Add PITCH.md and WHITEPAPER.md — elevator pitch + external white paper

PITCH.md

@@ -0,0 +1,68 @@
+# OpenBookMap — Pitch
+
+Three tiers, shortest to longest. Use the one that fits the moment.
+
+---
+
+## 10 seconds (the hook)
+
+**OpenBookMap is OpenStreetMap for used and indie bookstores — a free,
+public map built from phone photos of shelves.**
+
+---
+
+## 30 seconds (the elevator)
+
+> Used and independent bookstores hold millions of books no one online
+> can search. OpenBookMap turns a phone photo of any shelf into part
+> of a free, public catalog of every independent bookstore — built on
+> OpenStreetMap, with OCR that runs on the contributor's own device.
+> No platform middlemen. The result is open civic infrastructure for
+> the offline book economy: discoverable by anyone, owned by no one.
+
+---
+
+## 60 seconds (the meeting opener)
+
+> You ever try to find a specific used book? You call three stores,
+> drive past five more, and hope. Used and independent bookstores
+> between them hold millions of books, often deeply curated — and
+> none of that inventory is online anywhere a search engine can see.
+> When a store closes, that curation disappears with it.
+>
+> OpenBookMap is a free, open-source project that lets anyone with
+> a phone photograph a bookshelf, have the spines read on-device by
+> OCR, confirm the titles, and add them to a public, searchable map
+> of every independent bookstore in the world. The map data comes
+> from OpenStreetMap; the inventory layer is what we contribute.
+>
+> It's deliberately tiny: a static site on GitHub Pages, no servers
+> we run, every layer replaceable. The whole thing fits under a $20
+> /month budget and is MIT-licensed. The point isn't to disrupt
+> bookstores — it's to make them findable.
+>
+> Live at **openbookmap.org**. Source at
+> **github.com/tautme/openbookmap**.
+
+---
+
+## Spinoffs for specific audiences
+
+**For book lovers:**
+*"Find any used book near you without making twenty phone calls."*
+
+**For bookstore owners:**
+*"Get your store and a snapshot of your shelves discoverable online,
+without giving up your data to a platform. Contributors do the work;
+you keep your sovereignty."*
+
+**For the OSS / civic-tech community:**
+*"OpenStreetMap for the offline book economy. Static site, on-device
+ML, free tiers everywhere, every layer replaceable. Looking for
+contributors and partner projects in the open-data ecosystem."*
+
+**For funders or sponsors:**
+*"Independent bookstores are closing at a steady rate; their
+inventories are cultural memory that vanishes with them. We're
+building durable, open digital infrastructure for that memory.
+$0 to start, $20/month ceiling, MIT-licensed, no rent extraction."*

WHITEPAPER.md

@@ -0,0 +1,320 @@
+# OpenBookMap
+
+**A public, open-source map of used and independent bookstores —
+with their shelves searchable from a phone photograph.**
+
+White paper · v0.1 · 2026
+
+---
+
+## Abstract
+
+Independent and used bookstores collectively hold inventories of
+cultural and material value, yet that inventory is invisible to anyone
+not standing in the shop. When a store closes, the curation goes with
+it. OpenBookMap is a free, open-source project that lets anyone with
+a phone contribute a shelf photo, runs OCR on the device to extract
+titles, and publishes the result as part of a public, searchable map
+of every independent bookstore in the world. The location layer comes
+from [OpenStreetMap](https://www.openstreetmap.org/); the inventory
+layer is contributor-built and openly licensed. The implementation is
+deliberately small: a static site, no servers we operate ourselves,
+every layer replaceable. The aim is durable, open civic infrastructure
+for the offline book economy.
+
+**Live:** [openbookmap.org](https://openbookmap.org) ·
+**Source:** [github.com/tautme/openbookmap](https://github.com/tautme/openbookmap) ·
+**License:** MIT (code), open data licensing (contributions; see §10).
+
+---
+
+## 1. The problem
+
+A used or independent bookstore's value comes from two things: its
+*selection* (which is the owner's accumulated taste) and its
+*serendipity* (which is the customer encountering a book they didn't
+know they wanted). Both are invisible from the internet.
+
+- A reader looking for a specific used book has no efficient way to
+  find it. Calling stores is the working solution in 2026.
+- A bookstore looking to advertise its current stock has no
+  affordable, neutral place to list it.
+- A researcher or librarian interested in the geography of book
+  retail has no open dataset describing what's actually on shelves
+  in independent shops.
+- When a store closes, decades of curation disappear without trace.
+
+Existing platforms address pieces of this poorly:
+
+| Platform | Why it falls short |
+|---|---|
+| AbeBooks | Amazon-owned; transactional; not a public dataset |
+| Biblio.com | Commercial marketplace; not open data |
+| Bookogs | Defunct (folded into Discogs and then shuttered) |
+| Goodreads / Storygraph | Reader-facing; no bookstore inventory layer |
+| OpenLibrary | Bibliographic database; not bookstore-facing |
+| OpenStreetMap | Has bookstore *locations* (`shop=books`) but no inventory |
+
+OpenBookMap fills the inventory gap, and does so with the same open
+principles that make OpenStreetMap usable for everyone.
+
+---
+
+## 2. What we're building
+
+A two-layer public catalog:
+
+1. **The location layer** — every independent and used bookstore in
+   the world. This is read directly from OpenStreetMap via the
+   Overpass API. We do not duplicate the data into our database; we
+   *reference* it by OSM id. Improvements to OSM benefit the project
+   automatically.
+
+2. **The inventory layer** — contributor-uploaded shelf photos with
+   OCR-extracted, human-confirmed titles. This is the data we own
+   and openly license. Each contribution is keyed to an OSM bookstore
+   id so the two layers stay aligned.
+
+The public-facing site lets anyone:
+
+- Browse a map of every bookstore in OSM, with category dots for
+  *used / indie / chain*.
+- Click a shop to see its photos and indexed titles.
+- Search across shop names and book titles globally.
+- (With an account) contribute photos and titles.
+
+---
+
+## 3. How it works
+
+A typical contribution path:
+
+1. A volunteer walks into a bookstore (preferably with the owner's
+   blessing — see §8) and takes phone photos of shelves.
+2. They open `openbookmap.org/contribute`, sign in, and drop the
+   photos onto the page.
+3. The browser does everything else **locally**:
+   - Compresses each photo to a display and thumbnail derivative.
+   - Runs OCR on the original — currently PaddleOCR (more accurate)
+     or Tesseract.js (lighter, falls back automatically). A device-
+     capability auto-pick is in review.
+   - Returns candidate titles for the contributor to confirm, edit,
+     or discard.
+4. Confirmed titles, photos, and metadata are uploaded to a
+   Postgres database with Row-Level Security policies enforcing
+   "you can only edit / delete your own contributions."
+5. Within minutes, the shop's page on the public map reflects the
+   new contribution.
+
+The key architectural choice: **OCR runs on the contributor's
+device, not on a server.** This keeps operating costs at near-zero,
+sidesteps the privacy and licensing concerns of sending images to a
+cloud OCR API, and aligns with the project's "every layer
+replaceable" principle (any provider matching a small interface can
+be swapped in).
+
+---
+
+## 4. Principles
+
+The project makes design decisions against five tests, in order:
+
+1. **KISS** — the simplest thing that works, even if "fancier" is
+   available. A page is HTML until it has to be JS. A backend is one
+   binary until it has to be many.
+2. **Unix philosophy** — small pieces with simple data interfaces.
+   OSM owns *places*; our DB owns *contributions*; an OSM id joins
+   them.
+3. **Independence per section** — every layer of the stack is
+   replaceable on its own. OCR providers swap behind an interface.
+   The map page doesn't import the database client. Hosting,
+   analytics, OCR, and database are all candidate-swappable.
+4. **Open source / open data first** — software we can read, fork,
+   and self-host. Data we don't have to ask permission to use.
+5. **Don't migrate until it hurts** — when a simpler tool appears,
+   we document it as a migration target. We switch only when the
+   current tool stops earning its keep.
