design: should directive validate workflow outcomes, not just task completion?

[visionik]

Question

Directive tracks task completion via vBRIEF lifecycle and validates structural quality via task check, test coverage, and Greptile review cycles.

But it has no mechanism for asking: did this work actually create real value, or did it just feel productive?

The "theater vs real value" framing comes from a thread by @heygurisingh (https://x.com/heygurisingh/status/2055187352008716328), which attributes this distinction to an Anthropic researcher:

Identify what is saving you real time (measurable hours back) vs what feels productive but is not (theater) vs what you are doing manually that an LLM should be doing.

Why this might matter for directive

Directive's current success signal is: spec approved -> tasks completed -> tests pass -> PR merged. That is process quality. Outcome quality is a different axis -- did the shipped feature actually reduce friction, close the real bottleneck, or match what the user needed?

A few angles to explore:

1. vBRIEF outcome narratives: Should vBRIEFs include an optional post-completion OutcomeReview narrative -- a brief retrospective on whether the completed scope delivered the expected value? This would surface "theater" work before the next refinement cycle picks up related issues.

2. Refinement Phase 0 outcome check: Before triaging new issues, should deft-directive-refinement ask: "of the last N completed scopes, which ones are you actually using? Which ones sat on the shelf?" This is a lightweight signal for calibrating what gets priority next.

3. deft-setup delegation audit question: Related to feat(setup): add delegation-boundary questions to deft-directive-setup Phase 2 #1160 -- if setup captures the delegation boundary, a follow-up audit skill could periodically ask whether the actual boundary matches the intended one.

Open questions

• Is this directive's job, or the operator's job? Directive already trusts the operator to define what to build; outcome validation may be outside the framework's scope.
• What would "theater detection" even look like in a code framework? It may be more applicable to knowledge-work tools than to software build workflows.
• Could outcome narratives become noise that clutters vBRIEF history without adding signal?

Related

• feat(setup): add delegation-boundary questions to deft-directive-setup Phase 2 #1160 (delegation-boundary questions in deft-setup -- the input side; this is the output/feedback side)

Low-priority design question -- worth a conversation before deciding whether any implementation follows.