# Create and recover a community game candidate

Start with the [beginner creator guide](index.html), the [single-picture walkthrough](image-campaign-guide.md), or the [batch campaign walkthrough](batch-image-campaign-guide.md). The picture campaign creator accepts one or many images and videos, generates bounded levels with enemies and collision obstacles, prepares exact posters and optional victory stories, and exports or installs reviewed `.rlpack` editions. The phase acceptance records preserve detailed evidence; the [combined acceptance record](combined-release-acceptance.md) and [delivery register](delivery-plan.md) identify the remaining release and production gates. The historical manual trials below retain their original scope and identities.

The [current execution plan](delivery-plan-2026-10-15.md) records the desktop-first verification
and GitHub Pages decision. The [combined creator feature acceptance record](combined-release-acceptance.md)
retains the consolidated implementation history; individual phase records remain supporting
evidence and do not imply separate releases or fresh acceptance of current main.

Maintainers can rerun the creator, media, installed-play, community-client and self-hosted service
cohort with `npm run test:creator-feature`. The command discovers the owned test families, requires
the cross-boundary acceptance files and fails if either the browser/runtime cohort or service suite
fails. Install the locked root, authoring-fixture and service dependencies first:

```sh
npm ci --ignore-scripts
npm ci --prefix authoring/fpv-worlds --ignore-scripts
npm ci --prefix services/community --ignore-scripts
npm run test:creator-feature
```

The [video campaign review guide](video-campaign-guide.md) documents the mixed-media review contract, exact pairing corrections, poster choices, playback ranges and optional post-win story behavior. The main creator flow persists those exact dependencies before approval; the [Phase 3 acceptance record](phase3-acceptance.md) preserves the focused implementation evidence.

The account, resumable upload, exact validation, automatic catalog publication, report and
unlisting service is documented in the [Phase 4 acceptance record](phase4-acceptance.md) and
[`services/community/README.md`](../../services/community/README.md).
The [Phase 5 community-store acceptance record](phase5-acceptance.md) documents the integrated
client candidate, exact immutable install/update/offline behavior, reference-aware recovery
removal, and the remaining hosted-production gates.

The [optional media editor guide](media-editor-guide.md) documents the bounded Phase 8 slice:
playback ranges, evidence-backed decoded-frame stepping, and verified physical trim/conversion for
one silent browser-decodable MP4/WebM video track or one MP4 with exactly one AVC/H.264 video track
and one AAC audio track. Output is AVC/H.264 MP4 and retains verified AAC only on that bounded audio
path. [Its acceptance record](phase8-acceptance.md) preserves the explicit limits: additional
tracks and other audio codec combinations are rejected, while Firefox, Safari, Balanced portrait
and physical-mobile qualification remain open.

This is a tested local authoring guide for Reveal Line. It covers the automatic creator, a small
manual mission project, and a separate editable presentation collection. The repository includes a
community catalog client and self-hosted service, but no public production community deployment or
finished DroneAid edition is claimed here.

Use [Content Studio](../../game/studio/index.html), [Asset Studio](../asset-studio/index.html), and the [copyable community prompts](../prompts/community-creation.md). Start from the [example project](example-project.json) when reproducing this guide. It is a small greybox with Solo/Versus support, not a finished campaign.

The browser trials used exact source checkpoints `eab879c`, `a840a5e6`, `86618bfc` and `499d716e`. The later integrated candidate `dceafec3` contains the crop and geometry-review safeguards; its native Studio checks reconfirmed both. Source verification and public release acceptance remain separate. Check the [execution register](../../docs/cross-mode-execution.md) for delivered versions; do not infer publication from this guide.

## What each tool produces

| Tool or file                     | Purpose                                                                                                | What it does not do                                                            |
| -------------------------------- | ------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------ |
| Picture campaign creator         | Generate reviewed image/video campaigns, `.rlpack` files, source backups and installed Custom play     | Human balance review or public publication                                     |
| Content Studio                   | Maps, missions, campaign/pack ordering, rules and local checkpoints                                    | Publish arbitrary projects or certify playability                              |
| Content Studio **Export backup** | The current `ContentProjectV1` JSON snapshot                                                           | Include all local checkpoint history or the complete presentation media bundle |
| Asset Studio                     | Asset replacement, pixel edits, geometry, provenance, immutable presentation history                   | Change mission mechanics or player saves                                       |
| Asset Studio `.rltheme`          | Validated presentation records and retained original/derivative bytes                                  | Automatically make the public game use the theme                               |
| Community campaigns              | Browse/install immutable catalog editions and, on a deployed service, publish approved `.rlpack` files | Make the repository's undeployed service publicly available                    |
| Content CLI                      | Shared compiler output, mode/preset resolution and authored pacing diagnostics                         | Approve art, difficulty, offline readiness or public release                   |

Use a desktop keyboard/pointer and file chooser for this authoring workflow. Install a reviewed
`.rlpack` through the picture creator or use the compatible-content browser backed by a deployed
community service. Content Studio and Asset Studio imports remain authoring operations.

