01
Before truth comes vantage
The grounding question — where did this belief come from, how confident should I be — assumes the reasoning episode is already framed. It rarely is. Who is reasoning, from which role, toward which possibility space, at what level of zoom: these choices precede evidence evaluation and quietly decide what can be observed at all.
This is the deep half of the orientation step in any observe–orient loop. Orientation is not just assembling context; it is choosing the lens the context will be read through.
02
The postures agents fail to hold
Watch agent systems fail long enough and the failures sort into postures, not tasks. Stuck at one altitude: everything answered at 10,000 feet in vague strategy, or at 100 feet in line-by-line detail, with no movement between. Confused perspective: the frame the agent reasons from does not match the frame its tools act in. Wrong creativity mode, in Boden's taxonomy: grinding exploratory search inside a schema when the situation needs a combinatorial bridge between two schemas — or attempting a transformational schema change when plain search would do. And proposing unreachable moves: recommending actions from the abstract space of all actions rather than Kauffman's adjacent possible, the frontier actually reachable from the current configuration.
- Altitude: can the reasoner zoom, and does it know its current height?
- Perspective: does the reasoning frame match the acting frame?
- Mode: exploring a schema, bridging schemas, or changing the schema?
- Reach: is the proposal on the reachable frontier, or in the abstract action space?
03
Declare the posture
The intervention is unglamorous: make the posture explicit before reasoning begins. State the vantage, the mode, the altitude, and the radius of moves under consideration — and let the surrounding system serve views matched to the declared altitude instead of one flat context dump.
Declared posture does for reasoning what a plan contract does for action: it turns an implicit stance into an inspectable object. It is cheap to write down and expensive to skip, because an undeclared posture still exists — it is just unexaminable.
04
Where it sits in the stack
Orientation above grounding, grounding above memory: the lens chooses what counts as relevant, the gradient grades what is believed, the substrate remembers what happened. All three are infrastructure, not prompt garnish.
The honest open question: whether declared posture measurably improves outcomes, or whether it becomes post-hoc rationalization — a stance written down and then ignored. That is testable, and it is on the list.