2026-07-29 20:02:36 +00:00
7 changed files with 440 additions and 8 deletions
--- a/README.md
+++ b/README.md
@@ -20,6 +20,9 @@ ci/        cross-cutting CI/quality notes
 | [HERO_EDITOR.md](website/HERO_EDITOR.md) | Hero canvas editor feature spec |
 | [WIKI_UPGRADE.md](website/WIKI_UPGRADE.md) | Wiki subsystem upgrade notes |
 | [SHARD_VISIBILITY.md](website/SHARD_VISIBILITY.md) | Who sees which shard data — the admin-configurable audience framework |
+| [SPAWN_ATLAS.md](website/SPAWN_ATLAS.md) | The bestiary / spawn atlas: what the shard contains, parsed from its own ServUO tree |
+| [CLILOCS.md](website/CLILOCS.md) | UO's id → name table: converting one from your client so items have names |
+| [MARKETPLACE.md](website/MARKETPLACE.md) | The player-vendor index: how it is gathered, what it costs, how to tune it |
 | [website-README.md](website/website-README.md) | Snapshot of the website repo's README (setup/run reference) |
 | [PROJECT_TREE.md](website/PROJECT_TREE.md) | Auto-generated snapshot of the repo's tracked file layout |

--- a/link/INTEGRATION.md
+++ b/link/INTEGRATION.md
@@ -449,6 +449,78 @@ would carry ~25 zeroes. `maxPoints` follows the same `0 == uncapped` rule as the
 a points lookup stops at the character's own row, but a rank must count every row that beats them, in
 every system, on every profile build. Derive rank from `points.board` instead for anyone in the top N.

+#### Player-vendor marketplace (Protocol 3.0)
+
+The shard-wide shop index: every player vendor's shop name, owner, location and priced inventory —
+the same set the in-game **Vendor Search** gump reads, published so a site can offer the same search
+from outside the game.
+
+An **amortized round-robin diff sweep**, not a snapshot RPC, and the distinction is load-bearing:
+`rpc.rs::try_route` correlates a reply on the FIRST frame carrying a matching `reqId`, so a chunked
+reply sharing one `reqId` would deliver chunk 1 to the HTTP caller and leak chunks 2..N onto the
+broadcast feed. A whole-world snapshot could not fit in one frame inside the 10 s reply timeout
+either. The per-account `vendor.snapshot` RPC (§5) is unaffected and still serves the player portal.
+
+Each tick inventories at most `Bridge.MarketSweepBatch` vendors (default 25) starting from a
+persistent cursor, so **per-tick cost is bounded independently of world size**; full coverage takes
+`ceil(vendors / batch) × MarketSweepSeconds`. A vendor is emitted only when its contents, prices,
+shop name or location actually change.
+
+| kind | fields | notes |
+|------|--------|-------|
+| `vendor.listing` | `serial`, `shopName`, `ownerSerial`, `ownerName`, `location{}`, `count`, `total`, `truncated`, `items[]` | One vendor's complete shop — **never a delta**. The latest frame for a `serial` replaces the previous one outright. |
+| `vendor.listing.remove` | `serial` | The shop is gone from the index: dismissed, expired, or its owner switched off the in-game Vendor Search flag. |
+
+```json
+{"kind":"vendor.listing","serial":"0x40001234",
+ "shopName":"Darrow's Bargains","ownerSerial":"0x1A2B","ownerName":"Darrow",
+ "location":{"map":"Trammel","x":1421,"y":1699,"z":0,
+             "region":"Britain","house":"Darrow's Villa"},
+ "count":2,"total":2,"truncated":false,
+ "items":[{"serial":"0x40012ABC","itemId":3922,"hue":0,"amount":1,
+           "price":25000,"name":null,"cliloc":1023721},
+          {"serial":"0x40012ABD","itemId":7026,"hue":1157,"amount":3,
+           "price":500,"name":"a shard sigil","cliloc":1041243}],
+ "t":1752489280000}
+```
+
+**Six things consumers get wrong.**
+
+1. **`name` is `null` for nearly every item; `cliloc` is the real label.** Items carry a
+   `LabelNumber`, not a name. The plugin deliberately never calls `VendorSearch.GetItemName`, which
+   builds an `ObjectPropertyList`, serialises it and byte-parses the packet **per item** — a
+   multi-hundred-millisecond stall across a full pass. (It would not work anyway: every current
+   client ships its cliloc files compressed and ServUO's bundled `Ultima.StringList` cannot read
+   them, so the in-game gump has the same gap.) Resolve clilocs consumer-side; a non-null `name` is a
+   player-set literal and is strictly more specific, so **prefer it over the cliloc**.
+2. **`location` is one nested object, and it may be absent entirely.** It is nested so that a
+   consumer gating vendor whereabouts gates one field rather than five that can drift apart — the
+   website's `market.location` rule removes the whole object. Treat a missing `location` as "not
+   published", not as an error.
+3. **`truncated` means the shop holds more than the frame carries.** `count` is what was published,
+   `total` is what the shop actually holds, capped by `Bridge.MarketMaxListings` (default 250). A
+   commodity reseller with thousands of stacked resources is real and an uncapped frame for one is
+   measured in megabytes. Say "showing 250 of 3,104" rather than presenting a partial shop as
+   complete.