## 1. Create a small playable project

For an exact repeatable starting point, use **Inspect import** on `example-project.json`, inspect the source, then **Apply inspected source**. To practice creating it yourself, follow the steps below. Keep an export of any existing project before using the same ID. Copy the example and choose a new project ID for a different community; IDs are machine identities, while names are player-facing labels.

1. Open **Content Studio** from Workshop. Choose a unique local project ID, for example `community-guide-qa`, and choose **New project**.
2. Inspect the staged starter source, then choose **Apply inspected source**. Confirm the URL/project selector shows your project and the save status names a checkpoint.
3. Open **Organize missions, campaigns & packs**. Choose Campaign → Rename, select Horizon School, enter **Community Pilot**, then **Validate & apply structure**.
4. Choose **Save checkpoint / retry**. Confirm the campaign summary changes. Record the resulting checkpoint number; do not assume another workspace uses the same numbers.
5. Choose **Play exact Solo preview**, then **Start mission**. The starter's island provides a safe first exercise: travel down from the starting boundary to the reclaimed island. Our native trial closed a legal line at 0.6% coverage and 140 practice score with all three lives retained.
6. Press Escape to pause; Resume stays explicit. **Close preview** returns focus to the Play opener and retains the project. Reload and confirm the saved campaign remains.

Explicitly enabled optional scout/sentry editions can use this same exact Solo
practice preview; follow the [optional patrol authoring guide](../../docs/combat-pressure-authoring.md).
The preview awards no campaign progress and keeps authored rules. If a frame
fails after startup, a persistent message stops that child’s play and input;
Close/Return, correct the draft if necessary, then choose **Play exact Solo preview** again.
Team combat and paired-race Studio launches remain unsupported. This source
workflow does not establish production approval or public availability.

This starter is a greybox. Its disconnected-foundation and missing-background warnings are meaningful authoring diagnostics, not proof that it is a finished community mission. Add one deliberate route decision at a time; record the lesson, threat, counterplay, mastery opportunity and playtest findings before expanding the campaign.

## 2. Recover an earlier checkpoint safely

1. Enter the desired checkpoint and choose **Inspect saved**. The currently accepted workbench should remain unchanged while inspected source appears in the editor.
2. Choose **Apply inspected source** only when the inspected project is the intended one.
3. Save. Restoration creates a newer checkpoint; it does not erase the intermediate history. Our trial restored checkpoint 1 as checkpoint 3, then restored checkpoint 2 as checkpoint 4.
4. Inspect a newer historical checkpoint again to confirm it remains reachable.

Checkpoint numbers are local history. They are not the immutable project revision or a public game version.

## 3. Export and restore the mission project

1. Choose **Export backup** and verify the actual downloaded JSON exists. The UI says download requested because the browser owns the final download.
2. In another workspace, choose **Inspect import** and select that JSON. Inspection must not immediately replace the accepted draft.
3. Choose **Apply inspected source**, then **Save checkpoint / retry**. Reload to verify restoration.
4. Export the restored project again and compare the bytes. The guide fixture produced identical 1,773-byte files, SHA-256 `4a5290829dff769cc22bb3d383625cf8812fd8127dab96f8ac5e1dab870198e1`.
5. Test rejection with a separate malformed copy, never by altering your only backup. Our invalid-identity fixture left Apply disabled and preserved the accepted project.

A clean workspace begins new checkpoint history from this imported snapshot. Keep the source originals and theme bundle separately; the project JSON is not an all-media backup.

## 4. Create a presentation candidate

1. Open **Asset Studio** and wait for the release collection to load. Search for a stable slot ID; the guide trial used `hud.life`.
2. Read its exact dimensions, transparency, sampling, usage and geometry contract. Slot requirements belong to that element; an icon, actor body, illustration and font have different constraints.
3. For a small raster, open **Pixel editor → Edit current raster**. Arrow keys position the cursor; Space applies the tool. Exercise Undo/Redo before proceeding. Our test added one corner pixel solely as a QA fixture; it is not proposed production art.
4. Choose **Prepare edited sprite**. Alternatively use **Choose replacement file** to upload an original, set its crop, then **Prepare derivative**.
5. After changing any crop field, prepare again. Invalid crops and valid but unprepared changes must not stage old pixels. If preparation fails, correct the fields and explicitly prepare before continuing.
6. Enter actual Creator, Source, License/rights and the effective generation/edit prompt. Keep community branding provenance separate from gameplay mechanics. Do not invent authorship or rights from a generated image.
7. Choose **Validate and stage replacement**, inspect the comparison, then **Save local revision**. Reload and check the selected revision.
8. Metadata edits retain image bytes but create a new produced candidate with empty review evidence. A changed pivot, rotor layout or nine-slice needs its own review. The approved original stays intact in history.

