Decision Rights

Downstream-Instruction Framing Discipline

Problem

LLMs default to authoritative downstream instructions: "Approved Decisions — Do Not Re-Litigate," "Execute the following steps," "These are final." This is usually wrong.

The downstream reader often has evidence the author lacks: target-repo access, tests, and local rules. Command framing can make them silently follow a plan the code contradicts.

Test that asymmetry per decision: who can access the evidence that bears on it? Downstream often knows the target code and tests better; upstream may know legal rules, security policy, product intent, or external facts unavailable in the worktree.

Evidence and decision authority are separate. Better evidence licenses a revisit request, not the right to change scope, values decisions, or explicit user constraints.

Context / Trigger Conditions

Apply when writing handoffs, plans, dispatch prompts, workflow briefs, future-work ADRs, or instructions to a fresh session.

Symptoms: blanket "Do Not Re-Litigate" / "These are final" headers; evidence-sensitive decisions without "Revisit if:"; no per-decision evidence test; or proposed work written as commands.

Solution

1. Default framing: informed proposals from a less-informed reviewer

Open with a framing block adapted to the actual evidence asymmetry. Template:

## How to Treat This Document

You are not being given orders. You are being given the prior session's best research
output, written by a session that [name the evidence asymmetry — e.g., could not read
the target codebase directly / worked from summaries only / had no test-run authority].

Each decision below is labeled revisable or non-negotiable. A `Revisit if:` clause names
the evidence that can reopen a revisable decision and who can access it.

Therefore: treat the prior decisions and recommendations as *informed proposals from a
less-informed reviewer*. They reflect what looked right from outside [the relevant
context]. You are inside [the relevant context] now. This default applies only where you
have better evidence access for that decision.

You are explicitly licensed and encouraged to:
- Disagree with the recommendations if [evidence access] reveals the framing was wrong.
  Surface the disagreement to the user with reasoning; don't silently follow a bad plan.
- Restructure sequencing if the real dependency graph differs.
- Reject items that turn out to be solved already, duplicative, or premature.
- Redesign approaches if the proposed design conflicts with what the actual code shows.

What you should NOT do:
- Reopen instructions labeled non-negotiable. Values decisions and explicit constraints
  remain imperative unless their decision authority changes them.
- Silently deviate. The user is the decision authority on scope changes; surface
  disagreement with reasoning, then let them decide.
- Infer that better facts grant decision rights. New evidence licenses the requested
  revisit named below; it does not license unilateral scope changes.

2. Classify each decision; add "Revisit if:" only when evidence can change it

For every decision ask: what evidence could change this outcome, and who can access it? Label it Revisable with new evidence and add a specific Revisit if: only when such evidence exists. Label values decisions and explicit user constraints Non-negotiable; keep them imperative.

## Decision Status

- **Revisable with new evidence:** Phase X expands to include items A, B, C. *Revisit if:*
  repository inspection shows any item is already done,
  duplicative, or has a wrong cost estimate by >2x.
- **Revisable with new evidence:** Design Y was approved. *Revisit if:* the actual architecture
  makes it incompatible,
  or there's a simpler/better way you can see from inside the tree.
- **Non-negotiable:** Do not send customer data to third parties. This is an explicit user
  constraint; surface any conflict, but do not override it.

3. Acceptable narrow use of "do not re-litigate"

Use the phrase only for a specific question the user explicitly closed in the current conversation.

Acceptable: "User decided 'PostgreSQL not SQLite' in turn 14. Do not re-litigate this choice."

Not acceptable: a blanket "Approved Decisions — Do Not Re-Litigate" header.

4. Imperative → proposal mood

Imperative mood remains appropriate for legal and security boundaries, values decisions, NEVER-tier rules, and explicit non-negotiables. Better evidence does not soften them.

Verification

After drafting any downstream-instruction artifact:

• Each decision tests who can access the evidence that bears on it
• Evidence and decision authority are separate
• Every evidence-sensitive decision is labeled revisable and has a specific "Revisit if:"
• Values decisions and explicit user constraints are labeled non-negotiable and imperative
• Disagreement-with-reasoning is licensed only for revisable instructions
• "Do not re-litigate" is absent, OR scoped to a single explicitly-closed question
• Revisable recommendations use proposal mood; genuine hard constraints do not
• Downstream is required to surface disagreement rather than silently deviate

If any answer is no, revise before delivery.

Example

Anti-pattern (caught and corrected 2026-06-07, a private production project's security handoff):

## Approved Decisions (Already Made — Do Not Re-Litigate)

- Phase E expands to include E-ADD-1 through E-ADD-5.
- The security-item δ-pattern design is drafted. Recommended absorption as E-ADD-6.

Corrected version:

## User-Approved Directional Decisions (Revisable With New Evidence)

These were approved by the user against the framing this session presented. If reading the
actual codebase changes the framing, surface the change and let the user re-decide.

- Phase E expands to include E-ADD-1 through E-ADD-5. *Revisit if:* any item is already
  done, duplicative, or has a wrong cost estimate by >2x.
- The security-item δ-pattern design is drafted. Recommended absorption as E-ADD-6.
  *Revisit if:* the actual architecture makes the separation incompatible, or there's a
  simpler way to close the threat you can see from inside the tree.
- **Non-negotiable:** Preserve the user-set security boundary; surface conflicts, do not
  override it.

Notes

• Subagent prompts are riskiest. Agents read dispatches as near-system-tier and rarely push back; make their framing block explicit.
• Does not apply to genuine hard constraints. Security boundaries and user-stated non-negotiables are imperative for legitimate reasons. The skill governs the default framing of proposed work.
• Use "Revisit if:" only when evidence can change the outcome. Otherwise mark a values decision or explicit user constraint non-negotiable; a fake revisit condition is noise.
• Failure prevented: downstream silently executes a plan its evidence contradicts.