Miadi Review

Engineering Review: Transforming Screenwalk Reviews into Living Research Paths

Oct 15, 2026 · model: manual-import · channel: JGWill

Watch source

In the chronicle

Engineering Review: Transforming Screenwalk Reviews into Living Research Paths

Source: screenwalk-261009-ep120-iris-review-continuation-v1-v2 — JGWill https://youtu.be/WkBLJqhZdOs

Summary

William orchestrates the collaborative evolution of the Miadi production methodology across mobile terminal sessions, Slack agent interactions with Iris, and synthesized podcast dialogues with Mia and Ava. By differentiating intentions from assignments and unverified proposals from completed software, the workflow proposes 'Screenwalk Research Continuation,' a practice meant to turn static recordings into revisitable research paths.

Timecodes and figures in this review are as summarized from the video, not checked. No transcript is stored with this review, the video's own YouTube description could not be read, and some timecodes are out of order or overlap across sections. Statements marked "checked" were read on GitHub or on the live review service on 2026-10-15 UTC.

Academic Fields

Human–AI Interaction (Human–Computer Interaction)

Studies how people understand, direct, and correct AI systems. The video touches it where William revises Iris's review drafts and listens to a spoken layer of agent output. Amershi's guidelines are heuristics, not outcome measures, and Parasuraman and Manzey report complacency and bias in reviewers of automation that training has not removed.

Multi-Agent Systems (Computer Science)

Studies how autonomous agents coordinate, communicate, and share work. The video touches it where later agents resume an investigation earlier agents began, through prose handoffs. Guo et al.'s survey of language-model agent systems lists hallucination, collective intelligence, and scaling as open challenges, so a prose handoff does not guarantee understanding.

Provenance and Research Data Management (Information Science)

Studies how records of origin make data findable, reusable, and checkable. The video touches it where review versions, a Langfuse trace (the grouped steps of one request), and a GitHub issue link to Episode 120. Provenance records what produced what, not whether it is true: the service stores each version's model, not its editor.

Human Oversight in Software Process (Software Engineering)

Studies how people check software work before accepting it. The video touches it where proposals are kept apart from verified implementation and drafts are scored on five criteria. Bacchelli and Bird found code review yields understanding and knowledge transfer more than defect finding, so a score or approval records judgment, not verification.

Learning Sciences (Education)

Studies how people learn from materials, talk, and practice. The video touches it in the Mia and Ava dialogue, where Ava asks what a listener would. Craig et al. found overhearing a tutoring dialogue beat a monologue on recall and deep questions in a lab. ICAP ranks passive reception lowest. No listener test is reported.

Narratology and Transmedia Storytelling (Literary and Media Studies)

Studies how events are selected, ordered, and retold. The video touches it where one recording becomes a review, a podcast, and a Chronicle entry. Jenkins asks each medium to add something of its own, so repeating content in each layer is closer to adaptation. Genette's order and duration show a telling rearranges events.

Aspects

Defining Screenwalks and Research Continuation (00:05-01:45)

A screenwalk captures a situated human-system encounter across prompts, actions, and code branches. Rather than treating reviews as mere summaries or task lists, the framework introduces an 'Intended Use & Continuation' section. This anchors active work, reality versus proposal, and open inquiries for future human and autonomous agents.

Distinguishing Proposals from Implemented Verification (10:05-11:15)

Critical discipline requires separating proposed design prompts from verified software execution. In reviewing Episode 120 drafts, William and agent Iris identify where Langfuse traces and GitHub integrations were treated as accomplished facts rather than design requests, enforcing explicit boundaries between recorded evidence and prospective implementation. Checked on GitHub: jgwill/Miadi#754 states the same boundary in its own text, "No implementation is claimed here", and calls the work a "proposed pre-production ingestion workflow".

Evaluative Scaffolding and Five-Dimension Scoring (31:15-32:20)

Version 2 review drafts are scored across five criteria (as summarized from the video, not checked): evidence integrity, intended use, agent continuity, narrative integration, and community participation. Rather than imposing opaque numerical rankings, the framework treats evaluation scores as visible editorial judgments that invite future participants to interrogate evidence without mistaking curiosity for executive authorization. The summary gives no scoring scale and no second rater, so the scores read as one reviewer's judgments, not measurements.

Conversational Assimilation via Dual-Agent Audio Layers (20:15-22:30)

Through an automated dialogue between synthetic personas Mia and Ava, the system turns technical reviews into an audio narrative meant to be easier to assimilate. Ava challenges assumptions, distinguishing research notebooks from handoffs, so that listeners can enter the story through scenes, execution traces, or open research questions without erasing complex history. No listener test is reported in this summary.

