feat(ai): research-backed questionnaire improvements (high-ROI subset) - #21
Open
zyx1121 wants to merge 1 commit into
Open
feat(ai): research-backed questionnaire improvements (high-ROI subset)#21zyx1121 wants to merge 1 commit into
zyx1121 wants to merge 1 commit into
Conversation
Strengthen the AI pre-interview prompt with the highest-ROI findings from a citation-verified literature review (lib/ai/prompt.ts), and document the full plan + bibliography in docs/questionnaire-research.md. Landed in this pass: - Prediction-vs-judgment split per workflow step (Agrawal/Gans/Goldfarb 2018) — the most actionable suitability lens, beyond the existing three frameworks. - Ground-truth consensus probe in the Data phase (Lebovitz et al., MISQ 2021) — a feasibility blocker the prior SML "labeled data" framing hid. - Quantified baseline quad before advancing a task (volume×freq / time / cost-FTE / rework rate) so downstream prioritization isn't starved (McKinsey 2026). - Day Reconstruction Method for time use, replacing the biased "typical day" global estimate (Kahneman 2004; Kan & Pudney 2008; Belli 2000). - MUST-probe-on-vagueness + idiographic-probe default + a probe cap, since under-probing is the dominant AI-interviewer failure (Wuttke 2024; Jacobsen 2025; Xiao 2020). - Lightweight leadership-sponsorship + change-history signals in Context (Jöhnk 2021; MIT-BCG 2020). - Question-form anti-pattern rewrites (Zaremba & Liaskos, IEEE RE'21). The agent still only records signals; it does not score suitability (the consultant does, in phases 5-10). Deferred lower-ROI items + the verified bibliography (51 sources, 0 fabricated; 14 metadata corrections applied) are in the doc for you + CC to pick from. No prod/DB impact. tsc + eslint clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Not auto-merged — for your + CC's review, since the interview prompt is product IP and interview length is a real tradeoff. Run
git diff main -- lib/ai/prompt.tsto see the question-level changes.What & why
A multi-agent literature review (51 sources, citation-verified: 37 accurate / 14 metadata-corrected / 0 fabricated) on enterprise AI-adoption interviewing, task-suitability assessment, and self-report accuracy. This PR lands the high-ROI subset into
lib/ai/prompt.tsand documents the full plan + verified bibliography indocs/questionnaire-research.md.The agent still only records signals — it does not score suitability (the consultant does that in phases 5–10). Net interview length increase is bounded to ~+2–3 turns per task (probes are folded into existing steps + a per-point probe cap was added).
Changes to the prompt
Grounding comment expanded from 3 anchors to the full anchor set with citations.
Deferred (in the doc, for you to pick)
Context C3/C4/C6 (objective frame / external pressure / MGI triage) · Workflow W3–W5 (success criteria, reasoning-chain, Webb verb heuristic) · Pain P1–P6 (incident-driven CIT/CDM, error-tolerance, drift, negative-balance, workarounds) · Data D4/D5 (artifact-grounding, forced-choice) · Wrap-up WU1–WU4 (quantified read-back, E1/E2, readiness sheet, clearinghouse gating).
Verification
bunx tsc --noEmit✅ ·bun run lint✅ · prompt-only change, no DB / runtime / API surface impact.🤖 Generated with Claude Code