they think about it.
I was aware of it from the moment the biometric reader beeped me through the glass door. Heads turned but quickly looked away without eye contact. That was the tell. In a normal office, a new face prompts at least the involuntary flicker of curiosity—a glance, a held gaze, a whispered question to a neighbour. Here, everybody looked away. Not because they hadn't noticed me, but because they had all independently decided, in advance, that looking at me was a thing they were not going to do.
A young woman from HR named Pooja met me at the door with a lanyard and a visitor's temporary sticker to supplement my permanent ID—a formality, she explained, until the floor access was fully updated in the system. She was professionally cheerful in the way of someone executing a procedure rather than performing a welcome. She showed me the layout: the main analyst floor, the server access room which required a separate clearance I did not yet have, the meeting rooms arranged along the eastern glass wall, the pantry, and finally my cubicle.
My cubicle was at the far northwest corner of the floor.
It was, by any objective assessment, the worst position in the office. It sat behind a load-bearing pillar that partially blocked the view of both the main floor and the windows. The desk was older than the others; a slightly different shade of laminate that spoke of a lower quality. The chair had a wobble in the left rear castor. The monitor was a generation behind what I could see at the other workstations. The nearest kitchen was a considerable walk. The nearest natural light was effectively none.
Pooja showed me this location with the same cheerful efficiency with which she had shown me everything else and asked if I had any questions.
I said I did not, thanked her and sat down.
She left.
I looked at the partition in front of me, which was the back of someone else's cubicle and had a motivational poster on it that said Data is the new oil in a futuristic font. Then I opened my laptop, logged in and began setting up my workspace.
Nobody came over to introduce themselves. Nobody came over for any reason at all.
By the end of that first day I had a reasonably complete picture of the floor's social geography, assembled from observation rather than interaction since interaction was not on offer.
The BI division's Bangalore base was, by the standards of any company, an extraordinary concentration of talent. This was not modesty-adjusted hindsight—it was simply visible in the way people moved and spoke and approached problems, the kind of unconscious fluency that comes from years of operating at a level where imprecision is not tolerated. The team was almost entirely composed of people who had come through the Harvards, Stanfords, Indian Institutes of Technology or equivalent institutions, or who held postgraduate degrees in data science, mathematics or computer engineering from universities that received applications in the thousands for seats counted in the dozens. Many had PhDs. A lot were young. All of them had been selected through a recruitment process that, I would learn over the following days, was substantially more rigorous than anything the rest of the company employed.
They also knew it. There was a particular quality to the self-assurance here that was different from the confident salesmanship I had been used to. Sales confidence is performative by design; it is pointed outward at clients and prospects and the next deal. This was quieter and more total. The confidence of people who did not need to perform competence because competence was the baseline condition of being in the room.
I understood, sitting in my pillar-adjacent cubicle on the first day, that walking into this department as its newest and most junior member - a sales manager turned data analyst, recruited under circumstances that everyone on the floor already knew about in detail - was going to require a specific kind of endurance. Not the adrenaline-charged endurance of a bench press at its limit, where the effort is visible and the crowd is watching and the end is a rack and a moment of silence. The slower kind. The kind that looks, from the outside, like nothing at all.
Simran I located within the first hour. Her desk was positioned in the center of the analyst floor, not a corner office, the division was apparently not structured that way, but a desk whose position communicated a kind of informal authority. She was the Senior Analyst who had been assigned to oversee my onboarding. She was also, it was becoming clear, the reason the floor had achieved its current unified position on the subject of my arrival.
She did not look at me at all that day. It was a more pointed and deliberate version of the general avoidance—active rather than collective. I had seen it from her in Delhi, the tight smile and the assigned impossibilities, but there it had been one person's professional hostility on a floor where I had years of accumulated goodwill as insulation. Here, that insulation did not exist.
I already knew what had happened from my conversation with Simran back in Delhi. The BI team's mandate included, among its many functions, exactly the kind of internal audit that should have caught what I had done to the sales data. They had been tasked with investigating the anomaly—the numbers that didn't quite line up across quarters, the whisper in the data that something was being suppressed. Simran had led that investigation. It had gone on for months. And it had found nothing.
Not because the team was not good enough. They were very good. It was because I had been, in that particular endeavor, better - or at minimum, more specifically motivated. I had built the concealment from the inside, with full knowledge of exactly how an audit team would approach the data and exactly where they would look first and second and third. I had built the walls to match the shape of the scrutiny.
