RIGOR Agent-First Architecture Figure Prompt

Status: Visual concept only — not implementation requirements, paper claims, or qualification evidence

This brief helps a human illustrator or image-generation tool create a concise RIGOR architecture figure for documentation and presentations. Manuscript-specific RIGOR figure sources and claim boundaries live under research/rigor-paper/figures/.

The manuscript also maintains two paper-specific illustrated variants:

• figure_spec_agent_first_vision.md with rigor-agent-first-laboratory-vision.png for the Introduction;
• figure_spec_discovery_ecosystem_vision.md with rigor-discovery-ecosystem-vision.png for the Discussion target vision.

These variants use RIGOR terminology and explicit evaluated-versus-future scope. The broader RIGOR documentation images remain separate assets under docs/assets/.

Narrative

The figure should communicate one primary scientific narrative:

Human intent
  -> Agent planning
  -> exact-input Rehearsal
  -> deterministic execution
  -> semantic laboratory
  -> verified evidence
  -> scientific decision
  -> next iteration

The architecture should not be presented as six equal services or six peer Agents. ASCEND is the persistent goal owner, while Composer and PRISM are bounded specialist workers. Deterministic services retain authority for policy, execution, verification, evidence commitment, cancellation, and safe-stop. The current platform qualifies software, simulation, algorithmic analysis, and governed human work. Any pictured physical endpoint remains a target that needs separate device and effect qualification.

Primary Figure Prompt

Copy the following prompt into an image-generation system or use it as a design brief for a scientific illustrator. Figure text should remain in English.

Create a presentation-ready scientific architecture figure, titled:

“RIGOR: Rehearsal-bound, evidence-grounded agent-first experimentation”

FORMAT AND VISUAL STYLE
- Single-panel landscape scientific infographic, approximately 16:9 or
  180 mm × 100 mm.
- Clean editable vector appearance suitable for SVG/PDF reconstruction.
- White background, flat design, restrained Nature-style editorial aesthetics.
- Thin dark-gray strokes, no shadows, no gradients, and no 3D rendering.
- Use a color-blind-safe palette:
  blue for agent control and planning,
  amber for deterministic policy and execution,
  teal for the laboratory capability boundary,
  purple for evidence and scientific analysis,
  neutral gray for provenance and infrastructure.
- Use generous whitespace, consistent rounded rectangles, and professional
  sans-serif typography.
- Use no more than four large system blocks, one small researcher icon, five
  primary arrows, and one dashed shortcut.
- Make the information flow visually dominant and understandable within five
  seconds.
- Avoid dense infrastructure details, databases, API endpoints, ports, code,
  and excessive arrows.
- Preserve all requested English labels exactly. Do not invent additional text.

COMPOSITION
Arrange four large numbered blocks from left to right, forming one
evidence-driven closed loop. Place a small “Researcher” icon to the far left.
Use a single broad curved return arrow from the final block back to the first
block.

RESEARCHER
Use a small, restrained researcher icon outside the automated loop.

Input to RIGOR:
“Scientific goal”
“Constraints”
“Autonomy policy”

Information returned to the researcher:
“Progress & evidence”
“Needs you”

The researcher must not appear to manually advance routine lifecycle stages.

BLOCK 1 — GOAL-DIRECTED CONTROL
Header:
“1  Goal-directed control and planning”

Module subtitle:
“ASCEND Supervisor · Composer Specialist”

INPUT
“Goal, policy, current state,
and committed evidence”

INTERNAL LOGIC
Show two compact internal layers:
- “ASCEND: intent, revisable plan, next-round decision”
- “Composer: Skills, Procedures, capabilities, immutable RunSpec”

OUTPUT
“Semantic intent”
or
“Validated immutable RunSpec”

Add one short statement at the bottom:
“Plans may change; committed actions do not.”

BLOCK 2 — REHEARSAL-BOUND EXECUTION
Header:
“2  Rehearsal-bound deterministic execution”

Module subtitle:
“ASCEND Rehearsal · Policy · PACE · LabFlow”

INPUT
“Intent or RunSpec”
“Capability and Procedure contracts”

Show a compact policy gate:
“ALLOW  |  ASK  |  DENY”

Before the execution row, show a prominent isolated Rehearsal gate:
“Exact immutable intent”
“Frozen ExecutionSnapshot”
“Selected completed branch”

Use a short correction arrow from a failed branch back to planning. Mark all
Rehearsal outputs as “predicted” and keep them outside the evidence path.

Below it, use a single horizontal row of small symbols or short labels:
“Schema · Idempotency · Locks · Timeout · Cancel · Safe-stop · Verify”

OUTPUT
“Scheduled, bounded,
and verifiable operations”

A small amber exception arrow may point toward the researcher and be labelled:
“ASK → typed Attention”

Do not depict human approval as a mandatory stage. It is an exception channel.
DENY must visibly prevent execution rather than route around the policy gate.

BLOCK 3 — SEMANTIC LABORATORY BOUNDARY
Header:
“3  Semantic capability boundary”

Module subtitle:
“LabBridge · Controllers · Laboratory”

INPUT
“Typed capability call”
“Expected result and evidence contract”

Inside the block, show LabBridge as a narrow gateway between software and four
minimal endpoint icons:
- “Simulated”
- “Algorithmic”
- “Human-assisted”
- “Future physical”

Use solid outlines for the first three endpoint classes. Use a dashed gray or
gold outline for “Future physical” and add the small label “separate
qualification”.

The gateway may route internally through:
“MQTT · Device controller · Algorithm service”

These protocol labels must remain visually subordinate.