+4. **`child: true` means the price buys the ENCLOSING CONTAINER.** ServUO prices a container as a
+   unit and everything inside inherits that price with no `VendorItem` of its own; `DoSearch`
+   surfaces the same flag. A UI that prints the container's price against each item inside it is
+   lying about the shard.
+5. **Opted-out vendors are absent, and that is a privacy control.** `pv.VendorSearch` is the player's
+   own in-game toggle and the sweep honours it — hide your vendor in game and it is hidden here too.
+   The same goes for `Map.Internal` and a null backpack, matching `DoSearch`. Process
+   `vendor.listing.remove` promptly: it is how a player *revoking* that consent reaches you.
+6. **Prices are inherently stale, by design.** The round-robin sweep means a shop can be a full cycle
+   behind. Any UI over this must say how old the data may be — the website derives it from the oldest
+   vendor row.
+
+Entries carry `ownerSerial`/`ownerName` and **never `acct` or `webId`**, the same rule `points.board`
+follows. Absent entirely if the shard runs `Bridge.MarketEnabled=false` or an older plugin. Render
+from `GET /market` (§6) on connect, then keep live with these events — though note that a live
+firehose of whole vendor inventories is the largest stream the bridge produces, and a consumer that
+only needs a browsable index (as the website does) is better served by the REST read plus the
+periodic re-sweep.
+
 ---

 ## 5. REST — read queries
@@ -829,6 +901,35 @@ standings built over months and blanking them during a restart reads as data los
 one excluded by `Bridge.PointsSystems`). That is distinct from a published board nobody has scored in
 yet, which is **200** with an empty `top[]` — and the two are worth rendering differently.

+### Player-vendor marketplace (Protocol 3.0)
+
+```
+GET /market?limit=200&offset=0
+→ { "vendors": [ {"kind":"vendor.listing","serial":"0x40001234",
+       "shopName":"Darrow's Bargains","ownerSerial":"0x1A2B","ownerName":"Darrow",
+       "location":{"map":"Trammel","x":1421,"y":1699,"z":0,
+                   "region":"Britain","house":"Darrow's Villa"},
+       "count":2,"total":2,"truncated":false,"items":[ ... ],"t":...}, ... ],
+    "total": 137, "limit": 200, "offset": 0 }
+```
+
+Every vendor's latest shop, exactly as `vendor.listing` published it (§4 for the frame and its six
+gotchas). Served from the sidecar's projection, so it answers while the shard is down.
+
+**This is the only PAGED read the sidecar serves**, because it is the only board that can be a whole
+world's inventory. `limit` is clamped to 1..1000 (default 200); `total` is returned so a caller knows
+when to stop rather than paging until it sees a short page, which would race a concurrent sweep.
+Ordering is by **serial**, not by shop name — a serial is stable while a shop name is renameable, so
+a rename mid-walk cannot make a vendor skip or repeat a page.
+
+The route is `/market` and deliberately **not** `/vendors`: `/vendors/{account}` next door is the
+per-account RPC (§5), and two routes a prefix apart meaning "this player's shops" and "every shop on
+the shard" is a trap nobody wins.
+
+Frames are served **verbatim**, owner names and coordinates included. That is not an oversight: the
+sidecar defines no audiences. Deciding who may see what is the consuming site's job — see
+[`v3.md`](v3.md) §3 for how the website does it.
+
 ---

 ## 7. Status codes
--- a/link/PLAN.md
+++ b/link/PLAN.md
@@ -284,6 +284,15 @@ Counts in `hello` are a live snapshot taken on the Core thread, not a cached val

 `Item.Name` is frequently `null`; the display name is `LabelNumber`, a cliloc id. **There is no `Data/Cliloc.enu` in this repo** — `BRIDGE_FINDINGS.md` §IV.4 is wrong about this. Cliloc data lives in the client install, which `DataPath` resolves to `D:\Games\Electronic Arts\Ultima Online Classic\`. Ship **both** `name` (when non-null) and `cliloc`, and resolve the number **on the website** against a cliloc map. That avoids a server-side dependency on the client directory.

+**Update (3.0).** That recommendation held, and the reason it had to hold turned out to be stronger
+than "avoids a dependency": **ServUO cannot resolve clilocs either.** Every current client ships its
+`Cliloc.*` files compressed, and the bundled `Ultima.StringList` reads only the older plain layout —
+so `VendorSearch.StringList` is null and `VendorSearch.GetItemName` returns `item.Name` on any modern
+shard. The in-game Vendor Search gump has the same gap, which is why `vendor.listing` never calls it.
+Pushing name resolution to the plugin was never an option. See [`v3.md`](v3.md) §8.6 and
+`docs/website/CLILOCS.md` for how the site gets a table instead (the operator converts one from their
+own client, once).
+
 ---

 ## 8. Corrections to `BRIDGE_FINDINGS.md`
@@ -327,6 +336,17 @@ leaderboards. `BridgePoints` is the widest read the bridge performs: ten of Serv
 keep a row for every character ever created, so it selects the top N in a single bounded pass rather
 than sorting, and runs on a deliberately slow 300 s interval.

+Also shipped: **`vendor.listing`** ([`v3.md`](v3.md) §8), `BridgeMarket.cs`, the shard-wide
+player-vendor index. It introduces the one sweep pattern the bridge did not previously have — an
+**amortized round-robin**. Every other sweep walks its whole collection per tick, which is fine for
+tens of houses or a fixed set of point systems and is not fine for a world of shops whose inventories
+recurse into containers. `BridgeMarket` inventories at most `MarketSweepBatch` vendors per tick from
+a persistent cursor, so the per-tick cost is bounded by the batch rather than by world size, and full
+coverage takes `ceil(vendors / batch) x MarketSweepSeconds`. Measured at **15.4 ms** for a cold tick
+of 25 vendors x 40 listings and **0.3 ms** in steady state (the per-vendor diff), on a shard of 209k
+items / 43k mobiles. It is also the first stream to honour a per-player privacy toggle: ServUO's own
+`PlayerVendor.VendorSearch` flag, so a shop hidden in game is hidden on the site.
+
 ### Config keys (`Config/Bridge.cfg`)

 ```ini
@@ -342,7 +362,8 @@ Read in `Configure()` via `Config.Get<T>("Bridge.<Key>", default)`. Key scope is

 The set above is the 1.0 sample, not the current one — every later phase added keys (sweep intervals
 for each board, the town-crier/news caps, the admin write plane, account provisioning, and 3.0's
-`RulesetEnabled` / `PublicConnectAddress` / `RulesetIncludeSchedule`, and the `Points*` block).
+`RulesetEnabled` / `PublicConnectAddress` / `RulesetIncludeSchedule`, and the `Points*` and `Market*`
+blocks).
 **`servuo-plugins/overlay/Config/Bridge.cfg`
 is the authoritative, commented list**; `BridgeConfig.cs` holds the defaults.

--- a/link/v3.md
+++ b/link/v3.md
@@ -15,8 +15,8 @@ Each part is marked off here as it lands on `edge`. §9 carries the same state p
 | 2 | **B/1** — `world.ruleset` (§5) | ✅ **Done** | servuo-plugins [#3](https://gitea.whitlocktech.com/RunicGateway/servuo-plugins/pulls/3), link [#17](https://gitea.whitlocktech.com/RunicGateway/link/pulls/17), website [#111](https://gitea.whitlocktech.com/RunicGateway/website/pulls/111), docs [#66](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/66) |
 | 3 | **C** — spawn atlas (§6) | ✅ **Done** | website [#112](https://gitea.whitlocktech.com/RunicGateway/website/pulls/112) (parsers + CLI + tables) + [#113](https://gitea.whitlocktech.com/RunicGateway/website/pulls/113) (API + pages + admin panel), docs [#67](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/67) + [#68](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/68) |
 | 4 | **B/2** — `points.board` (§7) | ✅ **Done** | servuo-plugins [#4](https://gitea.whitlocktech.com/RunicGateway/servuo-plugins/pulls/4), link [#18](https://gitea.whitlocktech.com/RunicGateway/link/pulls/18), website [#114](https://gitea.whitlocktech.com/RunicGateway/website/pulls/114), docs [#69](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/69) |
-| 5a | **B/3 dependency** — cliloc table (§8.6) | 🟨 In review | website [#115](https://gitea.whitlocktech.com/RunicGateway/website/pulls/115), docs [#70](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/70) |
-| 5b | **B/3** — `vendor.listing` (§8) | ⬜ Not started | — |
+| 5a | **B/3 dependency** — cliloc table (§8.6) | ✅ **Done** | website [#115](https://gitea.whitlocktech.com/RunicGateway/website/pulls/115), docs [#70](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/70) |
+| 5b | **B/3** — `vendor.listing` (§8) | 🟨 In review | servuo-plugins [#5](https://gitea.whitlocktech.com/RunicGateway/servuo-plugins/pulls/5), link [#19](https://gitea.whitlocktech.com/RunicGateway/link/pulls/19), website [#116](https://gitea.whitlocktech.com/RunicGateway/website/pulls/116), docs [#71](https://gitea.whitlocktech.com/RunicGateway/docs/pulls/71) |
 | 6 | **Cutover** — `PROTOCOL_VERSION` 2→3 (§4) | ⬜ Not started | — |

 Order 5 split in two once §8.6's cliloc dependency turned out to be a client-format problem rather
@@ -639,7 +639,7 @@ if it is renamed back.

 ---

-## 8. Part B/3 — `vendor.listing`
+## 8. Part B/3 — `vendor.listing` 🟨 In review

 ### 8.1 It cannot be an RPC, and this is load-bearing

@@ -807,6 +807,75 @@ and text paths converge on identical content.
 driven by `staleAt` (the oldest `shard_vendors.updated_at`). The round-robin sweep means data is
 inherently up to one full cycle old, and the UI must say so.

+Shipped with a second page, `routes/public/MarketVendor.jsx` at `/site/market/vendors/:serial` —
+where a search result points. It is the only surface that can render the two states the result list
+cannot: a `truncated` shop (*"showing 250 of 3,104 — this shop holds more than the shard
+publishes"*) and a `location` an admin has gated away, which is a real answer rather than an empty
+coordinate.
+
+### 8.8 What the build changed
+
+Four things the implementation settled differently from §8 as written, all of them found by building
+against the live shard.
+
+**1. `location` is a nested object, not flat `map`/`x`/`y`/`region`.** §8.1's payload sketch had them
+flat, and it would have made `market.location` — a rule Part A pre-wired — **inert**, exactly like
+the `characterName` miss §7.5 records: `projectValue` matches literal JSON keys, so there is no
+`location` key for the rule to match. Flat keys would have needed five rules that could drift apart.
+Nesting makes one rule hide the facet, the coordinates, the region and the house together, on the
+live frame and the stored read model alike, because both now spell it the same way.
+
+The other pre-wired rule, `market.ownerName`, checked out — it is a real key on the frame. Owner is
+written as flat `ownerSerial`/`ownerName` rather than through `BridgeJson.Actor`, which would add
+`acct` and `webId`; same argument `points.board` makes. `ownerSerial` was **added** to the
+configurable fields alongside `ownerName`, because an admin who hides the owner's name and leaves a
+serial every other board resolves back to that name has not hidden anything.
+
+**2. The per-vendor diff signature is the full listing set, not §8.3's `count | Σ(serial ^ price)`.**
+That hash collides on the single most common change a shop makes: two items swapping prices, which
+is what re-pricing looks like. The signature is built over the same buffer the frame is written
+from, in the same order, so a match really does mean an identical frame.
+
+**3. There is no `payload` column on `shard_vendors`.** §8.5 implied the board pattern (whole frame
+in JSON, columns hoisted for display). It does not apply here: the items ARE the searchable rows, so
+they are normalized into `shard_vendor_items` and there is nothing left worth duplicating. The
+sidecar keeps the whole blob, because outage resilience is its job and search is not.
+
+**4. Sweep cost is reported, and a slow tick warns.** The batch cap is a *claim* about per-tick cost,
+and an operator tuning `MarketSweepBatch` was otherwise tuning blind. `[bridge status` now carries
+`lastMs`/`maxMs`, and a tick over 50 ms prints a rate-limited warning naming the knob.
+
+Measured on the live shard (27 vendors × 40 listings, 209k items / 43k mobiles):
+
+| | |
+|---|---|
+| First tick — 25 vendors emitted cold | **15.4 ms** |
+| Second tick — the remaining 2 | **3.4 ms** |
+| Steady state — nothing changed | **0.3 ms** |
+| Website `/market` search over 1,040 listings | 1,040 total, names resolved |
+| Cliloc re-resolution pass over 1,040 rows | **50 ms** |
+
+The diff is what makes the steady state ~free; the batch cap is what bounds the cold case. Note the
+arithmetic the warning exists for: at the default cap of 250 listings, a batch of 25 **full** shops
+is 6,250 items ≈ 95 ms — over budget. Real shops hold tens, which is why 25 is the default, but a
+shard of commodity resellers should lower the batch, and now it will be told to.
+
+Two smaller things worth not rediscovering:
+
+- **`BridgeJson.Escape` takes a NON-NULL string** — it dereferences `value.Length` immediately — and
+  `BridgeJson.Str` writes its own `,"key":` prefix, so neither serves a value inside a hand-built
+  object. Nearly everything this frame writes is legitimately null (an item's plain `Name` is null
+  for almost every item; a vendor in the street has no house), so that is the common path, not an
+  edge case. `BridgeMarket.Text()` is the two-line writer that was missing.
+- **The ServUO console writes in the OS code page**, so an em dash in a `Console.WriteLine` renders
+  as `???` in the log an operator would paste into an issue. Bridge console output is ASCII.
+
+Search-side, one thing the site had to fix rather than inherit: `%` and `_` in a user's query are
+**LIKE** metacharacters, not SQL ones, so parameterization does not neutralize them — a search for
+`%` would otherwise match every listing on the shard. `shardMarket.db.js` escapes them. (The atlas's
+`LIKE` searches predate this and have the same shape over a much smaller table; worth a follow-up,
+not a blocker here.)
+
 ---

 ## 9. Sequencing
@@ -817,8 +886,8 @@ inherently up to one full cycle old, and the UI must say so.
 | 2 | **B/1** — `world.ruleset` (§5) | all four | new kind | ✅ Done |
 | 3 | **C** — spawn atlas (§6) | website, docs | none | ✅ Done |
 | 4 | **B/2** — `points.board` (§7) | all four | new kind + `char.profile` field | ✅ Done |
-| 5a | **B/3 dependency** — cliloc table (§8.6) | website, docs | none | 🟨 In review |
-| 5b | **B/3** — `vendor.listing` (§8) | all four | new kinds | ⬜ |
+| 5a | **B/3 dependency** — cliloc table (§8.6) | website, docs | none | ✅ Done |
+| 5b | **B/3** — `vendor.listing` (§8) | all four | new kinds | 🟨 In review |
 | 6 | **Cutover** — `PROTOCOL_VERSION` 2→3, `edge` → `main` | all four | the bump | ⬜ |

 ---
--- a/website/BACKEND_DESIGN.md
+++ b/website/BACKEND_DESIGN.md
@@ -407,6 +407,50 @@ Two values carry non-obvious meanings, both set by the plugin and both documente
  cliloc rather than a literal. Listing therefore orders by `COALESCE(name, system)`, so boards
  awaiting cliloc resolution sort by their own key rather than clumping together under NULL.

+### shard_vendors / shard_vendor_items — the player-vendor marketplace (Protocol 3.0)
+
+The shard-wide shop index, fed by `vendor.listing` / `vendor.listing.remove`. One row per player
+vendor and one per priced listing. Full operator detail in [`MARKETPLACE.md`](MARKETPLACE.md); the
+design is `docs/link/v3.md` §8.