The team had not found it. And then Anjali had handed them everything, which meant the team had received, from an external source, the solution to the problem they had spent months on and had not solved themselves. The appraisal cycle had not been kind. Abhijit had not been kind, in his particular eccentric way, which I gathered from the residual tension around certain people's desks was more cutting than his mild manner suggested.
And now the man responsible for all of it was sitting in their northwest corner with his wobbling chair and his legacy monitor, drawing a salary from their division's budget.
I could not entirely blame them.
The first week established the pattern and I adapted to it.
No work was assigned. This was not an oversight but a statement. My name did not appear on any project roster, any task list, any briefing document that I could find. Simran, who was nominally responsible for my onboarding, had communicated nothing further since the Delhi meeting beyond the reading list and the single data task that had ended with Abhijit's intervention. No follow-up reading had been sent. No meetings had been scheduled. No introduction to team members had been arranged. The onboarding, such as it was, consisted of the HR walkthrough and a silence that extended from Monday morning through the following Friday afternoon without variation.
I kept my expression pleasant and did not react.
This was, I had learned over the years, the most effective available response to environments designed to produce a reaction. Anger, resentment, wounded withdrawal - all of these confirmed for the people who wanted to see them that their assessment of you was correct. Cheerful indifference, maintained consistently, had an eroding quality. It was harder to sustain hostility toward someone who appeared entirely unaffected by it. Not impossible - some people can sustain anything - but harder.
And the pleasantness did not have to be performed with any great expenditure of energy, because I had, genuinely, things to occupy myself with.
The access my ID card provided was, as Abhijit had mentioned in our brief and eccentric car conversation, considerably broader than what my title and seniority would normally confer. I did not know whether this was deliberate on his part or an administrative oversight, but I was not going to ask. Instead, I spent the empty hours of that first week doing what I would have done with any new and complex system: I mapped it.
The company's data infrastructure, seen from the inside of the BI division's access level, was significantly more intricate than I had understood from my years in sales. What I had known from that vantage point was the customer-facing architecture - the CRM systems, the sales tracking databases, the revenue reporting pipelines that fed the quarterly numbers that had once been my entire professional world. I had known them well enough to manipulate them, which I had taken, with the particular vanity of the technically skilled, as evidence of a thorough understanding.
It was not. It was evidence of a surface familiarity.
The actual system was layered in a way that a deep-sea map is layered - the surface visible and navigable, and then, below it, further structures organized by pressure and darkness and function rather than by the needs of the light above. The BI division sat at an interface between those layers, with access credentials that moved vertically through the architecture in ways that the sales division's permissions had never permitted.
I moved carefully, building my map in sections, never going further into a system than felt structurally natural for the purpose of understanding the framework. I made no changes. I extracted no data. I simply read the architecture the way you read a city by walking its streets, attending to what was named and how, where the connections ran, what fed what, where access narrowed and what that narrowing indicated about the sensitivity of what lay beyond.
By the third day I had a clearer picture of the company than I had assembled in three years of working for it. The software development services organization that my colleagues in sales had pitched to corporate clients; the B2B machinery of contracts and proposals and quarterly targets was the visible portion. A large, legitimate, highly profitable visible portion, but a portion nonetheless.
Beneath it, or perhaps more accurately alongside it in a parallel structure that shared certain infrastructure but operated under entirely different governance rules, was something considerably more substantial. The company built complex, bespoke, classified systems for organizations that did not appear in the public-facing client lists. Defense ministries. Military procurement bodies. Intelligence-adjacent agencies in multiple countries, most of whose names I did not recognize from the metadata alone and some of whose names I recognized immediately. Shipbuilders. Aircraft manufacturers whose primary customers were governments rather than airlines. Contractors whose sector, even without the specific project details that I did not have access to, was made apparent by the nature of what they needed built.
These were not the accounts that the sales managers in my former team had been chasing. These existed in a part of the organization that was structured specifically to ensure that people like my former team would not know they existed.
The BI division managed the data for all of it—the civilian business and the other business—with the same infrastructure and many of the same people. Which explained, among other things, the recruitment standards, the access controls, the non-disclosure agreements with their unusual specificity, and the instruction from Nidhi in HR that I was not allowed to have professional relationships within the company without written disclosure. You did not staff a data function with that kind of reach and then tolerate loose personal networks that hadn't been vetted.
