Chapter 12: The First Test
I read the yellow sentence a third time.
Do not proceed until recovery behavior is independently verified.
The words didn’t tell me what the problem was. They told me what the team didn’t know yet.
I pulled up the message that had arrived with the revised schedule. Priya had offered to walk through the new section during our call that morning. The proposed evaluation window was still listed for the following week, but the word provisional appeared beside it in bold.
I opened the draft terms in another tab. My signature line remained blank.
The system could not be tested just because a date had been penciled onto a calendar. I closed the schedule and made a list of what I needed answered before I would consider signing.
At the top: Who would verify the recovery behavior, and how?
By the time the call began, my coffee had gone cold again. Priya appeared first on my laptop screen, followed by two Meridian engineers I recognized from the review meeting. One had worked on the test environment’s security checks. The other had reviewed the project’s data-handling requirements. Neither had been part of the recovery analysis.
Priya looked tired, but not surprised to see my list.
“I saw the new note,” I said. “Before we discuss dates, I want to understand what changed.”
“That’s fair,” she said. “We found that the environment records the interruption itself, but we don’t have enough evidence to show that its recovery log captures every repeated interruption in the sequence.”
“Is that a bug in the prototype or the test environment?”
“We haven’t established that yet.”
I looked at the yellow sentence again. “Then we shouldn’t call the next window an evaluation window.”
The engineer handling security glanced down at his notes.
“It’s a planning placeholder,” he said. “We’re not treating it as authorization to run anything.”
“Then let’s label it that way in the documents. I don’t want a date turning into an assumption.”
Priya nodded. “We can make that change.”
I scrolled to the terms. “I also want someone who didn’t design this part to review the recovery process. Not just read the code and say it looks reasonable. I want a separate review plan, the exact evidence they’ll examine, and a written record of anything that remains uncertain.”
The other engineer leaned toward her camera.
“I can review the log format and the environment’s recovery records,” she said. “But I didn’t build the synchronization layer. I’d need the implementation notes and the interruption sequence.”
“Good. And if the records don’t prove what happened?”
“Then they don’t prove it,” Priya said. “We don’t infer recovery from a clean-looking display.”
I let out a breath I hadn’t realized I was holding.
The answer did not fix the problem. It made the problem clearer, which was the only useful thing an answer could do at that stage.
The engineer assigned to the security review shared a revised document. The schedule had been moved to a section called Potential Evaluation Window—Not Approved. The first task was now a review of the environment’s logs and the recovery sequence. No synthetic-data evaluation would begin until the independent review was complete, the revised scope was approved, and I had signed the terms.
I read the sentence twice.
“That matches what I understood,” I said. “Please make sure it appears in the version I’m asked to sign, not just the call notes.”
“It will,” he replied.
The changes were small on the screen. They meant the difference between a real boundary and a polite promise that might disappear when people got busy.
Priya brought up the recovery diagram. Repeated interruptions traveled through several steps: a session paused, a connection resumed, a record was checked, and the environment wrote an entry about what had happened. The new concern was whether those entries gave an independent reviewer enough information to tell what had been restored and what had merely been attempted.
I leaned closer to the screen.
“If the system says the session resumed, does that tell us the record recovered?”
“No,” Priya said. “That’s part of what we’re trying to separate.”
“Then the report needs to say that in plain language. ‘Session resumed’ can sound like ‘record restored,’ even if it doesn’t mean that.”
She typed a note.
“That’s useful,” she said. “We should define those terms before anyone reads the results.”
The engineers discussed which logs could be examined, how the review would be documented, and what would happen if the records proved incomplete. I took notes, stopping them when they drifted toward describing the planned evaluation as if it had already been scheduled.
Each time, they corrected the language.
It was not dramatic. No one gave me a speech about how important my concerns were. We made the document more accurate one line at a time.
Near the end of the call, Priya shared one more message.
“A Meridian liaison heard from the clinic that may eventually participate in a later phase,” she said. “They haven’t been given access to the system or its data. They asked what would be required before anyone there could rely on it.”
“What did we tell them?”
“That nothing is approved for clinical use. They’re asking for a written recommendation before they would consider relying on any part of the system.”
I stopped taking notes.
“From whom?”
“They haven’t specified. They want a recommendation from Meridian, and they want the limitations stated clearly.”
“Then they should get one that says we can’t recommend reliance yet.”
“That’s my view, too,” Priya said. “But we should be precise. We can explain what has and hasn’t been evaluated.”
I looked at the unsigned terms beside the revised schedule. The clinic’s request sounded reasonable. It also made the distance between a promising prototype and a dependable tool feel much larger.
“I don’t want the document to imply that the pilot has started,” I said. “It hasn’t. And I don’t want a recommendation for future consideration to get shortened into approval.”
“We can state both,” the security engineer said. “No clinical use, no clinic data, and no recommendation to rely on the prototype while the review is incomplete.”
“Put that in writing before it leaves Meridian.”
They agreed.
When the call ended, I stayed at my desk with the documents open. My job offer was nowhere in the conversation. It didn’t need to be. The signed agreement was its own thing, confirmed in writing, independent of this project and whatever the team decided to do with it.
That separation mattered.
At graduation, the CEO’s announcement had made the room sound as if a public moment had changed my future. It hadn’t created my job. It hadn’t made the work I’d done before anyone knew my name. And it did not make an unfinished system safe.
I reopened the proposed evaluation terms and marked the sections that still needed revision. Then I sent the team a short email: I was willing to keep reviewing the documents, but I was not agreeing to the evaluation until the independent verification plan, scope, and reporting requirements were clear in the written terms.
I added one more line.
Please do not describe the evaluation as scheduled or underway.
The reply came a few minutes later. Priya confirmed they would revise the language and send me the verification plan once the independent review had been scoped. The email also included the clinic’s request, forwarded for context.
It was one paragraph, courteous and direct. Before the clinic would consider relying on any part of the system, it wanted a written recommendation explaining what had been validated, what remained uncertain, and who would be accountable for those conclusions.
I read it once, then returned to the unsigned agreement.
The system had not passed its first test. It had not even reached the evaluation.
May you like
But now the question was no longer only what Meridian hoped the prototype could do.
Someone outside the company wanted us to put in writing what we could honestly stand behind.
