diff --git a/marrow/mongo/taxonomy.py b/marrow/mongo/taxonomy.py
new file mode 100644
index 00000000..12136c5e
--- /dev/null
+++ b/marrow/mongo/taxonomy.py
@@ -0,0 +1,544 @@
+# encoding: utf-8
+
+from bson import ObjectId
+from pymongo.errors import BulkWriteError
+from contextlib import contextmanager
+
+from marrow.package.loader import traverse
+from marrow.package.canonical import name as name_
+from marrow.schema import Attribute
+
+from .query import Ops
+
+if __debug__:  # In development, we track how long certain operations take and log this at the DEBUG level.
+	from time import time
+
+
+log = __import__('logging').getLogger(__name__)
+
+
+class Taxonomy(Ops):
+	"""A factory for queries relating to, and routines for manipulating taxonomic structure.
+	
+	This maintains an acyclic directed graph (tree) structure for you using an API inspired by the jQuery DOM
+	traversal and manipulation APIs. Some attempt is made to group these methods into logical sections, but there's a
+	surprising amount of code in here. Coalesced paths are provided for convienent lookup, but not used for general
+	querying beyond the `nearest` lookup. The `parent` reference and `ancestors` list are used for general queries.
+	
+	The following are fields this Ops specializaion maintains for you:
+	
+	* `ancestors` - An array of sub-documents storing cached data about all ancestors.
+	* `parent` - An ObjectID reference to the immediate parent of the document.
+	* `path` - The coalesced path to the document, UNIX-style. No leading slash? It's detached.
+	* `index` - The position of the document within its parent.
+	
+	Documents missing the `ancestors` and `parent` fields, and containing a `path` not beginning ith `/` are in a
+	"detached" state. They may have descendants attached to them. The true root node of the graph should have a `name`
+	of `/` in order for path coalescing to work correctly.
+	"""
+	
+	CACHE = ('_id', 'name')  # You can override this in your own specialization to record more about ancestors.
+	
+	collection = Attribute()
+	
+	# Internal Support
+	# The follwoing are principally meant for internal use.
+	
+	@property
+	@contextmanager  # _ordered
+	def _ordered(self):
+		"""Prepare an ordered bulk operation as a context manager to centralize logging and error handling."""
+		
+		ops = self.collection.initialize_ordered_bulk_op()
+		
+		if __debug__:
+			log.debug("Preparing ordered bulk operation.")
+			start = time()
+		
+		yield ops
+		
+		if not ops._BulkOperationBuilder__bulk.ops:  # FRAGILE INTERNAL MAGIC
+			if __debug__:
+				log.warn("Ordered bulk operation aborted.")
+			return
+		
+		if __debug__:
+			duration = (time() - start) * 1000
+			log.debug("Ordered bulk operation preparation complete; executing.", extra=dict(duration=duration))
+			start = time()
+		
+		try:
+			result = ops.execute()
+		except BulkWriteError:
+			raise  # TODO: Better exception logging.
+		
+		if __debug__:
+			duration = (time() - start) * 1000
+			log.debug("Ordered bulk operation complete.", extra=dict(result, duration=duration))
+	
+	@property
+	@contextmanager  # _unordered
+	def _unordered(self):
+		"""Prepare an unordered bulk operation as a context manager to centralize logging and error handling"""
+		ops = self.collection.initialize_unordered_bulk_op()
+		
+		if __debug__:
+			log.debug("Preparing unordered bulk operation.")
+			start = time()
+		
+		yield ops
+		
+		if not ops._BulkOperationBuilder__bulk.ops:
+			if __debug__:
+				log.warn("Unordered bulk operation aborted.")
+			return
+		
+		if __debug__:
+			duration = (time() - start) * 1000
+			log.debug("Unordered bulk operation preparation complete.", extra=dict(duration=duration))
+			start = time()
+		
+		try:
+			result = ops.execute()
+		except BulkWriteError:
+			raise  # TODO: Better exception logging.
