# Subnautica2Maps design/map regression review — 2026-05-20

## Verdict

NO-GO for launch/operations.

The current production site is functional as a source-backed POI/search tool, but it does not meet the expected quality bar for a game map / game information site.

## Evidence checked

- Production home: https://subnautica2maps.com/
- Production map: https://subnautica2maps.com/map/
- Design handoff: `/root/.hermes/kanban/boards/site-factory/workspaces/t_f6e7a59a/reports/site-pipeline/site-rerun-subnautica2maps-20260520-clean/subnautica2maps/design/HANDOFF.md`
- Stitch index: `/root/.hermes/kanban/boards/site-factory/workspaces/t_f6e7a59a/reports/site-pipeline/site-rerun-subnautica2maps-20260520-clean/subnautica2maps/design/stitch/screen-index.json`
- Frontend report: `/root/.hermes/reports/site-pipeline/site-rerun-subnautica2maps-20260520-clean/subnautica2maps/05a-public-reference-frontend.md`
- Browser visual screenshots:
  - Home: `/root/.hermes/cache/screenshots/browser_screenshot_ced96700e0294d9aa85a63344b6be595.png`
  - Map desktop: `/root/.hermes/cache/screenshots/browser_screenshot_94d9317446844bf9b04c07e533ac6c35.png`
  - Map mobile 390px: `/root/.hermes/cache/screenshots/browser_screenshot_c7147ff5afa14e9fa6fbe197844e2b1b.png`

## 1. Was every page truly designed through Stitch?

No.

The design package is explicitly `local-html-fallback`, not Stitch-generated.

Evidence:

- `HANDOFF.md` line 3: `local-html-fallback`.
- `HANDOFF.md` lines 22-25: Project ID is `local-html-fallback`; Stitch CLI failed with `can't resolve reference #/$defs/ScreenInstance from id #`.
- `verification.json` lines 6-8: `stitch_attempted: true`, but CLI failed.
- `screen-index.json` has 20 screen entries, but `mode` is `local-html-fallback` and `stitch_status` is `attempted_but_cli_failed`.

So the honest answer: pages had local HTML/screens/state coverage, but not actual Stitch project/screen/code output. This should not have passed as equivalent to “真正通过 Stitch 来设计” for a production-quality game site.

## 2. Asset / game atmosphere gap

Production home has a coherent dark sci-fi/HUD style, but it lacks a game information-site feel and lacks Subnautica-specific visual anchors.

Observed gaps:

- No real game screenshots.
- No YouTube guide thumbnails or embedded video cards.
- No biome cards with visual thumbnails.
- No game resource icons.
- No game-like map preview; home hero shows a decorative grid rather than usable map value.
- No news/guide/article module typical of a game information site.
- The visual language reads as a generic sci-fi tactical console, not a clearly game-world-aware Subnautica map/guide site.

Current implementation is intentionally asset-safe, but the pipeline overcorrected: avoiding copied official assets became “no useful game media inventory at all.”

## 3. Map product gap

The current `/map/` is better than a static SEO page, but still closer to “POI coordinate scatterplot + list/detail panel” than a competitive game interactive map.

Observed gaps:

- No real/abstract biome base map, terrain, cave boundaries, depth zones, landmark regions, or semantic map layer.
- `browser_console` inspection: `.map-canvas` has grid CSS background only; no image tiles, iframe, canvas/WebGL, or semantic SVG base layer.
- `browser_console` inspection: no wheel/pointer/mouse handlers on `.map-canvas`; the “Pan/zoom-style” label is not backed by real pan/zoom implementation.
- No standard map controls: zoom +/- buttons, reset, fullscreen, layer panel, scale, coordinate cursor, copy coordinate, share view.
- Marker design uses colored dots only; no resource/databox/biome icons or legend.
- Route calculation is textual; no current-position marker or route line on the map.
- Mobile is readable, but map is not the primary first-screen object; current structure is a long stacked page, not a mobile map app with bottom sheet/filter drawer.

Competitor baseline from web search:

- IGN and Game8 already have Subnautica 2 interactive maps with locations/resources/supply crates/black boxes and filter systems.
- MapGenie/SubnauticaMap style competitors provide full map UX: filters, progress tracker, notes, biome/resource layers, map-first layout.

## 4. Root cause

This is not a single-agent mistake. The workflow allowed three quality downgrades to pass:

1. **Stitch failure accepted too easily**
   - Local fallback was honestly disclosed, but DESIGN_GO treated it as acceptable without requiring user review or second design attempt.
   - For game/map sites, local fallback should be conditional at best, not equivalent to Stitch unless it passes visual competitive review.

2. **Asset policy was interpreted as “no official/third-party assets,” not “build a licensed/public/fair-use asset inventory.”**
   - The pipeline did not force a media inventory from public channels, YouTube guides, community sources, screenshots policy, thumbnails, or self-made derived visual substitutes.
   - Result: legally safer but visually empty.

3. **Competitive minimum was checked as boxes, not as real user experience.**
   - “search/filter/detail/route/progress/mobile” existed, so gates passed.
   - But the map lacked actual pan/zoom, semantic layers, map controls, icons, legend, route visualization, and map-first mobile UX.
   - QA/PM gates were still queued, meaning the earlier design/frontend gates did not block hard enough.

## 5. Immediate remediation plan

Block launch/ops and run a focused repair track before PM/QA/launch:

### A. Design repair — `P0`

Owner: moying + moce review.

Required output:

- Real Stitch attempt must be retried; if Stitch still fails, mark as local fallback and require explicit PM/user visual review before frontend.
- Redesign home, `/map`, resource page, guide page, mobile map.
- Add content/media inventory:
  - public YouTube guide cards with thumbnails, title, source, timestamps/chapters, lazy embed plan;
  - public/allowed game screenshot policy or self-made non-infringing underwater/biome visual substitutes;
  - resource/databox/biome icon set;
  - map visual system with biome/depth/landmark layers.
- Produce competitor delta matrix against IGN, Game8, MapGenie, SubnauticaMap.

### B. Map UX repair — `P0`

Owner: mojie.

Required output:

- Real pan/zoom state in code, not a label.
- Map-first layout for `/map`: left filter drawer, central map, right/bottom details.
- Semantic map base: at least self-made SVG biome/depth/region layer with labels.
- Layer panel + legend + marker icons.
- Route visualization: current position, target, line, distance/depth delta.
- Mobile map bottom sheet + filter drawer; map is first-screen primary object.

### C. Asset/media track — `P0`

Owner: moshu/modun.

Required output:

- Public-source media inventory with provenance/license/risk policy.
- YouTube guide/video data inventory.
- Allowed visual fallback plan for each risky asset.
- No hotlinking unknown assets; use stored thumbnails only if allowed or generated/derived safe substitutes.

### D. Gate changes — `P0`

Update skills/gates so future runs cannot pass with this class of issue:

- Stitch fallback requires explicit `fallback_visual_review` and competitor screenshot comparison.
- Game/map sites require asset inventory before design.
- Map pages require executable pan/zoom/controls/layers, verified by browser console/code inspection.
- PM/QA cannot proceed if `/map` is a scatterplot grid without semantic map layer.

## 6. Current launch decision

Block `07C Product Acceptance`, `08 QA`, `09 Launch`, and `10 Review` until repair track is complete.