FoCo Design Co. ↗

BUSINESS OS / ASSET STUDIO / REUSABLE OPERATOR SYSTEM

Reference. Create.
Review. Improve.

One system for comparing websites, Instagram posts, product photos, posters, signs, video and store concepts against the real references that inspired them.

Give another person or agent this index link. The instructions, source files, setup example, grading rules and regeneration process are all here, without private repository access.

Start on another machine

Download the complete portable kit ↓

Node.js 18 or later; no npm dependencies for this component. Unzip the kit and run these commands from its folder:

node build-before-after.mjs --manifest example/comparison.manifest.json --out example/compare.html
node serve.mjs example 8799
Open http://127.0.0.1:8799/compare.html

The setup example is deliberately synthetic. Replace it with the actual source and generated image files for your project. One served folder and one ledger per brand keep feedback separate. See the manifest fields →

When continuing an existing project, recover exact source/master bytes through its private asset rail. Keep rated recipes unchanged; use CREATIVE_ASSET_PATH_MAP or --asset-path-map to locate saved hashes on this machine. The helper checks recovered bytes and preserves the original provenancePath. Keep credentials, local paths and large masters offloaded. A drive move creates no new grade.

The public site hosts these instructions and source downloads. Run the included local server for disk saving; a static page opened by itself cannot append feedback to a file.

Review beside the pictures

Reference on the left, your version on the right, feedback alongside on desktop and directly beneath on a phone. All current pieces run down the page. Arrows, asset selection, type filters and an optional one-at-a-time view stay available.

Green · new, not rated Lavender · already rated

Use the 0–100 slider, No / Maybe / Good / Love shortcuts, comments, tags and Save. The calibrated score bands are 0–24 / 25–49 / 50–74 / 75–100; shortcuts use 12 / 37 / 62 / 87. Missing artwork is a comment-only gap.

Every save appends to ratings.jsonl in the served folder. GET /api/ratings reads latest feedback per exact version; POST /api/rating writes. A failed save keeps a browser draft and shows its status; retry Save when the server returns. A draft is never proof that disk saving succeeded.

Current round first. Keep the history.

A new round must contain actual new artwork. Set batch.id, a clear title/date and batchId on its pairs. The current batch stays visible; other pairs live in a collapsed Earlier rounds & open gaps section. Old deep links expand that section automatically.

Each asset gets a clear job title and reviewLabel: {revision, round, date, change}. These labels and batch membership do not change grade identities. A score updates the background without moving the card while someone types.

Good / Love keeps the highest-rated originals, including earlier favorites. Rating history runs best to worst and exposes individual saves. Never copy an old score onto new art. The version ID hashes brand, asset ID, output/download bytes, copy/caption, reference/crop and production recipe.

Preserve frames.json, versions/, comparison-receipt.json and the ledger. Before replacing any export, build once to freeze its original bytes. If a historical original is unavailable, show that honestly; never display newer pixels beneath an old grade.

Use feedback to finish the next round

  1. Read the actual version’s score and comments, real product/store photos, brand facts and accepted examples.
  2. Freeze the exact brief, input order and hashes, permitted changes and fixed facts in the asset’s recipe.
  3. Generate or edit with the image/video tool using those actual references. The comparison builder creates the review page; it does not generate artwork.
  4. Inspect the resulting image at its intended size. Correct visible misses before asking for ratings.
  5. Keep a new versioned master and exact submitted prompt, provider and available settings. Export small review/channel files and archive prior exports.
  6. Build a fresh batch, verify new IDs begin unrated and earlier favorites show original bytes, then walk desktop and phone controls.

Prepare one selected revision from checked sources and saved feedback:

node prepare-regeneration.mjs --manifest example/comparison.manifest.json --root . --ledger example/ratings.jsonl --asset campaign-poster --change "Specific feedback to apply" --out revision-brief.json

The helper checks source/output/master/prompt hashes and returns a brief. Complete the generation and export steps after it. Recipes may include prompt, promptSha256, sources: [{path,sha256}], master: {path,sha256}, brandInputs, feedbackBasis and fixedFacts. A changed brand input needs deliberate rebinding. Generation is repeatable as a method; the same prompt does not guarantee the same pixels.

Capture scoped lessons with the supporting exact-version IDs. A client’s ingredient, room geometry or wording does not automatically become a universal brand rule. Creative approval, factual clearance, final print proof, measured construction plans and publication each retain their own approval.

Capture real products. Audit every customer action.

Start with the brand’s existing product list and private photo intake. Staff choose the product, location, front/side/tub angle and actual container, then confirm its label beside a real capture example. Preserve unchanged originals and hashes; independent upload records keep simultaneous photos and later corrections.

Give each product its own enhancement job and execution receipt. Compare source beside result for count, silhouette, color, texture, markings and container. Inspect the actual website component and final print crop before expanding a batch; whole scoop edges and an actual container exit must survive both crops. Never recolor one generic serving across product identities. Register reviewed derivatives through the existing web/print pipeline. A competitor supplies framing; the client supplies product geometry. Upload alone does not publish, and a new result never inherits a rating. A partial batch cannot certify the whole menu.

Inventory separate signs and customer actions inside each reference: one photograph can contain pickup, order QR and service bell; catering and fundraising may need two different destinations. Distinguish rejected native drafts, missing art/photos, generated directions, operating prerequisites and print proof. Decode each QR from its actual final PDF, verify the destination, then walk the intended action. A page returning 200 does not prove a review button, checkout or inquiry delivery. Flavor browsing is not order-online parity.

Keep owner-only facts and archive-transfer needs in the existing client roadmap. Read the Asset Studio module, section 11, for the adopter contract. This extends the existing creative-review system; it creates no second rating backend or product database.

Source files & ownership

This is the existing Asset Studio operator component: owner-feature:asset-studio, capability:asset-studio-api, catalog D2/D3/D6 and the brand_and_creative work profile. It is not a separate SKU or an automatically installed admin app.

Module owner: client-factory/docs/features/asset-studio/CLIENT-CREATIVE-SYSTEM.md, section 11. Shared implementation: docs/design/guided-setup-flows/before-after/. Client adapters supply their own facts, media and manifest; improve the shared builder rather than copying its renderer into every project.

Readable sourceResponsibility
build-before-after.mjsBuild entry point for pairs or the existing screen Pass/Fix mode.
comparison-bench.mjsManifest validation, exact-version IDs, immutable archives and HTML composition.
comparison-bench.cssSide-by-side layout, new/reviewed colors, responsive controls and history gallery.
comparison-bench-ui.jsSlider, faces, comments, current round, navigation and ranked history.
rating-store.jsDisk API persistence, browser drafts, retry and concurrent edits.
prepare-regeneration.mjsRead saved feedback and checked source hashes into a revision brief.
build-system-index.pyRebuild this HTML index and portable kit from a pinned Factory source commit.
serve.mjsUnchanged pinned projection of Photography Services’ loopback SceneBench server.

Complete portable instructions · Pinned source provenance & hashes · Manifest example

Factory commit: f102c8669271902fe60d3189b0a54a566f64b07c
Projection generated from pinned source bytes. Server behavior remains owned by Photography Services; its donor pin is recorded in source.json. No private client imagery or feedback is included.