+		
+		if __debug__:
+			duration = (time() - start) * 1000
+			log.debug("Unordered bulk operation complete.", extra=dict(result, duration=duration))
+	
+	def _project(self, fields):
+		"""Construct a simplified MongoDB projection based on the provided iterable of field names."""
+		
+		if fields:
+			projection = {'_id': 0}
+			projection.update((field, 1) for field in fields)
+			return projection
+	
+	def _target(self, target):
+		"""Resolve a target reference.
+		
+		This will transform a Taxonomy query into one selecting the same elements by ID, wraps bare ObjectId in a
+		Taxonomy selecting for it, or, given an unsaved document, will save that document then provide a Taxonomy
+		selecting for it.
+		"""
+		
+		if isinstance(target, Taxonomy):  # Documents to attach as selected by Taxonomy.
+			return target
+		
+		elif isinstance(target, ObjectId):  # Document to attach by ID.
+			return self.id(target)
+		
+		elif target.get('_id', None):  # Docuent with existing ID.
+			return self.id(target['_id'])
+		
+		return self.id(self.collection.insert_one(target).inserted_id)
+	
+	def _move_target(self, target):
+		"""Resolve a target reference and ensure it is detached, ready for attachment.
+		
+		Identical to `_target` in semantics, but returns both a Taxonomy instance and count of affected items.
+		"""
+		
+		if isinstance(target, Taxonomy):  # Documents to attach as selected by Taxonomy.
+			target = target.detach()
+			return target, target.count()
+		
+		elif isinstance(target, ObjectId):  # Document to attach by ID.
+			return self.id(target).detach(), 1
+		
+		elif target.get('_id', None):  # Docuent with existing ID.
+			return self.id(target['_id']).detach(), 1
+		
+		return self.id(self.collection.insert_one(target).inserted_id), 1
+	
+	# Query Terminals
+	# These prevent further chaining by immediately executing the query to return a result.
+	
+	def first(self, *fields, **kw):
+		"""Retrieve the first record matched by our current query."""
+		
+		if 'sort' in kw:
+			kw['sort'] = [(i.lstrip('-'), -1 if i[0] == '-' else 1) for i in kw['sort']]
+		
+		return self.collection.find_one(self.as_query, self._project(fields), **kw)
+	
+	def all(self, *fields, **kw):
+		"""Retrieve a cursor selecting the records matched by our current query."""
+		
+		if 'sort' in kw:
+			kw['sort'] = [(i.lstrip('-'), -1 if i[0] == '-' else 1) for i in kw['sort']]
+		
+		return self.collection.find(self.as_query, self._project(fields), **kw)
+	
+	def one(self, *fields, **kw):
+		"""Retrive the record matched by our current query, requiring no more than one result."""
+		
+		if self.count(**kw) > 1:
+			raise ValueError("Requested one record, matched multiple.")
+		
+		return self.first(*fields, **kw)
+	
+	def count(self, **kw):
+		"""Retrieve the count of records matched by our current query."""
+		return self.all(**kw).count()
+	
+	def update(self, update, **kw):
+		"""Update the records matched by our current query."""
+		return self.collection.update(self.as_query, update, **kw)
+	
+	def delete(self, **kw):
+		"""Delete the records matched by our current query."""
+		return self.collection.delete(self.as_query, **kw)
+	
+	def scalar(self, *fields, **kw):
+		"""Retrieve the given field or fields from records matched by our current query.
+		
+		If only a single field is given, an iterable of just that field is returned. If multiple fields are selected,
+		you'll get back an iterable of tuples of those fields in the same order.
+		
+		Utilizes `marrow.package.loader:traverse`, allowing for deep lookup.
+		"""
+		
+		if len(fields) == 1:  # Special case for single field scalars.
+			field = fields[0]
+			
+			for result in self.all(*fields, **kw):
+				yield result.get(field)
+			
+			return
+		
+		for result in self.all(*fields, **kw):  # Slightly simpler multi-field scalars.
+			yield tuple(traverse(result, field, default=None) for field in fields)
+	
+	def nearest(self, path, projection=None):
+		"""Immediately look up the asset nearest the given path, utilizing any query built so far.
