Agent Harness Wiki

4 min read

Grounding Moldable Operations Studio Ideas in Real Research

Question

If the sci-fi and architecture passes gave us the right metaphors and primitives, what real research fills in the implementation details for how those ideas might actually work?

Short answer

The strongest result of this pass is that most of the desirable studio properties are already well-supported — just not usually assembled into one developer-facing control plane. The literature gives fairly direct recipes for:

  • one state, many synchronized surfaces
  • room-scale overview plus personal staging/inspection
  • hard mode separation between live and replay
  • click-to-explain evidence inspection
  • counterfactual/model artifacts with concrete cases attached
  • epochal reconfiguration
  • composite or local-namespaced identity
  • caveated delegated rights
  • labeled compartments
  • viewpoints and bidirectional projections over one substrate
  • lineage-preserving derived artifacts

The resulting screen model and interaction loops now live in moldable-operations-studio-wireframes.

1. One state, many surfaces

The combination of yang-wigdor-2014-panelrama, klokmose-et-al-2015-webstrates, bragdon-et-al-2011-code-space, and danielsson-alvinius-larsson-2014-common-operating-picture gives a fairly complete answer to the LCARS/DRADIS impulse.

The shared lesson is that one canonical operational substrate can support:

  • a public wallboard
  • private specialist panes
  • handheld or desktop staging surfaces
  • role-specific situational views

without needing separate truth stores. The room view is the common operating picture; the personal view is a local staging and interpretation surface.

2. Mode clarity is a formal UI problem

nandiganahalli-et-al-2014-mode-confusion-detection and honarmand-torrellas-2014-replay-debugging jointly sharpen the WarGames lesson. Live, replay, and simulation should not merely look different. They should be backed by different runtime semantics and validated mode transitions.

This suggests:

  • explicit mode enums on views and traces
  • assertions against illegal mode crossings
  • replay opened from immutable checkpoint data in isolated execution

3. Evidence should answer why, not only what

ko-myers-2009-java-whyline, stasko-gorg-liu-2008-jigsaw, andrews-north-2012-analysts-workspace, and groth-streefkerk-2006-provenance-annotation-visual-exploration make the Minority Report idea practical.

The recipe is:

  • linked evidence views
  • spatial arrangement of hypotheses and artifacts
  • replayable provenance trails
  • click-to-explain traversal from observed artifact to causes

The important move is to make the evidence workspace an active reasoning instrument rather than a graveyard of logs.

4. Counterfactual/model artifacts are not fantasy

amershi-et-al-2015-modeltracker and wexler-et-al-2019-what-if-tool make the Prime Radiant idea much less mystical. The workable version is not an oracle but an inspectable artifact that binds:

  • aggregate metrics
  • concrete failing cases
  • editable inputs or assumptions
  • immediate what-if consequences

That suggests a model or branch inspector that is simultaneously evaluative, diagnostic, and counterfactual.

5. Epochs and identity can be explicit

malkhi-lamport-zhou-2008-stoppable-paxos, rfc-9420-mls-protocol, and rivest-lampson-1996-sdsi fill in the Machineries-of-Empire and Ancillary Justice threads.

They suggest:

  • explicit stop-and-reconfigure barriers for policy or protocol epoch changes
  • group membership epochs with rolled security/context material
  • linked local names and memberships instead of one flattening actor id

The studio can therefore represent reconfiguration, embodiment change, and membership rollover as formal control-plane events rather than vague settings churn.

6. Bounded delegated rights and labeled memory are real

birgisson-et-al-2014-macaroons and efstathopoulos-et-al-2005-asbestos make the Quantum Thief idea concrete.

The practical recipe is:

  • caveated capabilities for delegated rights
  • label-aware runtime compartments for mixed-principal work

This fits elegantly with the existing authority_budget and label-set objects in moldable-operations-studio-schema-pass.

7. Multiple valid views over one substrate

finkelstein-et-al-1992-viewpoints, foster-et-al-2007-bidirectional-tree-transformations, and green-karvounarakis-tannen-2007-provenance-semirings operationalize the The City & The City and Diaspora lines.

They imply that:

  • a view is a first-class viewpoint
  • crossing between views should be explicit
  • editable views should be lenses back to canonical state
  • derived artifacts should preserve formal lineage rather than flattening ancestry

This is a rigorous basis for view_definition and branch lineage in the studio.

Most actionable design refinements

The architecture and schema pages already named the right large primitives. This research pass fills in how to sharpen them.

A. Refine view_definition

Add explicit fields or subcontracts for:

  • public vs private visibility
  • promotion-to-room actions
  • mode enum (live, replay, simulation, inferred)
  • viewpoint/lens metadata
  • surface capability constraints

B. Refine trace

Add:

  • mode
  • replay_isolation = true|false
  • whyline_entrypoints or causal query anchors
  • provenance bookmarks / annotations

C. Refine session_token and actor identity

Add:

  • local names or alias sets
  • embodiment/device membership
  • epoch binding

D. Refine authority_budget

Add:

  • caveats as structured predicates
  • attenuation-only delegation
  • compartment constraints

E. Refine artifacts and lineage

Add:

  • formal derivation expressions or lineage sets
  • concrete-case links for summary metrics
  • viewpoint-crossing receipts when content is promoted between surfaces

Main caution

The sci-fi ideas survive only when the underlying semantics are disciplined. Without explicit mode boundaries, lineage, epochs, and caveats, the studio becomes a theatrical dashboard over a pile of hidden mutable state. That is not a control plane. It is décor.

Read this with moldable-operations-studio-wireframes, sci-fi-audit-for-moldable-operations-studio, moldable-operations-studio-architecture-spec, moldable-operations-studio-schema-pass, legacy-distributed-systems-ideas-for-moldable-operations-studio, and web-patterns-for-non-linear-harness-interfaces.

Graph View

Backlinks