Most knowledge systems store ideas.
They don't know what they know.

Without explicit epistemic architecture, knowledge systems drift toward undifferentiation automatically

Any system that produces claims over time has the same structural defect by default: the boundary between confirmed and provisional erases — not because anyone is careless, but because there is no mechanism maintaining it.

Old hypotheses are never marked obsolete. Working assumptions become de facto canon — not through verification, but through repetition. Claims recorded without a falsification condition receive the same weight as claims that survived adversarial probe.

Three consequences that surface without explicit status tracking:

• Canonization by repetition — a claim becomes fact through citation, not verification
• Loss of falsification lineage — the system cannot explain why it believes what it believes
• Invisible decay — the assumptions underlying conclusions age out undetected

The problem accelerates in AI-assisted systems, where knowledge production outpaces epistemic verification.

The system does not filter material.
It labels it.

This distinction matters. Filtering removes material from the system — losing with it the information about what was considered and why it was set aside. Ranking creates hierarchy without explicit epistemic status. Curation depends on the curator.

The Epistemic Stack does something different: every record carries an explicit epistemic status at all times. Accumulation is not constrained — the status of what is accumulated is always legible.

It is a five-layer architecture. Each layer has explicit status values, transition conditions, and decay rules. A record cannot move to the next layer without satisfying the gate condition. A record that does not satisfy the gate stays where it is — correctly labeled.

How revision remains possible without collapsing trust

Fix everything as certain — the system stops learning. Revise everything — nothing can be trusted. The stack offers a third option: controlled revision. Revision happens only when a formal condition is met.

How the system generates pressure on its own conclusions

Publishing a case is not the end of the cycle. It is entry into an environment of epistemic pressure.

Each published case carries a structured signal field that tracks its relationship to the mechanism taxonomy: whether it supports current classification, introduces a potential counterexample, or marks a boundary condition. Counterexamples and unresolved edge conditions remain attached to the taxonomy and can trigger structured re-evaluation when pressure accumulates.

Multiple independent validation perspectives operate in parallel, each exposing different classes of failure and interpretive bias.

CDSA applied to its own knowledge architecture

The problem the stack addresses is not specific to StratoAtlas. Any research practice that produces claims over time faces the same challenge.

This is an architectural problem — and it has a CDSA diagnosis. The contradiction: accumulation (more material → richer system) conflicts with legibility (more material → harder to know what is actually confirmed). Standard responses choose one over the other. The Epistemic Stack resolves the contradiction structurally: accumulation is not constrained, but every record carries its status in explicit form.

The stack is applicable to any system that:

• produces claims over time
• operates with multiple participants or instances
• must answer the question "why do you believe this" — structurally, not from memory

StratoAtlas uses the stack as infrastructure for its own diagnostic practice. The same architecture applies wherever knowledge production runs faster than epistemic verification — which is most places now.