MiroResult — a complementary tool to visualize MiroFish outputs

[CyrilDEVIA]

Hi,

I've been using MiroFish for a few weeks and built a complementary tool that I just open-sourced: MiroResult.

The idea is straightforward: take MiroFish simulation ZIP exports and turn them into AI-generated visual HTML reports, with a two-pass critical analysis and a built-in chat to iterate on the result.

The tool works but there's definitely room to improve — which is why I opened it up. If it could be useful to the MiroFish community or if you see any interesting synergies, I'd welcome any feedback.

https://github.com/CyrilDEVIA/MiroResult

Best,

[666ghj]

Note

🤖 Automated maintainer response — Generated by the MiroFish triage agent and checked against main@4ebb2ed6 on 2026-07-21. Posted with maintainer authorization; a maintainer may revise it.

Thanks for sharing MiroResult. It is welcome as an independent community experiment, but it is not an official MiroFish component and has not been audited or compatibility-certified by the maintainers.

Current official MiroFish provides Markdown report download, but it does not expose a built-in “Export ZIP” workflow or a stable ZIP schema. Please document how users should produce the ZIP, its expected directory layout, supported MiroFish versions or commits, and how simulation data and API keys are handled.

Users should review the tool, its data handling, and compatibility independently. We will keep the Discussion open under Show and tell.

[CyrilDEVIA]

Thanks for the clarification.

I updated the README to make the project status and limitations clearer.

The documentation now explicitly states that:

• MiroResult is an independent community project
• it is not an official MiroFish component
• it has not been audited or compatibility-certified by the MiroFish maintainers
• MiroFish does not currently provide a stable built-in ZIP export workflow
• the ZIP archive must be prepared by the user
• compatibility may depend on the MiroFish output structure
• data handling and API key handling are documented

I also added sections about the expected ZIP structure, compatibility, known limitations, and data/privacy.

Thanks again for keeping the discussion open under Show and tell.

[DarkWorldCoder]

One regression case that may help with the ZIP/report contract: I checked the isolated scoring and summary functions at 9e44718. With this parsed input:

{
  "simulation_config": {},
  "state": {},
  "run_state": null,
  "env_status": null,
  "actions": [],
  "profiles": [],
  "raw_files": []
}

computeScore returns global: 100, with configuration: 100 and completion: 100; summarizeZip then describes a complete, well-structured simulation. Missing actions/profiles are excluded from the denominator, and the presence of a state object counts as completion. The parser accepts those two empty JSON objects by filename.

A small fixture containing only simulation_config.json and state.json, each {}, could guard against this. I'd keep file-presence checks separate from validated run status and show missing evidence explicitly before assigning a completion summary. Increasing agents or rounds would not repair an incomplete archive.

Scope: I ran only the two isolated pure functions after removing TypeScript annotations; I did not run the application or upload an archive. Disclosure: I maintain MarketMirror, another MiroFish/OASIS consumer. Codex assisted this review.

[CyrilDEVIA]

Thank you for this clear and reproducible report. I confirmed the issue in the current code and I’m preparing a targeted fix, including the regression test you suggested.

MiroResult is not currently used in production, but I want to correct this behavior before further use. Your review was very helpful, especially the distinction between file presence and validated simulation status.

Thanks again for taking the time to investigate this and document the scope of your test.

[DarkWorldCoder]

Thanks, Cyril. I rechecked the original regression against 23739bad.

The original simulation_config: {} / state: {} fixture now scores 0 in all four dimensions, receives an incomplete summary, and no longer recommends increasing agents or duration to repair missing evidence. I also checked that a state.status: "ready" marker does not count as completion, an unfinished-run fixture stays incomplete even with actions/profiles present, and a synthetic runner_status: "COMPLETED" positive control still receives the complete summary.

That addresses the empty-archive scoring and summary case I reported. Scope: I executed the pinned scoring/summary/recommendation functions after native Node TypeScript stripping. I did not import the application, exercise the ZIP parser/upload flow, connect a database, or test production deployment.

Appreciate the targeted fix and the regression coverage. MarketMirror maker; this verification and reply were assisted by Codex.

[CyrilDEVIA]

Hi Ayush,

Thank you for taking the time to thoroughly verify the fix and for your detailed feedback.

I really appreciate you reporting the issue and helping improve MiroResult. Your testing and clear explanations were very helpful.

Glad to hear the original regression is now resolved. Thanks again for your contribution!