Comparisons
Where does ClariLayer actually fit?
A project instruction file and a semantic layer are both useful. ClariLayer handles a different job: selected work that remains attributable, correctable and available across sessions. Analytics reconciliation adds a specialist trust path beside the data stack.
When someone wants continuity across AI sessions, two familiar tools come up. The first is writing the definitions down by hand in a CLAUDE.md or notes file. The second is reaching for a semantic layer or catalog — dbt, Cube, a metrics store, a data dictionary. Both are reasonable. Neither is the same thing ClariLayer does, and saying so honestly is more useful than pretending we replace them.
The thread through both comparisons is the same distinction. ClariLayer does not compute your numbers and it is not a warehouse of definitions you have to trust on faith. It is the trust layer: it reconciles a saved definition against the result your agent actually computed, shows provenance and status, and surfaces a caveat when the declared definition drifts from the real result — instead of handing your agent a confident, unchecked answer. We describe what is reconciled and what is merely asserted, and we never stamp a claim present-tense “verified.”
vs the free default
ClariLayer vs a hand-written CLAUDE.md
A project file is a useful place for stable local instructions. ClariLayer adds selected-space recall, source and applicability, correction and scoped forget across projects and sessions. Analytics reconciliation is a separate specialist advantage.
Asserted vs reconciledvs the execution layer
ClariLayer vs semantic layers & catalogs
A semantic layer (dbt, Cube, and the like) or a data catalog answers how a number is computed and where it lives. ClariLayer answers what your agent should trust — reconciled against your real result, not asserted. They sit at different layers, and they compose.
Execution vs trust layerThe one distinction that runs through both
A definition you typed and a definition that was reconciled against your warehouse are not the same thing — even when the words are identical. The first is asserted: nothing checked it, so when it drifts your agent keeps trusting the stale line, confidently. The second carries its provenance and status, and a declared-vs-actual mismatch shows up as a caveat you can act on.
That is the trust gap ClariLayer exists to close. A hand-written CLAUDE.md never closes it. A semantic layer makes the computation right but does not, on its own, tell your agent which saved definition has quietly diverged from what the warehouse now returns. Both comparisons are really the same story, told against two different starting points.
Evaluate it on your own data.
Keep your CLAUDE.md and semantic layer. Add selected memory across sessions, then use Analytics reconciliation where a live result needs an explicit caveat-aware check.
Connect your AI