End-results and Core Aspect

Core Aspect — Screenwalk Research Continuation Architecture (36:30-38:05)

The core proposal names a four-stage pipeline: Screenwalk encounter, Review interpretation, Intended Use & Continuation framing, and Chronicle community exploration. This aims to turn screen recordings into version-controlled research artifacts from which subsequent agents can resume investigation without confusing retrospective documentation with active assignments. It is a design as summarized from the video, not a measured result.

Results

  • Versioned Review Multi-State Deployment (43:35-47:20): As summarized from the video, not checked: review service endpoints for Episode 120 were upgraded to support multi-version inspection (v1, v2, and v3). The aim is to let reviewers move between the baseline Gemini draft and later corrected versions and see how the interpretation changed. Checked on the live service: the Episode 120 review e29a4fc3-302a-4309-8ad9-39085ee26283 (of video HMXMDDe3HIA, the screenwalk jgwill/Miadi#754 names for Episode 120) lists ten versions, version 1 by gemini-3.7-flash and versions 2 to 10 as manual-import or manual-edit. The service records a model and a date per version, not who edited, so the upgrade, the v1-v2-v3 labels, and the attribution of corrections to Iris remain unchecked.
  • Trace-Linked GitHub Issue Doorway (45:35-46:30): jgwill/Miadi#754 was opened on 2026-10-09 as a traceable research doorway. Checked on GitHub: the issue names Episode 120's three tmux branches, links a Langfuse trace and session, and (in a comment) a Langfuse managed prompt. Its 2026-10-09 comment calls the trace link "a submitted target, not yet verified as navigable", and the trace itself was not opened here. Entry from embedded QR codes is as summarized from the video, not checked: no QR code appears in the issue text or its five comments.
  • Synchronized Production Metadata Packaging (40:45-42:30): As summarized from the video, not checked: the workflow generated YouTube publication metadata, including structured descriptions, branch attribution, and copy-ready review drafts. The intent is to keep each published media artifact traceable to its originating agent consultations, bash sessions, and evidence logs. One generated description shows the method working once. It does not show that every artifact will be traceable.

Sources

Amershi, S. et al. (2019). Guidelines for Human-AI Interaction. CHI 2019. https://doi.org/10.1145/3290605.3300233
Parasuraman, R., Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors 52(3), 381–410. https://doi.org/10.1177/0018720810376055
Wooldridge, M. (2009). An Introduction to MultiAgent Systems (2nd ed.). Wiley. ISBN 978-0-470-51946-2.
Guo, T. et al. (2024). Large Language Model based Multi-Agents: A Survey of Progress and Challenges. IJCAI 2024, Survey Track, 8048–8057. https://doi.org/10.24963/ijcai.2024/890
Moreau, L., Missier, P. (eds.) (2013). PROV-DM: The PROV Data Model. W3C Recommendation, 30 April 2013. https://www.w3.org/TR/prov-dm/
Wilkinson, M. D. et al. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data 3, 160018. https://doi.org/10.1038/sdata.2016.18
Langfuse. Observability data model. Documentation, read 2026-10-15 UTC. https://langfuse.com/docs/observability/data-model
jgwill/Miadi#754 (opened 2026-10-09). [Episode 120] Plot threads ↔ execution branches: voice-layer ingestion and trace-linked research prototype. GitHub issue and five comments, read 2026-10-15 UTC. https://github.com/jgwill/Miadi/issues/754
Miadi Review service. Episode 120 review e29a4fc3-302a-4309-8ad9-39085ee26283, version list, read 2026-10-15 UTC. https://miadi-review-service.vercel.app/review/e29a4fc3-302a-4309-8ad9-39085ee26283/json
Bacchelli, A., Bird, C. (2013). Expectations, outcomes, and challenges of modern code review. ICSE 2013, 712–721. https://doi.org/10.1109/ICSE.2013.6606617
Craig, S. D., Gholson, B., Ventura, M., Graesser, A. C., Tutoring Research Group (2000). Overhearing dialogues and monologues in virtual tutoring sessions: Effects on questioning and vicarious learning. International Journal of Artificial Intelligence in Education 11, 242–253.
Chi, M. T. H., Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist 49(4), 219–243. https://doi.org/10.1080/00461520.2014.965823
Jenkins, H. (2007, March 21). Transmedia Storytelling 101. Henry Jenkins's blog. http://henryjenkins.org/blog/2007/03/transmedia_storytelling_101.html
Genette, G. (1980). Narrative Discourse: An Essay in Method (J. E. Lewin, Trans.). Cornell University Press.