+		
+		Allows for optional projection. This must immediately issue the query due to the fairly complex query.
+		"""
+		
+		if __debug__:  # Give useful error messages in development.
+			if 'collection' not in self.__data__ or not self.collection:
+				raise RuntimeError("Called method requiring bound collection without one.")
+		
+		if hasattr(path, 'split'):  # We accept iterables of individual elements, or a string.
+			path = path.split('/')
+		else:
+			path = list(path)
+		
+		# Remove leading empty elements.
+		while path and not path[0]:
+			del path[0]
+		
+		# Remove trailing empty elements.
+		while path and not path[-1]:
+			del path[-1]
+		
+		# Determine the full list of possible paths and prepare the query.
+		paths = [('/' + '/'.join(path[:i])) for i in range(len(path) + 1)]
+		criteria = (self & {'path': {'$in': paths}}).as_query
+		
+		if hasattr(projection, 'as_projection'):  # Get the bare projection.
+			projection = projection.as_projection
+		
+		if __debug__:
+			log.debug("Searching for asset nearest: /" + '/'.join(path), extra=dict(
+					query = criteria,
+					projection = projection
+				))
+		
+		# Issue the query against our bound collection.
+		return self.collection.find_one(criteria, projection, sort=(('path', -1), ))
+	
+	# Basic Queries
+	# These cover fairly basic attributes of asset documents, for convienence.
+	
+	def id(self, ref):
+		data = self.__data__.copy()
+		data['operations'] = {'_id': ref}
+		return self.__class__(**data)
+	
+	def named(self, name):
+		"""Filter to assets with a given name, useful primarily when chained with other criteria."""
+		return self & {'name': name}
+	
+	def of_type(self, kind):
+		"""Restrict to instances of specific Asset subclasses.
+		
+		Specify the kind either as a full dot-colon import path (such as is stored in the database) or by passing in
+		an actual Asset subclass to filter for.
+		
+		Unlike an `isinstance()` check, this is explicit and does not include subclasses of the target class.
+		
+		For example:
+		
+			context.taxonomy.of_type(Page)
+			context.taxonomy.of_type('web.component.page.model:Page')
+		"""
+		if not isinstance(kind, str):
+			kind = name_(kind)
+		
+		return self & {'_cls': kind}
+	
+	# Asset Management
+	# Basic management operations.
+	
+	def empty(self):
+		"""Delete all descendants of the currently filtered assets.
+		
+		Because this is removing every single descendant, there is no need to update asset ordering. Allows chaining
+		as the parent elements (currently filtered elements) remain.
+		
+		Because this is an operation against the descendants, not containers, no modification times are updated.
+		"""
+		
+		result = self.collection.delete_many(self.descendants.as_query)
+		
+		log.warn("Deleted %d document%s.", result.deleted_count, "" if result.deleted_count == 1 else "s")
+		
+		return self
+	
+	def detach(self):
+		"""Detach selected documents from the taxonomy."""
+		
+		with self._unordered as bulk:
+			for target, ancestors in self.scalar('_id', 'ancestors'):
+				if not ancestors: continue  # Skip already detached assets.
+				
+				target = self.id(target)
+				
+				# Update this immediate target, removing its path and parent/ancestors.
+				bulk.find(target.as_query).update_one({
+						'$unset': {'path': 1, 'parent': 1, 'ancestors': 1}
+					})
+				
+				# Update descendants of that target, eliminating the ancestors we've just detached from.
+				bulk.find(target.descendants.as_query).update({
+						'$unset': {'path': 1},
+						'$pullAll': {'ancestors': ancestors}
+					})
+		
+		(self | self.descendants).coalesce()
+		
+		return self
+	
+	def coalesce(self):
+		"""Recalculate the path and update the modification time for all selected documents."""
