[witbrock]

User outcome

Von coding agents should default to capable, low-friction help for ordinary administrative and scientific work, rather than preserving an aircraft-style control system because old tests and guidance name its stages. The root cause addressed here is accumulated mechanism-specific doctrine: category names, representedness scores, workflow choreography, evaluator taxonomies, and automatic closure procedures became architecture authority instead of evidence about the user job.

Material changes

• Make AGENTS.md say to bound available capability rather than model reasoning, let durable authority earn its layer, and put the burden of justification on compulsory semantic controls. Conceivable risk is insufficient: a restriction needs evidence or a clear causal demonstration of a specific credible material outcome, plus evidence that the least restrictive bounded, observable, and recoverable approach is inadequate. This does not require a prior incident or formal proof.
• Replace competing Copilot/Gemini instructions and the subsystem/risk-class PR checklist with small routing and outcome/evidence surfaces. Keep the root Gemini file as the single routing entrypoint.
• Prune domain-specific and controller-specific contracts from the VWL and replay manuals; direct function, tool, model, and workflow paths are peers whose value is judged end to end.
• Retire legacy weighted benchmark protocols and represented-fraction scores as policy, while keeping short compatibility notes for current consumers.
• Recast the current namespace resolver topology as an implementation compatibility record; private-data isolation is the protected outcome, not one permanent resolver and telemetry design.
• Make prompt, critic, browser-validation, Jira, and represented-authority guidance conditional on demonstrated need rather than standard architecture.

The diff changes 21 Markdown files and is net subtractive by 1,069 lines.

Evidence

• git diff --check
• Validated all changed Markdown files: relative link targets resolve and fenced code blocks balance.
• Searched active agent-facing guidance for the retired categorical-boundary wording, represented-fraction acceptance, workflow-driven-success requirements, and the weaker comparative-evidence wording.
• Live Jira descriptions JVNAUTOSCI-2595 through JVNAUTOSCI-2603 were updated separately and read back exactly. The strengthened evidential burden was applied to the Epic and relevant implementation/human-gate tasks; all statuses, hierarchy, labels, and the user-edited Epic paragraph were preserved.
• Both review threads were dispositioned: the Gemini routing mismatch was fixed, while duplicating the constitutional secret rule in the PR template was declined to avoid rebuilding checklist drift. Secret Scanning remains active.

No runtime or production code changed, so no code test suite was run.

Decision boundary

Michael explicitly accepted this constitutional rewrite on 25 July 2026 and authorised it to merge before the revised A1 experiment begins. The merge does not itself authorise A2, production cutover, or a human-gate decision.

[Copilot]

Pull request overview

This PR updates Von’s agent-facing documentation to replace category-driven “control doctrine” (stages, representedness scores, compulsory gates) with permissive defaults that emphasise bounded delegated capability, proportionate risk/recovery controls, and evidence-based justification for any new compulsory restriction.

Changes:

• Updates AGENTS.md, design index routing, and agent instruction entrypoints (Copilot/Gemini) to make “smallest adequate end-to-end path” and “bound capability, not reasoning” the baseline doctrine.
• Prunes / reclassifies legacy workflow/controller/benchmark protocols into compatibility notes and removes domain- and stage-specific contracts from general manuals.
• Simplifies PR authoring guidance and checklists to focus on user outcome, material changes, evidence, and decision boundary.

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/GEMINI.md	Repurposes Gemini notes into a routing stub pointing to AGENTS.md / docs/design_index.md.
GEMINI.md	Repurposes root Gemini notes into routing guidance aligned with AGENTS.md.
docs/engineering/von_workflow_language_manual.md	Reframes VWL semantics as optional features selected by capabilities; removes controller/UI/domain contract material from the language manual.
docs/engineering/Von_for_AgenticAI.md	Clarifies represented authority as optional and justified by evidence; avoids implying standard stage architecture.
docs/engineering/security_considerations.md	Replaces verb/category-based guardrails with delegated-capability and residual-risk guidance; clarifies outcome vs mechanism.
docs/engineering/real_path_server_replay_and_telemetry_loop.md	Makes replay protocol architecture-neutral (direct/tool/model/workflow paths are peers) and proportionate to the claim.
docs/engineering/prompt_programs_and_model_routing_playbook.md	Reframes prompt governance as conditional on deliberate selection; replaces certification/checklist doctrine with proportionate replay evidence guidance.
docs/engineering/operational_engineering_guide.md	Updates operational guidance to avoid fabricating unavailable authority while preserving safe alternatives and bounded progress.
docs/engineering/minimal_imposition_design_principle.md	Recasts minimal imposition as explanatory guidance under AGENTS.md rather than an independent gate taxonomy.
docs/engineering/minimal_imposition_benchmark_model.md	Retires the weighted benchmark model as policy; keeps only compatibility/interpretation boundaries for current consumers.
docs/engineering/maintaining_global_design_constraints_and_authority_alignment_with_coding_agents.md	Simplifies authority-boundary diagnosis to match the new constitutional framing and evidence expectations.
docs/engineering/intro_to_modern_agentic_ai_for_coding_agents.md	Updates primer to warn against both hidden Python policy and compulsory represented/staged architecture; adds permissive baseline framing.
docs/engineering/frontend_browser_user_view_validation.md	Replaces “validation ladder” with a menu; emphasises choosing the cheapest sufficient evidence for the claim.
docs/engineering/execution_correctness_benchmark_protocol.md	Converts the legacy benchmark protocol into a compatibility note with explicit non-authority boundaries.
docs/engineering/effective_namespace_contract.md	Recasts namespace doc as an implementation compatibility record; preserves the protected outcome (private-data isolation) over topology.
docs/engineering/agent_evaluation_and_research_uptake.md	Adds evaluation prompts for useful-action rate, false refusals, intervention burden, and residual harm weighted by consequence.
docs/design_index.md	Updates routing language and the status/authority descriptions of referenced docs to match the new doctrine.
AGENTS.md	Core constitutional update: permissive defaults, capability ceilings, proportional controls, and evidence requirements for compulsory restrictions.
.github/PULL_REQUEST_TEMPLATE.md	Replaces subsystem checklist with outcome/evidence/decision-boundary prompts (and removes explicit hygiene reminders).
.github/copilot-instructions.md	Converts Copilot instructions into routing to AGENTS.md + docs/design_index.md, removing duplicated rule surfaces.

[Copilot]

Pull request overview

Copilot reviewed 20 out of 20 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated no new comments.