Use this checklist when evaluating whether a runtime can implement VALP.
The question is not “can this runtime run an agent?” The question is whether it can export enough evidence for another person, agent, or CI job to audit the work.
Choose the closest shape:
| Adapter class | Typical runtime | Mode |
|---|---|---|
| pane controller | terminal panes or browser-controlled panes | Full Mode if proof is exported |
| daemon queue | local job queue or runner daemon | Full Mode if state and evidence are exported |
| hosted/local platform | web board plus agent workers | Full Mode if audit data is accessible |
| remote SSH | runtime state lives on another host | Remote Mode |
| manual | human copies prompts and results | Manual Mode |
A Full Mode adapter must export:
If any of these are missing, the adapter may still be useful, but it should not claim Full Mode yet.
Check each dispatch step:
dispatch_written: the task-local dispatch file or equivalent exists.dispatch_inserted: text was inserted into an input surface, if applicable.dispatch_submitted: the runtime proves the message/job was submitted.dispatch_completed: expected evidence exists after execution.dispatch_blocked: delivery or completion could not be proven.Text inserted into a pane is not enough. Queue accepted is not enough. Hosted job completed is not enough. The adapter must connect runtime state to expected evidence.
For each Leader-declared Agent, record:
Pane adapters should record:
Queue, hosted, or remote adapters should record equivalent job/session facts:
Do not fake pane fields for non-pane runtimes.
The adapter must stop or require approval for:
Approval evidence must be task-local and visible. A permissive runtime default does not override VALP approval gates.
If Leader-declared workers return recommendations:
Before claiming compatibility, run a small task through this sequence:
publish task
refresh capability passports
record the user-selected Leader
Leader writes assignment declaration
validate declared assignments
write concise dispatch
submit through runtime
record dispatch_submitted proof
wait for worker output
write expected evidence
record dispatch_completed
write final synthesis
run valp audit
The task is compatible only when the audit passes for the right reason.
In a runtime-adapter discussion, include: