docs/herb-ore-category-fix-spec.md

@@ -0,0 +1,114 @@
+# Fix: herb/ore categorization breaks for almost every zone in the v2 pool
+
+## The bug, confirmed by reading the code
+
+`SplitByCategory()` (`GoldRush.cpp:648`) decides whether a resource entry
+is a herb or ore purely by matching keywords in its display name:
+
+```cpp
+bool isHerb = ContainsWord(name, "lotus") || ContainsWord(name, "clover") || ContainsWord(name, "thorn") || ContainsWord(name, "bloom");
+bool isOre = ContainsWord(name, "vein");
+```
+
+This was written against the *original 5-entry Northrend default pool*
+(Goldclover→"clover", Icethorn→"thorn", Frost Lotus→"lotus", Titanium
+**Vein**→"vein") and never updated when the v2 zone-resources feature (PR
+#5) introduced ~70 new resource names across 57 zones.
+
+**Confirmed failure case**: tested live in Hellfire Peninsula. Its
+configured pool is Felweed, Dreaming Glory, Fel Iron **Deposit**,
+Adamantite **Deposit**. None of these match any herb keyword, and neither
+ore entry contains "vein" — most Outland/Classic ore uses "Deposit," not
+"Vein." So both `herbEntries` and `oreEntries` come back empty, and
+`SpawnHotspot()` (`GoldRush.cpp:998-1017`) falls back to
+`fallbackEntries` (the whole site pool, unsplit) for every node. The
+intended herb/ore alternation silently never happens for the large
+majority of the 57 configured zones — only the handful whose names happen
+to contain one of the four hardcoded herb words or "vein" categorize
+correctly at all.
+
+**Observed symptom**: the test run in Hellfire produced overwhelmingly
+Felweed with no observed ore, despite the fallback pool containing ore
+entries. Whether that's pure sampling variance or a second, separate
+issue (see "Worth checking after this fix" below) isn't confirmed from
+reading the code alone — don't assume the placement-bias theory without
+testing after this fix ships.
+
+## Fix: stop guessing category from the name, use a known lookup table
+
+Every entry ID used across the 57-zone pool was already individually
+verified against `acore_world.gameobject_template` when the pool was
+built (`docs/zone-specific-resources-spec.md`). Rather than continue
+patching an ever-growing, ever-fragile keyword list — the next zone added
+will just as easily introduce another unmatched name — replace the
+name-based guess with a static ID→category lookup built from data that's
+already verified correct.
+
+**Add a lookup table** (e.g. `static const std::unordered_map<uint32,
+bool> IsHerbById` or similar, `true` = herb, `false` = ore), populated
+from every entry in the tables below. Change `SplitByCategory` to check
+this table first; **keep the existing name-keyword check as a fallback
+only** for any entry not found in the table (defensive — covers the
+original global-pool default entries and anything added by hand later
+without an explicit table update).
+
+### Herb entries (mark `true`)
+```
+1618, 3724, 1617, 3725, 1619, 3726,           // Peacebloom, Silverleaf, Earthroot
+1620, 3727, 1621, 3729, 2045, 1622, 3730,     // Mageroyal, Briarthorn, Stranglekelp, Bruiseweed
+1623, 1624, 2041, 2042, 2046, 2043, 2866,     // Wild Steelbloom, Kingsblood, Liferoot, Fadeleaf, Goldthorn, Khadgar's Whisker, Firebloom
+142140, 180165, 142141, 176642,               // Purple Lotus, Arthas' Tears
+142142, 176636, 180164, 142143, 183046,       // Sungrass, Blindweed
+142144, 142145, 176637, 176587, 176641,       // Ghost Mushroom, Gromsblood, Plaguebloom
+176583, 176638, 180167, 176584, 176639,       // Golden Sansam, Dreamfoil
+180168, 176586, 176640, 180166, 176588,       // Mountain Silversage, Icecap
+191303, 176589,                                // Firethorn, Black Lotus
+181270, 183044, 181271, 183045, 181275,       // Felweed, Dreaming Glory, Ragveil
+183043, 181277, 181276, 181279, 181280, 181281, // Terocone, Flame Cap, Netherbloom, Nightmare Vine, Mana Thistle
+189973, 190171, 190172, 190176, 190170,       // Goldclover, Lichbloom, Icethorn, Frost Lotus, Talandra's Rose
+191019, 190169                                 // Adder's Tongue, Tiger Lily
+```
+
+### Ore entries (mark `false`)
+```
+1731, 2055, 3763, 103713, 181248,             // Copper Vein
+1732, 2054, 3764, 103711, 181249,             // Tin Vein
+1733, 105569,                                  // Silver Vein
+1735,                                           // Iron Deposit
+1734, 150080, 181109,                          // Gold Vein
+2040, 150079, 176645,                          // Mithril Deposit
+2047, 150081, 181108,                          // Truesilver Deposit
+324, 150082, 176643, 175404,                   // Small/Rich Thorium Vein
+165658,                                         // Dark Iron Deposit
+181555, 181556, 181569, 181570, 181557,        // Fel Iron, Adamantite (+ Rich), Khorium
+189978, 189979, 189980, 189981, 191133         // Cobalt/Saronite Deposit (+ Rich), Titanium Vein
+```
+
+(This is every entry that appears anywhere in `gold_rush.conf.dist`'s
+`GoldRush.ZonePool` plus the module's own default global
+`GoldRush.NodeEntries` — cross-check against both files while
+implementing, in case an entry was missed here.)
+
+## Worth checking after this fix — don't build blind for this part
+
+Once real ID-based categorization is in, re-test Hellfire Peninsula (or
+another Outland zone) with `.goldrush teststart` a few times. If ore
+still rarely/never appears despite `oreEntries` now correctly populated,
+look at the placement-validation loop (`GoldRush.cpp:1019` onward) —
+specifically the `MaxVerticalDrift` cliff/rock-face rejection. Ore
+deposits are commonly placed on steep terrain in the actual game world;
+if placement attempts for ore entries are failing validation far more
+often than herb entries, that's a second, separate bug worth its own fix
+— but confirm it's actually happening before changing that logic, rather
+than assuming.
+
+## Testing
+
+- Hellfire Peninsula (the confirmed failure case) should now show a mix
+  of Felweed/Dreaming Glory *and* Fel Iron/Adamantite, not all-herb.
+- Spot-check one or two zones whose original names *did* match the old
+  keyword list (e.g. Un'Goro Crater — Firethorn contains "thorn", Black
+  Lotus contains "lotus") to confirm the table-based path didn't regress
+  anything that used to work by keyword-coincidence.
+- Confirm entries not in either table (if any exist) still fall through
+  to the old keyword check rather than being silently dropped.