-
Synthetic sample: both stalls share one inventory; each packed title consumes an available copy.
-
Synthetic repair: the packed left crate stays locked while unavailable Kindred is replaced with Beloved on the right.
-
Earlier synthetic-mode evidence screen (October 17). Current live verification is shown in the project story.
Crate helps an independent bookshop pack for two different cultural events from one stockroom. The same single copy cannot leave in both crates. Shop reserves, editorial choices and an already-packed crate must survive last-minute changes.
Qloo provides live connections from artist and film interests to books in the available inventory. Crate turns that ordered evidence into feasible packing lists. It makes the cultural recommendation answer a practical question: which books can we actually take to each event?
Try the live app
Watch the live Qloo motion film, then try the same packing and repair flow below.
- Open Crate and choose Use live Qloo.
- Select Pack the stalls. The twelve starter books and four programme interests already have reviewed work-level Qloo identities.
- Open See the evidence to inspect both returned orders, coverage and retrieval provenance.
- Lock Soul & City, set The Dispossessed stock to zero in Manage stock, and repair. Future Frequencies replaces it with Kindred while the locked crate stays unchanged.
- Change a film interest, find and confirm its real Qloo match, then repair. Crate fetches new evidence for that programme and preserves protected choices.
- Export source-labeled manifests or print packing lists and display cards. Imported stock needs its own reviewed Qloo matches.
What is Qloo-powered?
The server calls typed entity search and GET /v2/insights, restricting book results to the supplied inventory UUIDs and using confirmed artist/film interests as the signal. Within-query order guides allocation; raw affinity scores are not added across requests or called sales probabilities. Missing candidates stay unknown, and failed live calls never become synthetic successes.
An exact bounded allocator respects shared quantities, shop reserves, table capacity, pins, vetoes and locked crates. Stock repairs minimize changed placements before improving rank fit. Programme changes trigger fresh cultural evidence and a new allocation under the protected constraints.
Three public MCP tools expose live search, ranking and planning to an external assistant at /mcp. Direct HTTP discovery and live planning passed on the Cloudflare deployment. An earlier installed-client call also fetched real Qloo evidence; that client used the prior deployment. This application does not call a language model itself or change external stock.
Verified live on October 22, 2026
- Genuine hosted searches resolved all twelve book/author pairs and the programme interests.
- Both initial cultural profiles returned 12/12 inventory titles.
- Changing the programme changed actual Qloo order and packed choices.
- The direct Cloudflare MCP HTTP integration passed live planning, with eight stock-safe placements in the initial plan.
- The public browser flow verified live packing, locked-crate stock-out repair, stale-export protection and a fresh query after a confirmed film change.
- All 95 automated checks pass, including 220 independent brute-force comparisons, CSV constraints, MCP contracts, UTF-8 exports, video byte ranges and Worker request limits. TypeScript, production build and deployment dry-run pass.
- The Cloudflare browser flow passed live packing, stale-export/print protection and locked-crate repair. Actual CSV/JSON link contents were saved and independently parsed; the JSON matched its visible snapshot byte for byte. The real plan produced inspected two-page A4 and Letter PDFs. Another 19 fixture pages passed content and bounds checks in a paged renderer.
The screenshot is the real public app after a programme change. Read the measured verification record and reproducible smoke check.
Free bookseller trial
Open the free trial guide and workbook. No payment or account is required. It records the usual manual method versus live Crate, including CSV preparation and Qloo matching, packing, protected-crate repair, failed attempts, editorial acceptance and negative feedback. Optional local save and JSON export do not send reports automatically. Practice and participant-reported observations stay explicitly unverified.
The trial protocol was defined before collecting results. Three individual invitations have been sent to event booksellers. No completed customer trials or measured benefits are available yet. This is recruitment and measurement preparation, not evidence of adoption, improved relevance or time savings.
Why this workflow is distinctive
Book recommendations and inventory filtering already exist. Crate combines cultural ranking with joint allocation of scarce physical stock across simultaneous stalls, then repairs shortages while protecting editorial choices and crates already packed. Its contribution is this operational combination, not a new optimization algorithm. Event-specific bookselling is real, but the frequency and value of this exact two-stall workflow still need bookseller validation.
Novelty, competing approaches and judging fit
Built and learned
React, TypeScript and Vinext provide the interface, deployed directly to the developer's Cloudflare Workers account. The key stays in a Worker secret. A native Cloudflare limiter guards credential-consuming requests at approximately 30 per IP per minute per location; this is not a global upstream quota. No external sign-in or connector gateway is required. Live testing exposed a Workers-specific redirect option that failed before network I/O; reproducing it in the Workers runtime led to a fix that still rejects redirects without forwarding credentials.
The two-stall constraint matters: an attractive recommendation is unusable if its only copy is already packed elsewhere. The oldest-stock comparison uses the same eligibility and stock constraints so the cultural allocation can be inspected alongside a simple baseline.
Motion demo and honest limits
Watch the narrated live Qloo motion film on Cloudflare. Real programme inputs and captured live evidence lead into a continuous 3D packing and repair sequence: The Dispossessed leaves Future Frequencies, Kindred fills its one vacancy, and the other seven placements stay. The moving books are explanatory drawings; the evidence is recorded from the working app. Rendered MP4, captions and editable source.
Sample events, stock quantities, prices and shelf ages are fictional. Cultural interests are audience hypotheses, not measured attendee preferences. Work-level identities do not certify ISBN editions. Bookseller adoption, reader acceptance, time savings and sales uplift have not been measured. CSV/JSON contents and paged A4/Letter layouts are tested. Completed native browser saving, Chrome Save as PDF and physical printing remain unverified.
Developed by Ajnas N B
Log in or sign up for Devpost to join the conversation.