The take is finished and the numbers are in: how much of it you spent looking at the lens, the longest single look away, how steady the framing held. The temptation is to read them the way you read a mark at school — a seventy is bad, a ninety is good, try harder next time. That reading wastes them. Trying harder is not a plan, and eye contact is one of the few delivery problems where effort is mostly beside the point. The numbers are not a grade on your sincerity; they are measurements of a physical arrangement — where the text was, where the lens was, how your body was set between them — and physical arrangements are fixable in ways that character is not. The useful question in front of each number is not how do I improve but what is this number pointing at. Each of the three points somewhere different, and the fixes barely overlap. Read separately, they turn a vague sense that something looks off into two or three specific adjustments you can make before the next take.
Start with the share of the take spent looking at the lens. When it comes back low, the instinct is self-blame — not disciplined enough, not practised enough — but a low share is almost never a discipline problem. It is a geometry problem. The number is telling you where your text was relative to the lens while you read it. Notes on the table drag your gaze down; a script in another window drags it sideways; each glance is small, but a take is made of hundreds of them, and the share records their sum. The fix is to move the text, not to try harder: bring the script to the camera. With the prompter laid directly over the camera view and the text narrowed so your eyes stay near the lens axis, reading and looking into the lens stop being two activities you switch between. People who jump from a fifty to the high eighties between two takes did not become sincerer overnight. They moved the words.
The longest look away is a different instrument. The share tells you about the take on average; the longest look away tells you about its worst moment — and viewers remember worst moments, not averages. One three-second dive to check a fact does more damage than a dozen flickers, because a flicker reads as thinking and a dive reads as a person consulting something you cannot see. The productive move is to ask what was happening in the script at that moment. Almost always the answer is a passage you did not trust: a number you had not memorised, a transition that was never written, a name you feared getting wrong. The longest look away is a bookmark into your weakest paragraph. Rewrite that passage, or rehearse just it, and the number shortens on the next take without your gaze habits changing at all. No other tool tells you as precisely which part of the script is not really written yet.
Framing steadiness sounds like a tripod statistic, but with the camera locked in a mount, the thing moving in the frame is you. A low steadiness number usually records one of two habits. The first is leaning in — creeping towards the screen as you read, which is a font-size problem: make the text bigger and the lean disappears. The second is the slow slump of a body that has stopped performing, which tends to arrive in the last third of a long script. The other physical lever is lens height. A camera below eye level makes you bow towards the text and look down on the viewer; a lens at eye height, with the device raised on a stand rather than propped at an angle on a stack of books, keeps the head level and takes the drift with it. Distance matters too: a prompter close enough to read without leaning is worth more than a bigger screen placed further away.
Two honesties keep the numbers useful. First, Talk2Camera measures what happened on the take, on the device — and when no face is found in the frame it reports zero rather than estimating. A zero after a take recorded in bad light, or with the camera pointed past your shoulder, means the camera never found you, not that you never looked; fix the light or the framing before reading anything into it. Second, the numbers describe the mechanics of attention, not the worth of the delivery. A take can hold the lens ninety-five per cent of the way and still be dull, and a warm, useful video can survive a wandering eye. The metrics earn their place as a loop, not a verdict: read one number, change one physical thing — text position, font size, lens height — and record again. When the number moves, you know which adjustment did it. That is more than trying harder ever tells you.