There was a third layer, or perhaps a better word was a third service line, visible from certain angles of the access I had been given. The BI division appeared to offer services of its own—not as a support function for the company's primary business, but as a commercial offering in its own right. These services were marketed and delivered to a specific client profile: government agencies, defence manufacturers, a few names in the maritime and aerospace sectors that I recognized as significant. The nature of the services was not visible to me at my current access level. I could see that they existed, could see the client category and the volume indicators and the billing architecture, but the project details themselves sat behind credential thresholds I had not been cleared for.
What I could infer, from the client profile and the structure and the classification markings on the service architecture, was that these were not conventional business intelligence offerings. Standard BI—dashboards, analytics, forecasting, process optimization—was what the company provided internally and what it provided to its commercial clients. What was being offered to governments and defense contractors and aircraft manufacturers under this third service line was something that occupied, as Abhijit had put it in his car with those expressive raised eyebrows and the unsettling driving, the grey areas.
I filed the inference and did not pursue it further. I had enough to absorb.
The reading material that Simran had sent me in Delhi - the list that had run to thousands of pages and had been delivered with the clear expectation that it would sit unread and provide grounds for a negative onboarding assessment. I worked through them in the evenings and late nights at the hotel and in the silent hours of each office day.
It was, stripped of the punitive intent with which it had been compiled, genuinely excellent material. Database architecture, data governance frameworks, statistical modeling methodologies, cryptographic protocols for data security, distributed systems design. The list was essentially a crash course for someone being brought up to operating standard in a serious data organization. I had strong foundations in several of these areas and had to build from scratch in others. The areas that required building from scratch were the most interesting. I kept notes. I cross-referenced against what I was observing in the live systems. By the end of the week I had covered everything on the list and had extended into several adjacent areas that the list had pointed toward without including.
On Friday afternoon I found three things that the list had not mentioned and the live systems had shown me.
The first was a redundancy in the data pipeline that aggregated usage metrics from a cluster of client-facing applications. Two separate processes were executing substantially similar transformations on the same dataset at different points in the pipeline, which meant that a portion of the computational load on that cluster was being run twice to produce identical intermediate results. It was not a dramatic inefficiency. The processes were fast and the cluster had capacity but it was unnecessary, and the duplication had a downstream effect on the timing of certain reports that several teams appeared to be waiting on each morning.
The second was a mismatch in a query that pulled together data from multiple departments into a single combined report—the equivalent of a roll call that called names from several different lists and counted the hands. The problem was that two of those lists used slightly different identifiers for the same group of people. Imagine a school that has one register listing a student as R. Sharma and another listing the same student as Rahul Sharma. A system that doesn't know these are the same person counts them as two. Except in this case the error ran in reverse—the system was failing to recognize which two entries were the same record and which were not and was therefore counting certain items once when it should have counted them twice. The undercount was small enough to fall within the noise threshold of the report it fed, which was presumably why it had not been flagged. It would not have affected any decisions. But it was wrong.
The third was a latency issue in the authentication handshake between two database systems that had been connected at different times and whose session timeout parameters had never been harmonized. What it means is that it was a timing problem between two database systems that had been built at different times by different teams and had later been connected to each other. Each system had its own rules about how long it would wait before deciding that a connection had gone stale and closing it - the equivalent of two offices that had agreed to keep a phone line open between them but had never coordinated on how long either side would wait before hanging up if nobody spoke. One office was willing to hold for five minutes. The other considered three minutes sufficient. During quiet periods this never mattered. The line stayed active because messages were passing through it constantly. But during peak hours, when both systems were busy and small pauses crept into the traffic, the shorter-patience system would occasionally hang up just before the other side was ready to speak again. The result was occasional failed connections during peak access periods, which would have presented to end users as intermittent login errors. Probably attributed to network variability. Not traced to what it actually was.
I documented the problems and respective solutions with detailed explanatory notes, and sent the documentation to Abhijit and Simran by email before leaving on Friday evening. No commentary. No bragging. Just the work, explained clearly.
Then I locked my computer and went to find somewhere to live, because the hotel was costing the company money it had agreed to spend for exactly seven days and the seventh day was tomorrow.