Main idea

A working frame is usually written at the beginning: the goal, the directives, the way material is loaded. What comes afterwards is supposed to be work, not frame. That chronology manufactures a false boundary — it classifies a rule by the moment it was found, whereas its place depends on what it commits.

A translation rule can't be invented before you have translated. It appears while rereading an English chapter that still smells of French: translate the meaning rather than the syntax, check the degrees of assertion, reread each text as if the French didn't exist. So it is born late, in the middle of the work, and its content is no less governing for that: from the moment it is validated, every subsequent translation will have to apply it.

The frame can therefore learn. It takes in what was discovered and settled along the way, on the same footing as what was laid down at the outset, on the single condition that it is now binding on what comes next.

Why it matters

This lifts a practical blockage: without that permission, a rule validated along the way stays in the session record where it was formulated, that is, in a place nobody rereads before acting. It is known to whoever was there and lost to everyone else, including the assistant that picks up the work the following week.

It also gives the right question to ask at the end of a session: what did we decide today that will bind us tomorrow? That is a filing question, not a memory question.

Nuances and limits

The permission quickly turns into a catch-all: not everything a mission learns has to become a rule, and a frame that takes in every lesson becomes a diary again.

A frame that learns also poses a dating problem. A rule that arrived in week six doesn't apply retroactively to translations already accepted, and nothing in the file says from when it holds.

Open questions

• What do you do with deliverables already produced when a rule of the frame arrives after them: rework them, or accept two quality regimes within the same piece of work?