Dayo replies at 10:40 AM.
Text, not voice. Which means she measured her words.
I thought about what to do with what you told me for fourteen hours. I decided to use the argument, not the data. If I had used the data — the overhear column, the specific cases — I would have had to explain where it came from. That would have put you at risk and made the Bureau aware of the index before you decided whether to tell them. I did not think that was mine to do.
Sonmat said no. The observers lost access. The consent form held, even though we both know it does not protect anything. Sometimes the fiction is load-bearing.
What happens next is yours.
I read it twice. Then I open the internal memo.
The header: Authentication Event Biometric Capture — Known Unknowns. Prepared by me. Date: February 2043. Marked internal only. Section 1: what the hardware collects. Section 2: what exists in the event logs. Section 3: who knows. Section 4: blank.
I wrote Sections 1 through 3 at 1:20 AM, when the question felt urgent. By 3:00 AM the urgency had hollowed out and I was left with the blank, which is the honest part of the document. The part that admits I do not know what comes after documentation.
The blank has been blank for nine hours.
In nine hours the question shifted.
Not: what do I put in Section 4?
But: is Section 4 mine to fill?
The memo was written for the Bureau — to document what the Bureau does not know. But the Bureau has not asked. There is no request number. No case file. No inbox waiting for it. A document that documents what no one is looking for is not a report. It is something else. I do not have a category for it in the authentication system. I should add it to the Overhear Index: things the institution captured without asking for them. This memo is one. My fourteen-hour silence between sending the message to Dayo and receiving her reply is another.
I type Section 4.
The question of whether this information constitutes an institutional liability requiring disclosure, or a private analytical finding that does not, is not one I can answer unilaterally. I am recommending a review by the Ethics and Compliance Division. I am not filing this memo with them. I am noting that the recommendation exists.
I save the document.
I close it.
The memo is complete. It is the most accurate document I have ever written, and it has no audience. It contains a recommendation that I am not making. It references a review I am not requesting. It says: someone should know this — without naming the someone, without naming this.
What it actually says, underneath all of that, is: I know this, and I do not know what knowing obliges me to do.
Dayo knew that too. That is why she used the argument and not the data. The argument protects; the data implicates. She gave me the choice of which one to be.
The Overhear Index has forty-seven entries. The most recent is entry forty-eight: myself. Date of capture: this morning. What the system heard: an authentication worker sitting with a completed memo and an empty ethics review queue, deciding whether the recommendation she wrote is the kind of recommendation that needs to be acted on.
Nothing in the index scores 1.0 on the intentionality gradient. Not even the most deliberate submission — the finished work handed over with full intent — escapes the ambient data of the handing. The gesture is always overheard.
I submitted this memo to no one.
The system has already logged the submission.