Running a title, and reading the report
The point of a run is not that it worked. It is what the guest asked for and what it got, which is the only thing that can be checked.
Why the report exists
An emulator of an undocumented platform can tell you that a guest died and almost never whether an answer was right. A function returns a number, the guest carries on or it does not, and forty thousand frames later something is wrong. There is no specification to test against, because the specification is the thing being reconstructed.
So a run records the calls, and comparing two runs is how a change is judged.
orbistoun run <title>
orbistoun report # what the last run asked for
orbistoun verify # this run against the lastHonest failure, and why a stub is not free
A stub that returns success is indistinguishable from working code until forty thousand frames later. Orbistoun's rule is that a function it has not implemented says so, loudly, at the call
- rather than returning zero and letting the guest carry on into damage it caused somewhere else.
That makes some titles stop earlier than they otherwise would. That is the intended trade: an early stop names its cause, and a late one costs a day of bisection.
When you see a run get less far after a change, that is not automatically a regression. It is what success looks like when the change replaced a lie with a refusal.
Levels, when you want more
An ordinary run is quiet. Everything below is off unless asked for:
OOPS_LOG=debug orbistoun run <title> # decisions and resolved configuration
OOPS_LOG=trace orbistoun run <title> # per-call detail
OOPS_LOG=warn,orbistoun_loader=debug orbistoun run <title>OOPS_LOG and RUST_LOG both work; the first
is the one the whole collection answers to.
Levels mean the same thing in every tool here: error is
giving up, warn is a surprise that did not stop the work,
info is an action with a side effect, debug is
decisions, trace is per-item.
The verdict, and what it is not
verify compares this run with the previous one and says
what moved. It is a differential tool: it knows that something
changed, not that the new answer is correct.
Nothing here can tell you an answer was right. That question needs an oracle, and the oracle is either a probe you wrote yourself (obSCEne) or real hardware reached through Prosperous. A commercial title can only ever tell you it stopped.
Where the artefacts go
Reports, traces and screenshots land under the data root - see paths, or run orbistoun paths to be
told exactly where, on this machine, in this mode.