Research

ABSD is an operating system first. It is also used to study a few questions that are easier to answer with a kernel, a filesystem, and real programs than in the abstract. None of this is a claim about other systems. The research program page gives the longer account: the question, the prior work it builds on, the method, and what is being investigated now.

Authority at the time of an effect

A check made before an action can go stale by the time the action happens. ABSD's primitive for this is the versioned cell: cell_transition succeeds only from the version the caller names, so an update based on stale state is refused rather than lost MILESTONE-QUALIFIED. Rights are checked at each operation, not when a handle was opened: making a file read-only withdraws writing from handles opened before MILESTONE-QUALIFIED. How far that can be carried, into effects that depend on evidence which expires, is an open question PLANNED.

Crash and durability semantics

Each storage claim names its failure model. The filesystem and an SQLite store were qualified under an abrupt-stop model and under a modeled power-loss law, with every cut point of each operation enumerated on the host (the device's possible outcomes sampled where there were more than 4096) and sampled cuts in the guest MILESTONE-QUALIFIED. The documents record where the model ends: torn sectors are characterized, not claimed; real power and real drives are untested.

Process authority

Authority as the set of handles and rights a process holds, narrowing at every start, with no ambient privilege and no user IDs MILESTONE-QUALIFIED. The open problem is several principals: independently authenticated operators holding different authority, without importing accounts and permission bits PLANNED.

Compatibility layers

How much ordinary software can run above native semantics that differ from Unix, and where it stops. The source-compatibility probe measures this for unmodified Rust programs MILESTONE-QUALIFIED; a Linux binary personality is a deferred experiment with a stop rule if it starts pulling the native design toward Linux PLANNED.

Qualification method

Claims are stated at the strength of the evidence: a named environment, a recorded run, counts from the check script itself, and deliberate absences checked to be absent. Hardware support is tiered and never inferred from emulation (tiers). The evidence page is an example of the record kept for each milestone.

A real native workload

ABSD runs programs from Constellation, a separate project by the same author, as ordinary native workloads: an SQLite-backed authority store, a Pulse journal and support resolver, the NQ diagnostics tool, and the Nightshift writer and resolver, each compared with its Linux build MILESTONE-QUALIFIED. These programs live above the operating system; nothing in the kernel depends on them, and ABSD's semantics are not defined in their terms. They are useful because they were not written for ABSD. Public material on Constellation: unpingable.com/constellation.

By the same author. These are not evidence about ABSD.