What Astrolabe does not do

Stated limits

1. It does not rank pedagogies. Layer 3 eliminates on constraints and orders survivors by how near they are to what you already do.
2. It does not draft your work. No field where you design offers a generated draft, and there is no setting that adds one.
3. It does not issue a verdict where the evidence is contested. You get both cases and the question that would settle it.
4. It does not certify teaching competence. It supplies evidence; an institution that wants a credential warrants that itself.
5. It holds no hidden profile. Every inference we hold is readable, correctable and exportable by you.

The evidence base

Every pedagogical claim Astrolabe makes is registered with its strength, its sources, and — where the evidence is contested — the dispute. Contested claims show both cases wherever they are acted on. Rejected claims are kept rather than deleted, so they can be recognised and set aside.

The taxonomy currently holds 6 genuine rivalries — pairs that cannot both be true of one design decision. That number is computed from the taxonomy each time this page renders, so it cannot drift out of date.

The instrument reports its own reliability

Descriptor confidence numbers are meaningless without agreement evidence behind the coding scheme. Ours is PROVISIONAL — synthetic figures pending the panel study. Public launch is gated on the real study (§1b.2), not on this file.

Threshold: 0.667 — Krippendorff's conventional lower bound for tentative conclusions. Configurable per deployment.. Descriptors below it may be used and are never reported as a standalone finding; that is enforced in the profile service rather than in this paragraph.

Two kinds of operation

Everything Astrolabe does is one of two things, and every screen says which:

Extendable A model may do this. Reading a document, posing a contrast case, generating a question, critiquing something you wrote.

Anchored A person must do this. Your reasoning, your design attempts, your judgements of quality, and your decisions about what your students must do themselves.

Anchored operations do not reach a model at all. It is not that the system declines — there is no path from those surfaces to a provider, and a test asserts that passing one to the gateway fails.