+
+| Table | Shape |
+|---|---|
+| `shard_vendors` | `serial` (PK), `shop_name`, `owner_serial`, `owner_name`, `map`/`x`/`y`/`z`, `region`, `house`, `item_count`, `item_total`, `truncated`, `t`, `updated_at`. Indexes on owner, map, region and `updated_at`. |
+| `shard_vendor_items` | `id` (PK), `vendor_serial`, `serial`, `item_id`, `hue`, `amount`, `price`, `name`, `cliloc`, `display_name`, `child`. Indexes on `vendor_serial`, `price`, `item_id`, `display_name`, and `(display_name, price)`. |
+
+**Ingest is per-vendor and authoritative**: the frame is the whole shop, so ingest is
+delete-then-insert of that vendor's listings inside one transaction. All-or-nothing matters
+specifically because the two writes are "the shop" and "what is in it" — a failure between them
+leaves a shop advertising an inventory it no longer has, which is visibly wrong and indistinguishable
+from a genuinely empty shop. No foreign keys, consistent with every other `shard_*` table.
+
+**There is deliberately no `payload` column**, unlike `shard_points_boards` directly above. The
+board's top-N is a fixed-size list read whole, so it lives in JSON; here the items *are* the
+searchable rows, so they are normalized and nothing is left worth duplicating. The sidecar keeps the
+whole blob — outage resilience is its job, search is ours.
+
+Market state, not events: neither kind is in `LOGGED_KINDS`, and this is the strongest case of the
+three v3 kinds. One frame carries up to 250 listings and the sweep re-emits a shop on any price
+change, so logging would turn `shard_events` into a price history nobody reads.
+
+Two columns carry non-obvious meanings:
+
+- **`item_count` vs `item_total`.** `item_count` is what the frame published; `item_total` is what
+  the shop actually holds. They differ when `truncated` — the shard caps listings per frame
+  (`Bridge.MarketMaxListings`, 250 by default), and a commodity reseller with thousands of stacks
+  genuinely exceeds it. Any UI must show both or it presents a partial shop as complete.
+- **`display_name` is denormalized at ingest**, resolved from the item's literal `name` (preferred —
+  a player set it, so it is more specific) else its `cliloc` against `shard_clilocs`. Resolving at
+  query time would put the cliloc table on the hot path and make search-by-name impossible. Because
+  the shard's diff sweep will not re-send an unchanged shop just because the site learned what its
+  items are called, **a cliloc import triggers a bulk re-resolution** of this column (after a boot
+  import and after an admin import; ~50 ms per thousand rows, never throws).
+
+`updated_at` is written explicitly on every upsert rather than left to `ON UPDATE CURRENT_TIMESTAMP`,
+which MariaDB does not fire when every column is written back unchanged. A shop re-published
+identically is still *freshly confirmed*, and without this the staleness banner would age a perfectly
+current shop forever.
+
 ### shard_feature_visibility  — per-feature audience config (Protocol 3.0)

 One row per shard feature: `feature` (PK), `enabled`, `audience` (a rung on the ladder in §6.5),
@@ -745,6 +789,9 @@ from the per-route **siteMode** middleware (§5), never from an auth gate.
 | GET | `/shard/ruleset` | the shard's own published ruleset (Protocol 3.0 `world.ruleset`): expansion, which optional systems are on, skill/stat caps, account and house limits, champion scroll rules, the save/restart schedule. Served from `shard_ruleset`, so it renders while the shard is down; live via `world.ruleset` on `/shard/stream`. Behind `requireFeature('ruleset')`. **`null`** means the shard has never published one — a real answer, distinct from a published ruleset. `caps.skill` / `caps.totalSkill` are in **tenths** (1000 = 100.0). |
 | GET | `/shard/points` | every points/loyalty leaderboard the shard publishes (Protocol 3.0 `points.board`) — Queen's Loyalty, Void Pool, the nine city loyalties, Clean Up Britannia, … Served from `shard_points_boards`, so it renders while the shard is down; live via `points.board` on `/shard/stream`. Behind `requireFeature('leaderboards')`, ordered by display name. **`maxPoints: 0` means uncapped** (the common case), and `nameString` is usually `null` with `nameNumber` holding a cliloc — resolve client-side or humanise the `system` key. |
 | GET | `/shard/points/:system` | one board by the shard's `PointsType` name (e.g. `QueensLoyalty`); `:system` must match `/^[A-Za-z][A-Za-z0-9_]{0,47}$/` or **400** before any query runs. **404** = the shard has never published that system, which is distinct from a published board nobody has scored in yet (**200** with an empty `top`). |
+| GET | `/shard/market?q=&minPrice=&maxPrice=&itemId=&map=&region=&sort=&limit=&offset=` | search the player-vendor marketplace (Protocol 3.0 `vendor.listing`). Returns **listings**, not vendors — "who sells X and for how much" is the question, and a vendor-shaped result would make every caller flatten the shops back out. Served from `shard_vendors` + `shard_vendor_items`, so it renders while the shard is down. Behind `requireFeature('market')` **and rate-limited** — the first genuinely expensive public read on the site (a `LIKE` scan plus a `COUNT` over what is typically the largest `shard_*` table, reachable with no session). `sort ∈ {price_asc, price_desc, recent}`. `q` matches the resolved display name **or** the item's literal name, with `%`/`_` escaped: they are `LIKE` metacharacters, not SQL ones, so parameterization alone would let `?q=%` match every listing on the shard. Every response repeats `staleAt` (the oldest vendor row) because the shard sweeps round-robin — a banner that ages with the results it labels, not one fetched once. |