For generated alternatives, use the selected slot's **AI prompts** and preserve its effective contract: dimensions, alpha, palette, silhouettes, anchors, dependencies, output names and actual-size checks. A candidate is technically prepared before it can be visually reviewed; copying a prompt or passing a decoder never grants review approval.

## 5. Transfer a theme and restore its original

1. Export the actual `.rltheme` file. Keep it with your project JSON and source originals.
2. In a clean Studio workspace, choose **Import collection**, select the bundle, wait for complete validation, then save.
3. Re-export and verify retained payload hashes. Our test retained all 128 payloads exactly. The complete bundle differed because imported records were correctly namespaced and their history changed; whole-file equality is not the right test for a presentation adoption.
4. Open **Asset revision history**, locate the exact approved original and choose **Bind this revision**. This stages another immutable theme revision.
5. Exercise Undo/Redo, save and reload. Both the edited candidate and approved original must remain in history.
6. A corrupted bundle must fail before adopting anything. A one-byte-corrupted payload produced a hash-mismatch error in the native trial and left the accepted saved collection unchanged.

Do not use **Record reviewed revision** merely to remove a warning. Review the actual affected screens, supported modes, native-size rendering, contrast, motion and accessibility; record the checks you actually performed.

## 6. Recover when two tabs disagree

1. If Save reports that another tab changed the draft, keep the unsaved tab open.
2. Choose **Export .rltheme** in that tab, verify the actual download, and retain it. Our exported manifest kept the stale tab's distinct unsaved description.
3. The Studio's **Reload saved** refuses to silently discard staged work. Once the export is safe, choose **Reset to saved**, then **Reload saved**.
4. Confirm the other tab's accepted revision is now loaded. Recover or combine the exported work through a separate deliberate import/review step.

This is conflict protection, not an automatic merge. The native trial proved that an older tab could not overwrite the accepted newer revision, and that its own work remained exportable.

## 7. Check the project through the shared CLI

From a checkout of the exact source being qualified, run the following against the supplied example. Substitute your downloaded project filename for your own project:

```sh
node scripts/compile-content-project.mjs authoring/community/example-project.json --check
node scripts/compile-content-project.mjs authoring/community/example-project.json --mission nearby-shore --mode solo --difficulty gentle
node scripts/compile-content-project.mjs authoring/community/example-project.json --mission nearby-shore --mode solo --difficulty standard
node scripts/compile-content-project.mjs authoring/community/example-project.json --mission nearby-shore --mode solo --difficulty expert
node scripts/compile-content-project.mjs authoring/community/example-project.json --mission nearby-shore --mode versus --difficulty gentle
node scripts/compile-content-project.mjs authoring/community/example-project.json --mission nearby-shore --mode versus --difficulty standard
node scripts/compile-content-project.mjs authoring/community/example-project.json --mission nearby-shore --mode versus --difficulty expert
node scripts/compile-content-project.mjs authoring/community/example-project.json --journey --pack opening --mode solo --difficulty standard
node scripts/compile-content-project.mjs authoring/community/example-project.json --pacing --pack opening --mode solo
```

All commands above passed for the guide fixture on Node 20.19.5. Gentle/Standard/Expert resolved to 5/3/2 lives; each preset retained the same simulation identity across Solo and Versus. The check deliberately reports `readyForRelease: false`, and pacing reports authored ratings rather than measured enjoyment or route quality.

The starter declares Solo and Versus only. Requesting `--mode team` fails with **Mission does not support this mode.** Author and qualify a compatible Team mission; do not relabel a Solo result as Team evidence. Studio's Team test export is a separate geometry/rules workflow and must retain its stated artwork limitations.

## 8. Create a Team campaign from pictures and videos

Open **Create a Team campaign** from the picture campaign creator. This path creates a separate
cooperative campaign; it never relabels a Solo/Versus level as Team content.

1. Enter a collection name, a lowercase campaign ID and a generation seed. Choose **Generate and
   verify**. The current registry creates Mirrored crossing and Relay pincer in deterministic seed
   order and replays all three difficulties with both supported teamwork presets. The review must
   report 12 exact completion routes.
2. Choose one PNG, JPEG or WebP reward picture for each level. Pick **Fit whole picture** or **Fill
   board** and review the description. Preparation removes source metadata and produces an exact
   1152 × 576 PNG; the private original is not added to the portable campaign.
3. Optionally choose one browser-decodable MP4 or WebM victory video per level. Blank playback
   times use the complete video. A range changes playback only: the complete inspected original
   remains in the package. Enter the credit and sharing terms that actually apply.
4. Choose **Prepare review**. Inspect every exact reward derivative, level binding, optional video
   range, package size, staging size and shared managed-media usage. Errors remain local; correct
   the affected row and prepare again.
5. Choose **Approve exact package** only after the visible review is correct. Any later gameplay,
   media, range, description or credit change invalidates approval. Choose **Install approved
   campaign** or **Download .rlteammedia**.
6. Open Team and select the installed edition under Custom missions. A legal win earns the exact
   reviewed picture. An optional video offers Play, Skip and Replay without blocking Next. Retry,
   unfinished recovery and clears remain scoped to the immutable package SHA-256.

