Artefact is a media platform dedicated to the value and culture of 20th-century antique design, featuring a smart valuation tool. It serves as a digital catalog devoted to the world of designer antique furniture and interior objects. Here, we do not sell—we tell stories, showcase, and appraise.

For Content:

1. Auction Databases and Aggregators.

• Invaluable The largest database of auction results, featuring filters by category, auction house, price, and date. It allows users to download data on sold lots, including realized price, estimate, description, condition, and photographs. For media outlets, this is the foundation for «Auction Digest» and «Trend Report» sections—allowing for the analysis of price dynamics across specific designers and eras.

• Wright Auctions (wright20.com) A specialized auction house in Chicago focused specifically on 20th-century design and mid-century modern furniture. The database includes designer names, materials, year of creation, condition, and hammer price. This is a key resource for articles covering Scandinavian Modern, Bauhaus, and Italian design.

2. Primary Sources (Original Content)

• Subscriber Stories The «Subscriber Case Study» and «Object History» features are unique content that no competitor offers. Users submit photos, the story of how they found an item, and the results of an AI-camera valuation.

• Expert Interviews Designers, restorers, auction experts, and collectors—these are the guests on our podcast and the subjects of our long-form articles.

For AI Cameras:

Model Recognition Training

• Auction Archives with Photographs and Metadata Wright Auctions provides lot images along with attribution: designer, material, year, country, and condition. Invaluable and LiveAuctioneers also contain photos with full descriptions. This forms the basis for training the model: «this is the photo—this is the attribution.»

• Museum and Gallery Catalogs Open collections (such as MoMA, Vitra Design Museum, Musée des Arts Décoratifs) are a source of benchmark images with precise item attribution. This is the basis for recognizing form and style.

• Condition and Rarity Assessment Wright includes a condition report in the lot’s metadata. This allows the model to adjust pricing based on condition: the same chair in perfect condition versus one with a crack—different price ranges.

1. Human Loop — Human ProcessingWho: editors, invited experts (designers, conservators, auction appraisers), and the community in the Telegram group.

2. Machine Loop — Automated ProcessingWhat: The «Eye» AI camera, cost calculation algorithm, and comparables database.

1. AI Camera «The Aprisal» (Recognition and Valuation)

   Can Do:

   · Provide a price range (± margin of error), not a single figure. · Indicate a confidence level (high / medium / low). · Suggest probable attribution—style, era, designer—with a «likely» qualifier. · Reference comparable sold lots as justification.

   Cannot Do:

   · Assert authenticity («this is the original»)—only state «features are consistent with.» · Publish a price without specifying a range and an effective date. · Value items with obvious signs of restoration without marking them «requires expert assessment.»Mandatory Human Review:· Valuations above a threshold (e.g., €10,000)—the editorial team verifies before publishing the case study. · Contested items (replicas, disputed attribution)—consultation with an expert is required.

2. Comparables Database (Finding Comparable Sales)Possible:· Automatically pull auction results from the past 24 months. · Filter out lots lacking photos or complete metadata.Not Possible:· Using data without specifying the source (auction house, date, hammer price). · Mixing hammer price and price with buyer’s premium without labeling it.Mandatory Verification:· Discrepancies between platforms exceeding 30% require manual review by the editorial team.

3. Auto-posting and Categorization (Channel)You can:· Automatically generate post drafts using a template (headline + 2–3 sentences + photo + price). · Publish a «Price of the Day» on a schedule—after editor approval.You cannot:· Publish without human approval. · Automatically generate «Item Stories» or long-form content.Mandatory Review:· Every post with a price. · Monthly PDF catalog—full manual layout and review.

4. Telegram Group Moderation (Bot)Possible Actions:· Welcome newcomers and display the rules. · React to keywords («rate, ” „sell“) by suggesting a relevant section. · Filter spam and advertisements using stop words.Prohibited Actions:· Deleting participant messages without cause (only according to established rules). · Automatically assigning the „Expert“ status—this must be done manually.Mandatory Review:· Disputed cases—handled by the lead moderator. · Participant complaints—reviewed manually.

5. Feedback and Model RetrainingYou can:· Collect verified case studies (confirmed by an expert) for dataset retraining. · Flag cases where the AI made an error—marking them as high priority for analysis.You cannot:· Retrain the model using unverified user data. · Use auction data without checking the reuse license.Mandatory review:· Every dataset for retraining requires editorial and expert sign-off. · Legal vetting of sources is required before every model update.