+| GET | `/shard/market/meta` | index size, staleness (`staleAt`/`freshAt`) and which facets and regions actually hold vendors, so a client builds its filters without running a search it will discard. |
+| GET | `/shard/market/vendors/:serial` | one shop and its listings; `:serial` must match `/^0x[0-9A-Fa-f]{1,16}$/` or **400** before any query runs. **404** = a serial the index has never seen, which also covers a vendor since dismissed or hidden — to an anonymous caller those are the same answer, and distinguishing them would leak that a hidden vendor exists. `truncated` (with `total` exceeding `count`) means the shop holds more than the shard publishes per frame. |
 | GET | `/shard/features` | the shard features **this caller** may reach plus the audience rung they resolved to (§6.5), so a client hides nav it can't follow. Reports only what the caller can see — the list itself never discloses a gated feature. Consumed by the SPA header and (pending) the Android nav. |
 | GET | `/atlas/creatures?q=&facet=&limit=&offset=` | the bestiary, most numerous first, with an unpaginated `total`. Static content parsed from the shard's ServUO tree — **not** sidecar-backed, which is why the atlas sits outside `/shard`, and unlike `/shard/*` it **is** site-mode gated. Behind `requireFeature('atlas')`. `?facet=` is matched exactly and never validated against a list (no facet name exists in the code); the filter is an `EXISTS` over the points rather than a JSON path or `JSON_SEARCH` built from caller input, whose `%`/`_` wildcards would make `?facet=%` match everything. |
 | GET | `/atlas/creatures/:slug` | one creature: `places` (the point-in-rect aggregate — "lizardman → Shrines, Isamu-Jima, Yew"), `spawners` (the bounded raw list, with `spawnersTruncated`), `alsoHere`. **`points` is a COUNT and `spawners` is the LIST** — named apart so one key never means a number on one route and an array on another. `minDelay`/`maxDelay` are in **seconds**, normalised at parse time from the source's per-record minutes-or-seconds. 404 = no such creature in this atlas. |
@@ -835,7 +882,11 @@ who"; `activity_log` provides the history feed.
 - **Cookie**: `httpOnly`, `sameSite=Lax`, `path=/`, and **`secure` decided per-request** (`COOKIE_SECURE=auto` → `secure: req.secure`).
 - **Trusted-device MFA.** A second, separate httpOnly cookie (`rg_trust`, default 30d) — opaque, sha256-hashed server-side in `trusted_devices` — lets a browser/app **skip the TOTP step** (never the password) on future logins. It is a server-side, per-row-revocable record (never a JWT claim), so the stateless session JWT is unchanged and trust stays revocable. It only ever gates the **second factor**; it deliberately outlives logout, and is cleared on untrust / password change / password reset / TOTP disable. **Recovery codes** (bcrypt, single-use) are the 2FA-lockout fallback. All admin trusted-device/MFA actions and the self actions (`auth.login.trusted_device`, `account.trusted_device.*`, `account.recovery_code*`, `admin.trusted_device.*`, `admin.user.totp.reset`) are audit-logged. See `docs/website/TRUSTED_DEVICES_MFA.md`. This is the key to dual access: the cookie is `Secure` when reached through Pangolin (HTTPS, `X-Forwarded-Proto: https`) but **not** `Secure` when reached directly over the LAN IP on plain HTTP — so login works in both. `COOKIE_SECURE=true|false` can force it. Requires `trust proxy` (below). `localhost:5173` (Vite) and `localhost:3000` are same-site, so the cookie flows in dev too.
 - **bcrypt** hashing (cost 10+); plaintext passwords never stored, logged, or returned.
- **Rate limiting** (`express-rate-limit`) on `/auth/login` and `/public/contact`.
+- **Rate limiting** (`express-rate-limit`) on `/auth/login`, `/public/contact`, and — the only limited
+  *read* — `/public/shard/market` and `/public/shard/market/vendors/:serial` (60/min/IP). Every other
+  public read is an indexed lookup of bounded size; the marketplace search is a `LIKE` scan plus a
+  `COUNT` over the largest `shard_*` table, anonymous by default, so it is the one public GET that is
+  worth money to serve.
 - **Validation** (`express-validator`) on all writes; centralized error handler.
 - **helmet** with a Content-Security-Policy tuned for the built React SPA. The policies now live in
  **`server/src/config/csp.js`** (`app.js` only wires them up):
--- a/website/MARKETPLACE.md
+++ b/website/MARKETPLACE.md
@@ -0,0 +1,167 @@
+# Marketplace — the player-vendor index
+
+**Status:** On `edge` — servuo-plugins [#5](https://gitea.whitlocktech.com/RunicGateway/servuo-plugins/pulls/5), link [#19](https://gitea.whitlocktech.com/RunicGateway/link/pulls/19), website [#116](https://gitea.whitlocktech.com/RunicGateway/website/pulls/116).
+**Design:** [`docs/link/v3.md` §8](../link/v3.md) — Protocol 3.0 Part B/3.
+**Depends on:** [`CLILOCS.md`](CLILOCS.md) — without a cliloc table, listings render as item ids.
