Skip to content

sample-report.md

Latest commit

 

History

History
57 lines (42 loc) · 2.66 KB

sample-report.md

File metadata and controls

57 lines (42 loc) · 2.66 KB

Sample PromptIQ Report

PromptIQ Review

Metric Value
Score 7.4/10
Band competent
Confidence high
Complexity high
Delta +0.3 vs last compatible session

Why It Is Not Higher
This session had real steering and decent structure, but the finish line stayed too soft. Scope control was solid and course correction happened on time, yet the user still left verification implicit instead of saying how the final answer should be checked.

Dimension Breakdown

Instruction Clarity    ███████░░░  7/10
Context Provision      ███████░░░  7/10
Iteration Quality      ████████░░  8/10
Task Decomposition     ████████░░  8/10
Output Specification   ███████░░░  7/10
Example Usage          ──────────  N/A
Reasoning Elicitation  ████████░░  8/10
Tool Awareness         ███████░░░  7/10

Strongest Evidence

"Please review the failing deploy flow. First isolate whether the regression is in build, runtime, or auth middleware. Then propose the minimum fix and list the files you would touch."

This prompt is strong because it locks three things at once: scope, reasoning order, and output shape. The model is given a lane and a deliverable before it starts moving.

Course Corrections

  1. Input: "That diagnosis is too broad. Narrow it to the middleware path only and compare against yesterday's behavior."
    Patch: "Limit the analysis to middleware only. Compare today's behavior against yesterday's middleware diff and exclude build/runtime."
    Effect: Better boundary control. It closes off nearby branches the model would otherwise wander into.

  2. Input: "Do not refactor unrelated modules. I only want the smallest change that restores production behavior."
    Patch: "Give me the smallest safe patch. No refactors, no naming cleanup, and no unrelated file edits."
    Effect: Better output control. It specifies the acceptable shape of the answer, not just the general preference.

Next Session Drill
Add one explicit verification rule to every corrective follow-up. Do not only narrow scope. State how the result should be checked before the model writes it.

Recent Trend

Date Score Weakest
2026-04-10 6.8 Output Specification
2026-04-12 7.1 Context Provision
2026-04-13 7.4 Output Specification

Focus Area
Output Specification is still the system bottleneck across recent sessions. Tighten the finish line earlier, and pair it with a concrete check so the result can be proven instead of assumed.