The seam you will never find

Inside the firmware a recording is not one long file. It is a run of five-minute segments. Every five minutes the current file is sealed, a new segment starts immediately, and the two overlap by one to two seconds.

No tone, no buzz, no pause. The software merges them automatically at sync, and you will never discover that any of it happened.

Why segment at all

Because a file that is being written to is fragile. A power cut, a storage fault, an unexpected restart — anything that happens mid-write damages the file that has not been sealed yet.

Without segmenting you could lose forty minutes. Cut into five-minute pieces, you lose five at most, because every earlier segment has already landed intact.

Why five minutes and not ten

Ten-minute segments cost slightly less overhead and produce fewer files. We still chose five.

The reason is simple: this is a decision the user cannot see. When an invisible decision goes wrong, the user has no way to recover from it and no way to see it coming. Where nobody can look, take the conservative number.

Five minutes means the worst-case loss is half what the ten-minute version would be, and the price is a little storage overhead and a little merge logic — both of which we carry, and you do not.

Those one to two seconds of overlap

If two segments met exactly end to end, the word sitting on the cut would be sliced in half, and after merging it would either repeat or vanish.

With one to two seconds of overlap, the audio around the cut exists in full in both segments, and the software has enough slack to align them and put it back together. Those seconds are the only genuinely wasted part of the whole mechanism, and they are the reason it can be imperceptible.

A large part of good engineering is doing the things nobody hears about when they go right.