The House with a Million Windows: Interactive Fiction for Narrative Restorying
Cody Kommers, Sarah G Immel, Drew Hemment, Mina Lee
cs.CL · cs.HC
Quick verdict
LLM interactive-fiction tool for restorying personal stories; NISE gains (dz=0.67, N=40) lack a control group—engaging design, causal evidence still weak.
Summary
The paper presents The House with a Million Windows (HWAMW), an LLM-based interactive fiction prototype for personal narrative restorying. Users tell a true personal story to a fictional Gatekeeper, explore three of five LLM-generated 'windows' that retell it in canonical literary styles (e.g., Pride and Prejudice, To the Lighthouse), then retell the story in their own words. Built on Llama-3.3-70B-Instruct as a Next.js app, the system reframes AI output as perspective-opening windows rather than drafts to accept. A within-subjects study (N=40, Prolific, UK/IE) found NISE narrative-identity scores rose from 4.55 to 4.75 (t(39)=4.24, Holm-corrected p<0.001, dz=0.67), driven by autobiographical reasoning and positive themes; LIWC-22 metrics on retold stories (N=32) rose for staging and plot progression, and VADER sentiment increased. An expert review (N=10) found the framing preserves authorship while surfacing LLM flaws like clichéd prose and hallucinations. Main limitations: no control condition weakens causal claims; the narrative and sentiment analyses lack inferential statistics; a collected in-story meaningfulness measure goes unreported; effect durability is unknown; code is promised but not yet released.
Review complete
Assessment at a glance
5
risk flags identified
● high 1
● medium 3
● low 1
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- A single session with HWAMW significantly increases self-reported narrative identity (NISE 4.55→4.75, t(39)=4.24, Holm-corrected p<0.001, dz=0.67, N=40).
- Retold stories shift measurably after the experience: LIWC staging (21.27→32.62) and plot progression (9.16→28.27) rise, and VADER sentiment increases from 0.32 to 0.61.
- Framing LLM outputs as 'windows' rather than drafts lets users restory personal narratives while retaining authorship and agency (expert review, N=10).
- Metafictional framing differentiates the system from therapy and makes engagement with vulnerable personal stories feel safer and more natural.
Strengths
Novel and well-argued design synthesis
Combines restorying psychology (Rogers et al. 2023), interactive fiction, and AI writing interfaces into a genuinely new framing: LLM outputs as 'windows' onto the user's story rather than drafts, explicitly linked to Vallor's AI mirror (§3.2.2) and metafiction (Borges, Haig) to avoid therapy-bot framing (§3.1).
Exemplary prototype documentation
Model (Llama-3.3-70B-Instruct, locally hosted), stack (Next.js, SQLite), full style/structure prompt rules for all five windows (Appendix B), word limits, and the complete three-phase user flow are published — rare transparency for a prototype paper.
Disciplined self-report analysis
Honest mixed-methods reporting
Careful ethics positioning
Weaknesses
No control condition
Descriptive-only narrative analyses
Unreported collected measure
Window allocation uncontrolled
Small, single-instrument, possibly transient effect
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
HWAMW (arXiv:2609.12537, cs.CL/cs.HC) is an LLM-based interactive fiction prototype for personal narrative 'restorying': users tell a true story to a fictional Gatekeeper, explore three of five LLM-generated 'windows' recasting it in canonical literary styles (Austen, Woolf, Hemingway, Poe, Hughes; full prompt rules in Appendix B), then retell it in their own words. The paper claims three contributions: (1) a metafictional restorying design framing LLM outputs as windows rather than drafts; (2) quantitative evidence that one session strengthens self-reported narrative identity (NISE 4.55→4.75, dz=0.67) and measurably changes retold stories; (3) a qualitative expert account (N=10) of how the effect is achieved. The headline claim — that HWAMW 'increases' narrative identity — is causal language the study design cannot fully support.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is The House with a Million Windows?
An LLM-based interactive fiction prototype in which users tell a true personal story to a fictional Gatekeeper, explore AI-generated retellings of it in the style of canonical literary works (presented as 'windows', e.g. Pride and Prejudice), then retell the story in their own words. It runs on Llama-3.3-70B-Instruct and is designed for reflection and meaning-making, not as a writing autocomplete.
Did using it improve users' well-being or narrative identity?
In a study of 40 participants, scores on the Narrative Identity Self-Evaluation (NISE) scale rose from 4.55 to 4.75 after one session — statistically significant with a moderate effect size (dz=0.67). However, there was no control group, so the change cannot be firmly attributed to the system itself, and how long any effect lasts is unknown.
Is The House with a Million Windows a therapy tool?
No. The authors are explicit that it is a narrative/creative-writing experience with no clinical aims or oversight, and they warn that it should not be used by psychologically vulnerable populations or treated as a therapy tool (the study had ethics approval, ref 921742).
Which AI model and prompts does HWAMW use?
The window narratives are generated with Llama-3.3-70B-Instruct, hosted locally at the University of Edinburgh, inside a Next.js web application with SQLite storage. The paper publishes the complete style and structure rules for all five literary windows in Appendix B.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figure 4's plotted data points are not recoverable from extraction; exact pre/post values rely on the in-text statistics (and the extracted axis tick labels read 1–8 for a 1–7-point instrument, likely a rendering artifact).
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.