We study.
Evidence, authority, recovery, review, workflow reconstruction, and public-safe proof packets from real runs.
SF-WEB-13 / Research + Joint Research / Source-backed work
Joint research for evidence-bound AI work.
Standing Framework studies evidence, authority, recovery, and review in consequential AI-assisted work.
We partner with teams that have a consequential workflow, reviewable evidence, and a decision that needs to be trusted. When the evidence supports release, we turn that material into bounded findings, reusable review objects, and public-safe research outputs.
Joint research / collaboration terms
Joint research with Standing Framework.
We collaborate on real AI-assisted work: agent runs, review loops, approval paths, recovery, source-state drift, and benchmark evidence.
You bring.
A bounded workflow or workflow set, accountable reviewer, reviewable artifacts or traces, and clear sharing boundaries.
We produce.
A scoped collaboration can produce a research note, mechanism study, failure reconstruction, evidence packet, design-partner finding, or public-safe method artifact.
Not a fit.
No sponsorship wrapper, customer quote extraction, benchmark leaderboard shortcut, or market validation from local proof.
Practical first move
Start with a bounded case rather than a broad theme.
Bring the case, decision owner, reviewable evidence, hard-to-trust decision, and sharing boundary.
Research map / source to question
Current research questions.
Each topic connects a source area, research question, and planned treatment. Review the paper record for the current research position.
What may this score actually decide?
Treatment: authority-boundary evidence cards and invalid-promotion failure states.
Review paper 02What makes yesterday's approval unsafe today?
Treatment: lease timeline with scope, expiry, revocation, and refresh controls.
Review paper 08Where did the work really happen?
Treatment: human-agent-tool choreography mapped to source-state changes.
Review paper 17Is blocked the same as failed?
Treatment: failure classifier and safer-next-move drill.
Review paper 19What should travel with an agent-made change?
Treatment: annotated proof receipt with missing-evidence states.
Review paper 15How do context, evidence, authority, and verification connect?
Treatment: bounded claim-and-evidence graph.
Review paper 32Why is retrospective distance growing?
Treatment: dual timeline comparing execution density with later reconstruction.
Review paper 31When can more context make decisions worse?
Treatment: context-on versus context-off counterexample gallery.
Review paper 07What did the benchmark actually test?
Treatment: matched-condition notebook with denominator and failure views.
Review paper 06Why are local proof and public release different?
Treatment: observed-to-accepted-to-released state ladder.
Review paper 13