+		
+		with self._unordered as bulk:
+			for target, name, ancestors in self.scalar('_id', 'name', 'ancestors'):
+				if ancestors:
+					path = '/'.join([ancestor['name'].lstrip('/') for ancestor in ancestors] + [name])
+				else:
+					path = name
+				
+				bulk.find({'_id': target}).update_one({
+						'$currentDate': {'updated': {'$type': 'date'}},
+						'$set': {'path': path}
+					})
+	
+	def insert(self, index, target):
+		"""Attach the given target as a child of the filtered asset, at the given index.
+		
+		If the target is an unsaved document (no `_id` defined) it is first saved. If the target is an ObjectId
+		instance, minimal relevant details are first loaded. Negative indexes are permitted.
+		
+		Chaining from this method results in operations affecting the inserted Asset, not the containing one.
+		"""
+		
+		container = self.one('_id', 'name', 'ancestors')
+		parents = container.pop('ancestors', []) + [container]
+		
+		# Identify the largest possible index and calculate the index to attach at.
+		
+		last = self.children.first('index', sort=('-index', ))
+		if last: last = last['index']  # We only care about the actual index.
+		else: last = 0  # No existing children.
+		
+		if index < 0:  # Allow insertion at indices relative to the end.
+			index = last - index + 1
+		
+		index = max(0, min(index, last + 1))  # Limit to the available range.
+		
+		# Identify the documents we are attaching.
+		
+		target, nTargets = self._move_target(target)
+		targets = target | target.descendants
+		
+		# Process the attachments.
+		
+		# These must happen first, since we don't want to accidentally alter the indexes we add later.
+		with self._unordered as bulk:
+			# Add a gap to the indexes of siblings.
+			bulk.find(self.children.gte(index).as_query).update({'$inc': {'index': nTargets}})
+			
+			# Associate ancestors.
+			bulk.find(targets.as_query).update({'$push': {'ancestors': {'$each': parents, '$position': 0}}})
+		
+		# Associate parent and assign index.
+		with self._unordered as bulk:
+			for i, child in enumerate(target.scalar('_id')):
+				bulk.find({'_id': child}).update_one({'$set': {'parent': container['_id'], 'index': index + i}})
+		
+		return self
+	
+	def append(self, child):
+		"""Append the given child to the selected parent."""
+		return self.insert(2**24, child)
+	
+	def appendTo(self, parent):
+		"""Append the selected documents as children of the given parent."""
+		
+		ids = list(self.salar('_id'))
+		parent = self._target(parent)
+		parent.insert(2**24, self)
+		
+		if len(ids) == 1:
+			return self.id(ids[0])
+		
+		return self.id({'$in': ids})
+	
+	def prepend(self, child):
+		"""Prepend the given child to the selected parent."""
+		return self.insert(0, child)
+	
+	def prependTo(self, parent):
+		"""Prepend the selected documents to the given parent."""
+		
+		ids = list(self.salar('_id'))
+		parent = self._target(parent)
+		parent.insert(0, self)
+		
+		if len(ids) == 1:
+			return self.id(ids[0])
+		
+		return self.id({'$in': ids})
+	
+	def after(self, target):
+		"""Insert the target document after the selected document."""
+		
+		sibling = self.one('_id', 'index', 'parent')
+		self.id(sibling['parent']).insert(sibling['index'] + 1, target)
+		return self.id(sibling['_id'])
+	
+	def before(self, target):
+		"""Insert the target document before the selected document."""
+		
+		sibling = self.one('_id', 'index', 'parent')
+		self.id(sibling['parent']).insert(sibling['index'], target)
+		return self.id(sibling['_id'])
+	
+	def insertAfter(self, sibling):
+		"""Insert the selected documents after the target sibling."""
+		
+		sibling = self._target(sibling).one('_id', 'index', 'parent')
+		self.id(sibling['parent']).insert(sibling['index'] + 1, self)
+		return self  # TODO: Need to re-select IDs like prependTo; lazy.
+	
+	def insertBefore(self, sibling):
+		"""Insert the selected documents before the target sibling."""
+		
+		sibling = self._target(sibling).one('_id', 'index', 'parent')
+		self.id(sibling['parent']).insert(sibling['index'], self)
+		return self  # TODO: Need to re-select IDs like prependTo; lazy.