+
+These principles are operationalized in
+[ROADMAP.md](./ROADMAP.md) and [CONTRIBUTING.md](./CONTRIBUTING.md).
+
+---
+
+## 5. Architecture (one page)
+
+| Layer | Choice | Why |
+|---|---|---|
+| **Hosting** | GitHub Pages (static) | Free. Cacheable. No servers we operate. |
+| **Build** | [Vite](https://vitejs.dev/) multi-page | Each page is a real `.html` entry; no SPA shell. |
+| **Framework** | Vanilla JS, ES modules | Seven pages, mostly read-only. A component runtime would not earn its keep. |
+| **Map** | [Leaflet](https://leafletjs.com/) + OSM tiles | Mature, light, no API key. |
+| **Bookstore data** | [OpenStreetMap](https://www.openstreetmap.org/) via Overpass API | Source-of-truth for places; community-maintained; free. |
+| **Backend** | [Supabase](https://supabase.com/) (Postgres + Auth + Storage) | Open-source platform. Schema versioned as SQL migrations in the repo. |
+| **OCR** | PaddleOCR (ONNX) + Tesseract.js, both client-side | No cloud OCR billing. Swappable interface. |
+| **Spine detection** | YOLOv11n, in-browser | Fetched at runtime from Hugging Face; not committed to the repo. |
+| **Analytics** | [GoatCounter](https://www.goatcounter.com/) | Cookieless, privacy-respecting, GDPR-clean by default. |
+| **Tests / lint** | Vitest, ESLint, Prettier | Standard, runs in CI. |
+
+Operating cost target: **$0 to start, ~$20/month ceiling** (Supabase
+Pro tier only when free tier is outgrown). No paid infrastructure
+required to operate, fork, or self-host the project today.
+
+---
+
+## 6. Current status (as of 2026-Q2)
+
+The site is live and the core contribution loop works end-to-end:
+sign in, pick a shop, upload photos, confirm titles, see the result
+on the map.
+
+Recent work has focused on:
+
+- **Shipped** — a swappable OCR provider interface with a `?ocr=`
+  URL flag for A/B testing engines (PR #17); profile-page resilience
+  with per-section timeouts (PR #14); a standalone shelf-labeling
+  tool (PR #13).
+- **In review** — a YOLOv11n in-browser scan prototype with
+  tap-to-capture and IndexedDB persistence (PR #19); a KISS live
+  spine counter (PR #21); a public map of every `shop=books`
+  filtered by used/indie/chain (PR #20); device-aware OCR auto-pick
+  with a visible UI switch (PR #23).
+- **Documented** — a forward roadmap in [ROADMAP.md](./ROADMAP.md);
+  a no-JavaScript simplification plan; known operational blockers
+  in [BLOCKERS.md](./BLOCKERS.md).
+
+The project is intentionally still small. The right next milestone
+is **a smooth, reliable contributor flow on a smartphone, end-to-end**,
+not breadth of geographic coverage.
+
+---
+
+## 7. Roadmap
+
+See [ROADMAP.md](./ROADMAP.md) for the live version. In summary:
+
+- **Near term** — toolchain reproducibility, enabling cookieless
+  analytics, finishing the open PR queue, resolving HTTPS / iOS
+  Safari blockers, merging the device-aware OCR picker.
+- **Mid term** — move OCR model weights off git (runtime fetch,
+  same pattern proven by the scan prototype); port used/indie
+  category dots into the live map; daily ops digest via a scheduled
+  GitHub Action.
+- **Long term** — if and only if it earns its keep: contributor
+  re-capture workflows for inventory freshness, deeper bibliographic
+  linking (Wikidata, Open Library), partnerships with adjacent
+  archives.
+
+---
+
+## 8. Ethics: bookstores as partners
+
+OpenBookMap photographs the inside of private businesses. That has
+implications, and we name them rather than paper over them:
+
+- **Contributors should ask.** A shelf photo with the owner's
+  blessing is a contribution; a shelf photo taken without consent is
+  not what this project is for. We treat this as a contributor
+  norm, not a technical control.