+
+The marketplace is a searchable index of every player vendor on the shard: what
+each shop is selling, for how much, and where it is standing. It is the same set
+the in-game **Vendor Search** gump reads, offered from outside the game — so a
+player can find the vanquishing kryss they want before logging in, and someone
+who does not play at all can see that the economy exists.
+
+Page: `/site/market`, plus `/site/market/vendors/:serial` for one shop.
+
+## Three things the pages must say out loud
+
+Everything below follows from how the data is gathered, and each has a visible
+consequence the UI is required to surface.
+
+**1. The prices are not live.** The shard sweeps vendors **round-robin** — at
+most `Bridge.MarketSweepBatch` shops per tick — so a given shop can be a full
+cycle behind. The page carries a *"prices last refreshed N minutes ago"* banner
+driven by the **oldest** vendor row, not the newest: the one stale shop is the
+one that wastes somebody's trip.
+
+**2. A shop can be truncated.** `Bridge.MarketMaxListings` (250 by default) caps
+how many listings one frame carries. A commodity reseller with thousands of
+stacked resources is a real thing, and an uncapped frame for one is measured in
+megabytes. Over the cap the shop reports `truncated`, and the vendor page says
+*"showing 250 of 3,104 — this shop holds more than the shard publishes"* rather
+than presenting a partial shop as complete.
+
+**3. An item may have no name.** Items on the wire carry a cliloc id, not a name.
+On a shard whose operator has not converted a cliloc table
+([`CLILOCS.md`](CLILOCS.md)) the honest render is the item id — never an invented
+label, which would be indistinguishable from a real one.
+
+## Privacy: the player's own toggle wins
+
+Only vendors whose owner left the in-game **Vendor Search** flag ON are ever sent
+to the site. A player who hides their shop in game is hidden here too, and no
+admin setting overrides that. When they hide one that was already indexed, the
+shard emits `vendor.listing.remove` and the row is deleted — so revoking consent
+takes effect, it does not merely stop refreshing.
+
+Shop name, owner character name and location default to **Everyone**, because the
+stock Vendor Search gump already shows exactly that set to any player in game.
+They remain admin-configurable; see [`SHARD_VISIBILITY.md`](SHARD_VISIBILITY.md).
+Account names and website user ids never cross the wire at all.
+
+## How it is put together
+
+```
+ServUO                          uo-link sidecar               website
+──────                          ───────────────               ───────
+BridgeMarket.cs                 vendors table                 shard_vendors
+  round-robin sweep    ──────►    (whole frame blob)  ──────►  shard_vendor_items
+  per-vendor diff                 GET /market (paged)          + display_name
+  vendor.listing                                               resolved at ingest
+  vendor.listing.remove
+```
+
+**The shard side** walks at most `MarketSweepBatch` vendors per tick from a
+persistent cursor, diffs each against what it last published, and emits a whole
+frame for any shop that moved. Per-tick cost is therefore bounded by the batch,
+not by how many vendors the world holds — full coverage takes
+`ceil(vendors / batch) × MarketSweepSeconds`.
+
+**The sidecar** stores each frame whole and serves `GET /market`, its only paged
+read. It normalizes nothing and defines no audiences: it is a dumb forwarder, and
+search is the website's job.
+
+**The website** splits each frame into a vendor row and its listings, replacing
+that vendor's whole listing set inside one transaction (the frame is
+authoritative for that vendor, never a delta). Item names are resolved against
+the cliloc table **on the way in** and stored denormalized, which is what makes
+search-by-name possible and keeps the cliloc table off the hot path.
+
+## Operating it
+
+Everything is in `Config/Bridge.cfg` on the shard. There is nothing to configure
+on the website.
+
+| Setting | Default | What it does |
+|---|---|---|
+| `MarketEnabled` | `true` | Master switch. Off publishes nothing; the page shows an empty index. |
+| `MarketSweepSeconds` | `60` | Tick interval. |
+| `MarketSweepBatch` | `25` | Vendors inventoried per tick. Clamped 1..500. |
+| `MarketMaxListings` | `250` | Per-shop listing cap, after which `truncated`. Clamped 1..5000. |
+
+**Faster coverage vs. per-tick cost.** Lowering `MarketSweepSeconds` or raising
+`MarketSweepBatch` both refresh the index sooner and both cost more per tick.
+The expensive part is the item walk, which recurses into every container a vendor
+is selling — so a shard of big shops should raise the interval rather than the
+batch.
+
+`[bridge status` reports the sweep, including `lastMs` and `maxMs`:
+
+```
+market(enabled=True sweeps=42 scanned=108 emitted=27 removed=0 skipped=0
+       truncated=0 tracked=27 vendors=27 cursor=2 batch=25 lastMs=0.31 maxMs=15.40)
+```
+
+A tick over **50 ms** prints a rate-limited warning naming the knob:
+
+```
+[Bridge] market sweep took 82.4 ms (budget 50 ms) - lower Bridge.MarketSweepBatch (now 25) if this persists
+```
+
+Measured on a shard with 27 vendors × 40 listings (209k items, 43k mobiles):
+**15.4 ms** for the first cold tick of 25 vendors, **0.3 ms** in steady state —
+the diff is what makes an unchanged world nearly free. Note the arithmetic: 25
+*full* shops at the 250-listing cap is 6,250 items ≈ 95 ms, over budget. Real
+shops hold tens, which is why 25 is the default and why the warning exists.