Team editions share the same 256 MiB managed-media ledger as creator pictures, stories and audio.
An interrupted cross-database install leaves a pending byte claim; the next Team inventory or
install reconciles that claim against the exact installed editions before accepting more data.

## 9. Export and test a legacy Team mission

This additional native trial uses exact source `499d716e` and the bundled **Inspect Team Signal greybox** action. Inspect first, then **Apply inspected source** to accept the separate Signal partners project. The current project remains unchanged during inspection.

1. Select **Shared detour** and the intended Gentle, Standard or Expert preset. Read the terrain warnings and both spawn locations.
2. Choose **Export exact Team test mission** and retain the actual `shared-detour-<preset>-team-test.json` download. This is geometry/rules only; it does not carry the authored mission picture or grant Journey awards.
3. Open the linked **Team test player**. Expand **Play a created co-op pack**, then **Open a Team pack or artwork bundle** and choose the downloaded file.
4. Wait for **Imported Team picture ready** and the Shared detour selection. Verify the locked Challenge matches the exported preset, then choose **Start together**. Our three exports started with 4/2/1 reserves for Gentle/Standard/Expert respectively.
5. On Standard, steering Sunflower upward from its foundation created a legal 0.4% capture without losing reserves. Pause, cancel Retry with **Stay**, then deliberately retry. Stay preserves the paused attempt; confirmed Retry resets the same mission.
6. For recovery testing, use a disposable malformed copy. A width-zero pack was rejected visibly, retained Shared detour and left Start available. A subsequent valid import recovered normally.

All three actual downloaded files matched the shared compiler using the Studio's `campaignTheme: true` candidate option. These checks prove export/import and basic play, not a complete clear, fairness, custom Team artwork, physical controls or public/offline readiness. The Team mission preset is fixed by its export: choose another preset in Studio and export again rather than expecting the player to rewrite authored rules.

## 10. Brand the community without changing its mechanics

Write a short edition brief before replacing assets: community name and rights holder, authorized source links, palette, ornament rules, player roles, threat meanings, supported modes, language requirements and content framing. For DroneAid, keep the requested Support/Combat framing explicit; do not assume all community imagery or missions use combat. For cultural, retro and spend-management editions, explain hazards using their own visual language while retaining readable warning/counterplay cues.

Resolve each slot from the selected theme and collection. Replace actors, HUD, terrain, effects, objectives, reveal pictures and audio through their respective contracts. Keep player number/shape cues and warning contrast across every palette. Translate UI through real text; do not bake labels into images. A logo, palette change or alternate picture alone is not a complete game theme.

Begin with one reviewed Solo mission, separately qualify each advertised multiplayer path, and add
one compatible Team arena. Generated v3 editions can use the installed two-board Versus host only
after the exact package and equal-board evidence are revalidated. Installed Versus keeps the
verified difficulty and runtime seed, records edition-scoped progress and pictures, and continues
to the next exact mission; human fairness and balance review still remain campaign-specific. New
reveal pictures receive new revisions; Retry, saved flights and earned originals keep their
accepted pins. Never replace old records in place to make a new community look complete.

## 11. Prepare the release handoff

The preceding local workflows are tested. The following acceptance remains open for a finished community edition:

- Complete original/authorized community branding, artwork, actors, audio and exact presentation bindings; no required source recipe or placeholder is counted as finished.
- Qualify every advertised mode and difficulty with real routes, fair warning/counterplay, progress/Retry/Next and appropriate multiplayer objectives.
- Inspect shared components, fonts, global preferences and all affected screen states, including portrait/landscape, keyboard, touch and physical controllers separately.
- Verify the actual catalogue installation, all pinned offline dependencies, recovery/export/import and unchanged historical ownership.
- Update version, source documentation and prompts; run all six release gates plus applicable build/production/artifact checks on the exact committed source.
- Freeze and publish through the existing release coordinator; verify the actual deployed bytes and play the public routes before accepting delivery.

Selecting media in an authoring page changes only the local workspace. Installing a reviewed pack
adds its immutable edition to that browser. Publishing is a separate authenticated upload to a
deployed community service; the repository implementation alone does not create a public catalog
or complete production acceptance.