+- **Stores can opt out.** A bookstore that doesn't want a
+  contribution on its OSM-id page can request removal via an issue
+  or email; we will honor it.
+- **No price scraping, no purchase routing.** We index *what's on
+  shelves*, not *what's for sale at what price*. The goal is
+  discoverability, not arbitrage.
+- **Contributors retain their copyright** to the images they
+  upload; they license the use of the image to the project under
+  terms compatible with open-data norms. (Final licensing details
+  are being settled — see §10.)
+
+This is the load-bearing ethical stance of the project. We exist
+because used bookstores are valuable; treating their owners as
+partners (and not as subjects of extraction) is non-negotiable.
+
+---
+
+## 9. Why this is a good fit for open infrastructure
+
+OpenBookMap is the kind of project that **does not exist** unless
+it's built as a commons:
+
+- A commercial competitor would silo the data and charge for it.
+- A bookstore-owned consortium would never form (too fragmented,
+  too small per shop).
+- A national library or archive would build something heavier and
+  more centralized.
+- A platform play (think "Yelp for books") would extract rent and
+  put bookstores in a bad bargaining position.
+
+A small, KISS, open-source project — that bookstores can ignore,
+verify, fork, or self-host — is the right shape for the problem.
+
+---
+
+## 10. Licensing & contribution
+
+- **Code**: MIT (see [LICENSE](./LICENSE)).
+- **Contributor data** (shelf photos, transcribed titles): the
+  project is finalizing licensing terms compatible with OSM/Wikidata
+  precedent (Open Database License + Creative Commons attribution
+  for media). Contributors will be asked to confirm the license at
+  upload. Current users see a license note on the upload page;
+  watch [PRIVACY.md](./PRIVACY.md) and CONTRIBUTING.md for updates.
+- **Bookstore location data** comes from OpenStreetMap under ODbL
+  and is not re-licensed by us.
+
+Contribution is open:
+
+- File issues, fix typos, ship code: see
+  [CONTRIBUTING.md](./CONTRIBUTING.md).
+- Upload shelf photos: [openbookmap.org/contribute](https://openbookmap.org/contribute).
+- Improve bookstore data: directly on [OpenStreetMap](https://www.openstreetmap.org/).
+- Partner discussions: open a discussion on the GitHub repo.
+
+---
+
+## 11. Who we'd like to talk to
+
+- **Independent bookstore owners and associations** — to test the
+  contributor flow on real shelves and refine the ethics of how it
+  shows up in their store.
+- **Open-data and civic-tech projects** — particularly anyone working
+  in adjacent commons (OSM contributors, Internet Archive volunteers,
+  Open Library, Wikidata).
+- **Researchers studying book culture and retail geography** — the
+  dataset, however small today, is publicly queryable.
+- **Funders and sponsors** willing to support open digital
+  infrastructure at a small budget. Cost ceilings, principles, and
+  honest limitations are documented openly; we're not asking for
+  scale capital.
+
+Reach the maintainer via the [GitHub repo](https://github.com/tautme/openbookmap).
+
+---
+
+## 12. Honest limitations
+
+Worth naming up front so nobody is surprised:
+
+- **OCR accuracy is the gating constraint.** Stylized spines,
+  rotated text, and faded covers all degrade extraction.
+  Contributors confirm titles before publish, which mostly hides this
+  from the public dataset — but it slows the contribution flow.
+- **Inventory freshness has no automation.** A shelf photographed
+  today is correct today. We don't yet have re-capture workflows.
+- **Coverage is wherever contributors are.** No central seeding.
+  Cities with active contributors will look great; cities without
+  will look empty.
+- **The contributor base is small.** This is a v0 project, not a
+  scaled platform.
+- **Dependencies on third-party services** (Supabase, GitHub Pages,
+  OSM Overpass) are real. We have migration targets documented for
+  each (see ROADMAP §Migration candidates).
+
+We surface these because the alternative — overclaiming and
+under-delivering — is a worse failure mode for an open project than
+shipping honestly.
+
+---
+
+*OpenBookMap — built KISS, open, and in public. Read the code; file
+a bug; photograph a shelf. The point of the project is the shelves.*