OUTPUT
“Operation result”
“State · Measurement · Artifact”
“Verification status · Evidence references”

Add a compact emphasized statement:
“Agents select capabilities, not device protocols.”

Do not expose GPIO, PWM, MQTT topics, serial commands, robot poses,
trajectories, joints, or device SDK calls to the Agent-facing side.

BLOCK 4 — EVIDENCE COMMITMENT AND ANALYSIS
Header:
“4  Evidence commitment and scientific analysis”

Module subtitle:
“Immutable PACE delivery · PRISM Analyst”

INPUT
“Verified operation results”
“Measurements, artifacts, and provenance”
“Versioned Analysis Skill”

Show a small evidence gate before analysis:
“Verify checksum and completeness”

Then show an isolated analysis workspace:
“Read-only evidence input”
“Validated structured output”

OUTPUT
“Committed result”
“Report · Uncertainty · Evidence citations”
“Minimal scientific projection”

Add a compact emphasized statement:
“Evidence before claims.”

MAIN INFORMATION FLOW
Use exactly these four prominent arrow labels:

Block 1 → Block 2:
“Exact intent / immutable RunSpec”

Block 2 → Block 3:
“Rehearsal-attested semantic operation”

Block 3 → Block 4:
“Verified result and evidence”

Block 4 → Block 1:
“Committed findings inform the next iteration”

The last arrow should curve cleanly above or below the blocks to communicate
closed-loop autonomous experimentation without crossing other arrows.

DIRECT ATOMIC PATH
Add only one thin dashed shortcut from Block 1 directly to Block 3, labelled:

“Low-risk atomic action”

Add a very small note:
“Same Rehearsal, policy, Operation, verification, evidence, and trace”

This path communicates progressive formalization: a simple bounded action does
not require a workflow or experiment package, while reproducible multi-step work
continues through PACE and LabFlow.

END-TO-END PROVENANCE
At the bottom, add one narrow neutral-gray ribbon spanning all four blocks:

“Goal & Policy → Profile / Context / Tool versions → PlanRevision / RunSpec
→ Operation → Evidence / Artifact → Scientific Decision”

Label the ribbon:
“Append-only lineage · audit · deterministic replay”

Do not draw a single central database. Indicate that domain records remain owned
by their authoritative modules and are connected through stable references.

CENTRAL SCIENTIFIC MESSAGE
In the whitespace near the closed-loop arrow, place one concise statement:

“Agent plans.
Deterministic code executes.
Committed evidence drives the next decision.”

The final composition should communicate:
- the researcher defines or changes goals and constraints;
- ASCEND is the persistent goal owner;
- Composer and PRISM are bounded specialist workers, not peer autonomous
  controllers;
- policy and deterministic execution remain separate from model reasoning;
- human involvement occurs only for important decisions, genuinely physical
  work, safety boundaries, or exhausted automatic recovery;
- transport dispatch is not equivalent to physical success;
- every scientific conclusion is linked to verified evidence and provenance;
- the system can autonomously continue to the next scientific iteration when
  policy permits;
- Rehearsal predicts the exact intent but never becomes scientific evidence;
- physical instruments and effects remain future qualification targets.

OUTPUT REQUIREMENTS
Use crisp vector-like lines, balanced margins, legible text at single-column
print size, CMYK-safe colors, no decorative complexity, and no visual clutter.
Use no more than two short text levels inside each block. Produce
publication-ready scientific storytelling.

Negative Prompt

Avoid generic AI brain icons, humanoid robots representing the Agent,
cyberpunk aesthetics, neon colors, holograms, dark backgrounds, photorealistic
laboratory scenes, cloud-computing diagrams, microservice spaghetti, network
topology, database cylinders, many crossing arrows, tiny unreadable text,
decorative circuitry, six equal peer Agents, mandatory human approval at every
stage, raw GPIO or MQTT commands, unverified data flowing directly into a
scientific conclusion, and optimistic “command sent = experiment succeeded”
semantics.

Recommended Figure Caption

RIGOR agent-first experimentation model. Researchers define scientific goals, constraints and autonomy policy, while an ASCEND supervisor delegates bounded planning and analysis tasks to Composer and PRISM specialists. Every experimental write first carries exact immutable Rehearsal lineage. Deterministic services govern policy, scheduling, device operations, verification and safe stopping through semantic capabilities. Only verified, provenance-linked evidence is committed for scientific analysis and subsequent experiment decisions. Low-risk atomic actions may bypass workflow formalization, but retain the same policy, operation, verification, evidence and trace contracts.

Interpretation Notes

• The main figure should communicate scientific and operational logic rather than process topology.
• ASCEND, Composer, PACE, LabFlow, LabBridge, and PRISM should not appear as six equal Agents.
• The normal reproducible route and the low-risk atomic route converge on the same Rehearsal, Operation, verification, evidence, audit, and trace semantics.
• Human Attention is an exception channel, not a routine approval stage.
• A transport acknowledgement is not evidence of physical success.
• The physical endpoint is aspirational and visually distinct from the current software, simulated, algorithmic, and governed-human scope.
• The CoN4Cl three-historical-rounds-to-fourth-design benchmark is better suited to a separate case-study figure or a Panel B than to this architecture panel.

Design Sources

• Agent-First System Design PRD
• Agent Work Information Management
• Agent-First Target Responsibility Architecture: diagrams/agent-first-target-architecture.puml
• Agent-First Design Showcase: diagrams/agent-first-design-showcase.puml
• CoN4Cl Control-Matrix Benchmark