+
+`[bridge sweepnow` runs one tick immediately; `[bridge reload` re-reads the
+settings above without a restart.
+
+## Names arriving late
+
+Item names come from the cliloc table, and the market sweep will **not** re-send
+an unchanged shop just because the site learned what its items are called. So a
+cliloc import triggers a bulk re-resolution of every stored listing — otherwise
+an operator who configures clilocs after the first sweep would see item ids until
+every shop happened to change on its own. It runs after a boot import and after
+an admin import, takes ~50 ms per thousand listings, and never throws: a failure
+leaves names exactly as they were.
+
+## API
+
+All under `/api/v1/public/shard`, gated by the `market` feature and
+**rate-limited** — these are the first genuinely expensive public reads on the
+site (a `LIKE` scan plus a `COUNT` over what is typically the largest `shard_*`
+table, reachable with no session).
+
+| Route | What |
+|---|---|
+| `GET /market` | Search. Returns **listings**, not vendors — "who sells X and for how much" is the question. `?q=&minPrice=&maxPrice=&itemId=&map=&region=&sort=&limit=&offset=`, `sort ∈ {price_asc, price_desc, recent}`. |
+| `GET /market/meta` | Index size, staleness, and which facets and regions actually hold vendors — so a client builds its filters without running a search it will discard. |
+| `GET /market/vendors/:serial` | One shop and its listings. **404** for a serial the index has never seen, which also covers a vendor since dismissed or hidden — to an anonymous caller those are the same answer. |
+
+`q` matches the resolved display name **or** the item's own literal name, because
+an item with a player-set name (most of what is worth searching for on a
+player-run shard) may carry a generic cliloc. `%` and `_` in a query are escaped:
+they are `LIKE` metacharacters, not SQL ones, so parameterization alone would let
+a search for `%` match every listing on the shard.
+
+Full schemas are in the OpenAPI spec (`ShardMarketPage`, `ShardMarketVendor`,
+`ShardMarketMeta`, `ShardMarketListing`, `ShardMarketLocation`).
+
+## Tables
+
+`shard_vendors` (one row per shop) and `shard_vendor_items` (one row per priced
+listing). Both are ingest-owned; nothing else writes to them. No foreign keys,
+in keeping with every other `shard_*` table — the ingest transaction is what
+keeps them consistent, and an FK would turn a malformed frame into a failed write
+rather than a dropped row.
+
+There is deliberately **no `payload` column** on `shard_vendors`, unlike the
+points board next door. The board's top-N is a fixed-size list read whole, so it
+lives in JSON; here the items *are* the searchable rows, so they are normalized
+and there is nothing left worth duplicating. The sidecar keeps the whole blob,
+because outage resilience is its job.
+
+`shard_vendor_items.display_name` is denormalized and indexed (alone, and
+composite with `price` for "cheapest matching X"). See "Names arriving late"
+above for how it is kept current.
--- a/website/SHARD_VISIBILITY.md
+++ b/website/SHARD_VISIBILITY.md
@@ -60,12 +60,32 @@ board while holding back one column. See the table in §3.
 | **Shard rules** | Skill/stat caps, house limits, vet rewards, the ruleset | Everyone | Connect address → Everyone |
 | **Spawn atlas** | Bestiary and spawn locations (static content) | Everyone | — |
 | **Leaderboards** | Point and loyalty standings | Everyone | Character names → Everyone |
-| **Marketplace** | The shard-wide player-vendor index | Everyone, **live updates off** | Vendor owner name → Everyone · Location → Everyone |
+| **Marketplace** | The shard-wide player-vendor index | Everyone, **live updates off** | Vendor owner name → Everyone · Vendor owner character id → Everyone · In-game location → Everyone |

 **Why the marketplace ships with live updates off.** A live feed of every vendor's full inventory
 would be the single largest thing the site sends. No page needs it — the marketplace is a search over
 stored data with a “prices last refreshed N minutes ago” stamp. Turn it on only if you want it.

+**Why the marketplace's fields default to Everyone.** A vendor's shop name, its owner's character
+name and where it is standing are *already* visible to every player in game: the stock Vendor Search
+gump surfaces exactly that set to anyone who opens it. Publishing them on the site is not a new
+disclosure. They stay configurable because a shard may still prefer to keep its economy behind a
+login — and because "already public in game" is a judgement about your shard, not ours.
+
+**Location is one setting covering four things.** Hiding it removes the facet, the coordinates, the
+region *and* the house name together. That is deliberate: those are four ways of saying the same
+thing, and a setting that hid the coordinates while publishing the house name would not have hidden
+anything.
+
+**Hiding the owner name also hides the owner character id.** They are separate settings so you can
+be explicit, but leaving the id published while hiding the name achieves nothing — the leaderboards
+and guild boards resolve that same id back to a character name. Set both.
+
+**What hiding a vendor cannot do.** Only vendors whose owner left the in-game *Vendor Search* flag ON
+are ever sent to the site, so a player who hides their shop in game is hidden here too — and no
+setting on this page can override that. It works the other way as well: these settings control who
+sees the index, not whether players can find each other's shops in game.
+
 **Why house owner/price default to Staff.** The public Houses page has always been a "where are the
 falling houses" board — location only. Owner and price are the staff view. That split is preserved.