1  #+title: Amazon Author-Velocity Extension — Idea and Roadmap
  2  #+author: Design note
  3  #+date: [2026-04-17 Fri]
  4  #+startup: indent
  5  #+options: ^:{} toc:nil num:nil
  6  #+description: Browser extension that surfaces an author's publication rate on Amazon book pages as a trust signal against AI-authored fraud, with a staged feature roadmap.
  7  
  8  * Motivation
  9  The /Blake Whiting/ case (see [[file:blake-whiting-trust-brief.org][blake-whiting-trust-brief.org]]) shows that most
 10  trust signals on Amazon — bio, reviews, cover quality, prose voice — can be
 11  fabricated. One signal resists fabrication: /publication velocity/. A single
 12  human cannot research, write, edit, and publish ten non-fiction books a week. If
 13  a buyer can see that number at the point of purchase, they can judge for
 14  themselves whether they are likely looking at a bot.
 15  
 16  The extension turns that latent signal into a visible one.
 17  
 18  * The Core Signal
 19  - Publication velocity :: number of books published by this author over the last
 20    N months, shown alongside a human baseline.
 21  - Design intent :: inform, never accuse. The widget displays a number and
 22    context; the user decides what it means.
 23  - Why this signal first :: it is objective, cheap to compute, hard to fake
 24    without spawning many identities, and easy to explain in one sentence.
 25  
 26  * MVP Scope (v0.1)
 27  The minimum shippable product. Answers one question: /how many books has this
 28  author published in the last twelve months?/
 29  
 30  - Trigger :: content script detects an Amazon book product page
 31    (=amazon.{com,co.uk,fr,...}/dp/=).
 32  - Extraction :: read author name from the DOM (one well-known selector, fallback
 33    to byline parsing).
 34  - Lookup :: query Google Books API with =inauthor:"Name"= and filter to the last
 35    12 months.
 36  - Cache :: store per-author result in =chrome.storage.local= with a 7-day TTL.
 37  - Display :: small info card injected near the author byline — count, "typical
 38    human non-fiction author publishes ≤3/year," link to the full list.
 39  - Manifest :: V3, minimal host permissions (Amazon domains + =googleapis.com=).
 40  
 41  * Roadmap — Features for Later Releases
 42  Each release adds one axis. Ship v0.1 first, use it for a week, then reassess
 43  priority.
 44  
 45  ** v0.2 — Publication timeline and bursts
 46  - Full year-by-year publication chart rendered in the info card.
 47  - Burst detector :: flag when N books appear within a short window (e.g. ≥5
 48    books in 30 days).
 49  - Genre-aware baselines :: separate thresholds for romance, pulp, academic,
 50    children's — a prolific romance author is not a red flag.
 51  - Settings page :: let the user adjust the "suspicious" threshold.
 52  
 53  ** v0.3 — Identity-verification signals
 54  - ORCID lookup :: for books tagged as academic, check whether the author has an
 55    ORCID record.
 56  - Wikipedia / Wikidata presence :: binary signal, surfaced as "known / not
 57    known."
 58  - Personal site or publisher page :: checked via a curated whitelist, not open
 59    web crawl.
 60  - Topic-spread entropy :: measure how unrelated an author's subjects are (e.g.
 61    Byzantine archaeology + quantum computing + knitting).
 62  
 63  ** v0.4 — Composite trust score
 64  - Weighted score combining velocity, identity signals, and topic spread.
 65  - Transparency panel :: every signal shown with its contribution, no black box.
 66  - User-tunable weights for the skeptical vs. generous reader.
 67  - A/B comparison between two authors on the same topic.
 68  
 69  ** v0.5 — Community layer
 70  - User reports of suspected bots (local-first, opt-in to share).
 71  - Shared cache of confirmed/disputed authors hosted on a lightweight backend.
 72  - Moderation workflow to prevent review-bombing of real authors.
 73  - Export of the public dataset for researchers.
 74  
 75  ** v0.6 — Cross-platform coverage
 76  - Goodreads, Barnes & Noble, Apple Books, Kobo adapters.
 77  - Firefox build (Manifest V3 compat).
 78  - Safari extension (separate Xcode project).
 79  
 80  ** v1.0 — Publisher and provenance integration
 81  - Publisher self-serve :: authenticated authors can claim their page and show a
 82    verified badge.
 83  - [[https://c2pa.org/][C2PA]] provenance reading :: surface cryptographic provenance when the manifest
 84    is embedded in the book file.
 85  - Integration with [[https://orcid.org/][ORCID]] as an optional identity proof for academic authors.
 86  
 87  ** Long-term (speculative)
 88  - Opt-in aggregate telemetry feeding a public dashboard of suspicious-author
 89    statistics.
 90  - API for libraries and retailers to query the same trust signals.
 91  - Browser-level "trust mode" that blurs AI-suspected listings until the user
 92    opts in.
 93  
 94  * Technical Risks and Open Questions
 95  - Google Books rate limits :: 1000 req/day free tier may not be enough at scale
 96    — plan for per-user API keys or a proxy with shared cache.
 97  - Author disambiguation :: common names collapse multiple real authors into one
 98    count, inflating velocity — need heuristics (publisher overlap, topic
 99    coherence) before trusting the number.
100  - Amazon DOM drift :: selectors will break; content script needs a self-test
101    that fails quiet and logs to a local error buffer.
102  - Legal framing :: never label an author "fake" or "AI" without proof — stick to
103    neutral language ("published N books in the last 12 months") to avoid
104    defamation exposure.
105  - Privacy :: no user data leaves the browser in MVP; community features must be
106    explicit opt-in with clear data-handling copy.
107  
108  * Decision Points Before Starting
109  - Confirm MVP is one-signal only — do not pre-build the composite score
110    architecture.
111  - Pick data source :: Google Books is the default; OpenLibrary is the fallback
112    if rate limits bite.
113  - Pick deployment target :: Chrome Web Store first, Firefox second, Safari only
114    if demand appears.
115  - Name the project — /WriterAIScore/ is the working title; consider whether to
116    brand around the /score/ (judgmental) or the /signal/ (neutral).