+	
+	# Taxonomy Querying
+	# These are properties where possible to make chaining more elegant.
+	
+	@property  # children
+	def children(self):
+		"""Select for all children of the currently filtered assets."""
+		
+		extra = {k: v for k, v in self.__data__.items() if k != 'operations'}
+		parents = list(self.scalar('_id'))
+		
+		if len(parents) == 1:
+			return self.__class__({'parent': parents[0]}, **extra)
+		
+		return self.__class__({'parent': {'$in': parents}}, **extra)
+	
+	@property  # descendants
+	def descendants(self):
+		"""Select for all desdcendants of the currently filtered assets."""
+		extra = {k: v for k, v in self.__data__.items() if k != 'operations'}
+		parents = list(self.scalar('_id'))
+		
+		if len(parents) == 1:
+			return self.__class__({'ancestors._id': parents[0]}, **extra)
+		
+		return self.__class__({'ancestors._id': {'$in': list(self.scalar('_id'))}}, **extra)
+	
+	# Sibling queries.
+	
+	@property  # next
+	def next(self):
+		"""Select the sibling immediatly following this one."""
+		
+		child = self.one('index', 'parent')
+		sibling = self.id(child['parent']).children.gt(child['index']).first('_id', sort=('index', ))
+		
+		if not sibling:
+			return
+		
+		return self.id(sibling['_id'])
+	
+	@property  # nextAll
+	def nextAll(self):
+		"""Select all siblings following this one."""
+		
+		child = self.one('index', 'parent')
+		return self.id(child['parent']).children.gt(child['index'])
+	
+	@property  # prev
+	def prev(self):
+		"""Select the sibling immediately prior to this one."""
+		
+		child = self.one('index', 'parent')
+		sibling = self.id(child['parent']).children.lt(child['index']).first('_id', sort=('-index', ))
+		
+		if not sibling:
+			return
+		
+		return self.id(sibling['_id'])
+	
+	@property  # prevAll
+	def prevAll(self):
+		"""Select all siblings prior to this one."""
+		
+		child = self.one('index', 'parent')
+		return self.id(child['parent']).children.lt(child['index'])
+	
+	@property  # siblings
+	def siblings(self):
+		"""Select all siblings of this document."""
+		
+		child = self.one('index', 'parent')
+		return self.id(child['parent']).children.lt(child['index']).gt(child['index'])
+	
+	# Querying by position.
+	
+	def eq(self, index):
+		return self & {'index': index}
+	
+	def gt(self, index):
+		return self & {'index': {'$gt': index}}
+	
+	def gte(self, index):
+		return self & {'index': {'$gte': index}}
+	
+	def lt(self, index):
+		return self & {'index': {'$lt': index}}
+	
+	def lte(self, index):
+		return self & {'index': {'$lte': index}}
+	
+	# Complex queries.
+	
+	def has(self, filter):
+		"""Find the element (out of those currently selected) which contains a descendant matching the filter."""
+		raise NotImplementedError()
+	
+	def closest(self, filter):
+		"""For each document selected find the closest ancestor matching the given filter."""
+		raise NotImplementedError()
+	
diff --git a/setup.py b/setup.py
index 25770561..e508af5b 100755
--- a/setup.py
+++ b/setup.py
@@ -1,9 +1,9 @@
 #!/usr/bin/env python
 # encoding: utf-8
 
+import codecs
 import os
 import sys
-import codecs
 
 try:
 	from setuptools.core import setup, find_packages
@@ -89,6 +89,7 @@
 			development = tests_require + ['pre-commit'],  # Development-time dependencies.
 			scripting = ['javascripthon<1.0'],  # Allow map/reduce functions and "stored functions" to be Python.
 			logger = ['tzlocal'],  # Timezone support to store log times in UTC like a sane person.
+			taxonomy = ['pathlib2; python_version < "3.4"'],  # Path manipulation utility lib; builtin in 3.4 and 3.5.
 		),
 	
 	tests_require = tests_require,