diff --git a/e2e/ruby/rails/demo/test/integration/observability_instrumentation_test.rb b/e2e/ruby/rails/demo/test/integration/observability_instrumentation_test.rb
new file mode 100644
index 0000000000..9c79a41206
--- /dev/null
+++ b/e2e/ruby/rails/demo/test/integration/observability_instrumentation_test.rb
@@ -0,0 +1,45 @@
+# frozen_string_literal: true
+
+require 'test_helper'
+
+# End-to-end coverage for the LaunchDarkly observability plugin's Rails
+# auto-instrumentation.
+#
+# Background: the plugin configures OpenTelemetry from `Plugin#register`, which
+# runs when the LaunchDarkly client is created. In this app that happens in
+# `config/initializers/launchdarkly.rb` — i.e. DURING Rails boot — so the OTel
+# Rails-family instrumentations (Rack, ActionPack, ActiveRecord, ...) install
+# correctly. A customer who instead creates the client lazily AFTER boot sees
+# "Instrumentation: OpenTelemetry::Instrumentation::ActionPack failed to install"
+# because the ActiveSupport.on_load hooks those instrumentations rely on have
+# already fired. These tests pin the working boot-time behavior so a regression
+# (or a change that breaks instrumentation install) is caught in CI.
+class ObservabilityInstrumentationTest < ActionDispatch::IntegrationTest
+  # The Rails-family instrumentations that must attach during a boot-time init.
+  # These are exactly the ones that report "failed to install" on the lazy path.
+  RAILS_INSTRUMENTATIONS = %w[Rack ActionPack ActiveRecord ActiveSupport Rails].freeze
+
+  def instrumentation_instance(name)
+    Object.const_get("OpenTelemetry::Instrumentation::#{name}::Instrumentation").instance
+  end
+
+  test 'rails auto-instrumentation installed during boot' do
+    RAILS_INSTRUMENTATIONS.each do |name|
+      assert instrumentation_instance(name).installed?,
+             "#{name} instrumentation should be installed after a boot-time plugin init " \
+             '(it reports "failed to install" when the client is created lazily after boot)'
+    end
+  end
+
+  test 'http request produces a server span via the rack instrumentation' do
+    exporter = OpenTelemetry::SDK::Trace::Export::InMemorySpanExporter.new
+    processor = OpenTelemetry::SDK::Trace::Export::SimpleSpanProcessor.new(exporter)
+    OpenTelemetry.tracer_provider.add_span_processor(processor)
+
+    get pages_home_url
+    assert_response :success
+
+    server_spans = exporter.finished_spans.select { |s| s.kind == :server }
+    refute_empty server_spans, 'expected an HTTP server span from the Rack/ActionPack instrumentation'
+  end
+end
diff --git a/e2e/ruby/rails/demo/test/test_helper.rb b/e2e/ruby/rails/demo/test/test_helper.rb
index 0c92e8e881..967eed2a76 100644
--- a/e2e/ruby/rails/demo/test/test_helper.rb
+++ b/e2e/ruby/rails/demo/test/test_helper.rb
@@ -1,6 +1,14 @@
 # frozen_string_literal: true
 
 ENV['RAILS_ENV'] ||= 'test'
+
+# The observability plugin only configures OpenTelemetry (and installs the Rails
+# auto-instrumentation) when the LaunchDarkly client registers it, which requires
+# a non-empty SDK key. Set a dummy key BEFORE the app boots so the instrumentation
+# attaches during initialization. The key is invalid, so the client never connects
+# (background connection attempts fail gracefully and do not affect tests).
+ENV['LAUNCHDARKLY_SDK_KEY'] ||= 'sdk-test-0000000000000000000000'
+
 require_relative '../config/environment'
 require 'rails/test_help'
 
diff --git a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability.rb b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability.rb
index dec63a7c40..f1f421d83a 100644
--- a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability.rb
+++ b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability.rb
@@ -13,7 +13,13 @@
 
 require_relative 'launchdarkly_observability/middleware'
 require_relative 'launchdarkly_observability/otel_log_bridge'
-require_relative 'launchdarkly_observability/rails'
+
+# NOTE: rails.rb is required at the *bottom* of this file, after the
+# LaunchDarklyObservability module body has been fully defined. Its Railtie
+# registers a `config.after_initialize` hook that runs synchronously when the
+# gem is required lazily after Rails has booted. That hook references module
+# constants and `class << self` methods, so they must already exist when the
+# require runs. See the require at the end of this file.
 
 module LaunchDarklyObservability
   # Default OTLP endpoint for LaunchDarkly Observability
@@ -179,3 +185,9 @@ def otel_logger_provider_available?
     end
   end
 end
+
+# Required last, on purpose: the Rails Railtie's `config.after_initialize` hook
+# can run synchronously during this require (lazy require after Rails has
+# booted), and it depends on the fully-defined LaunchDarklyObservability module
+# above. See the note next to the other require_relative calls at the top.
+require_relative 'launchdarkly_observability/rails'
diff --git a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/hook.rb b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/hook.rb
index 630336b501..f0709a5de2 100644
--- a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/hook.rb
+++ b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/hook.rb
@@ -91,12 +91,23 @@ def build_before_attributes(series_context)
         FEATURE_FLAG_PROVIDER_NAME => 'LaunchDarkly'
       }
 
-      context = series_context.context
-      if context
-        attrs[FEATURE_FLAG_CONTEXT_ID] = context.fully_qualified_key
-      end
+      attrs[FEATURE_FLAG_CONTEXT_ID] = context_id_for(series_context)
 
-      attrs
+      # OpenTelemetry rejects nil attribute values and logs an error for each
+      # one. Drop nils centrally so individual call sites cannot reintroduce the
+      # "invalid attribute value type NilClass" noise (e.g. an invalid context
+      # whose fully_qualified_key is nil).
+      attrs.compact
+    end
+
+    # Returns the fully-qualified context key, or nil if unavailable.
+    #
+    # An invalid context (e.g. a non-string context kind) is still a non-nil
+    # object but yields a nil fully_qualified_key. OpenTelemetry rejects nil
+    # attribute values ("invalid attribute value type NilClass"), so callers
+    # must omit the attribute entirely when this returns nil.
+    def context_id_for(series_context)
+      series_context.context&.fully_qualified_key
     end
 
     def serialize_value(value)
@@ -127,10 +138,7 @@ def add_feature_flag_event(span, series_context, detail)
         FEATURE_FLAG_PROVIDER_NAME => 'LaunchDarkly'
       }
 
-      context = series_context.context
-      if context
-        event_attributes[FEATURE_FLAG_CONTEXT_ID] = context.fully_qualified_key
-      end
+      event_attributes[FEATURE_FLAG_CONTEXT_ID] = context_id_for(series_context)
 
       if detail.variation_index
         event_attributes[FEATURE_FLAG_RESULT_VARIANT] = detail.variation_index.to_s
@@ -146,7 +154,9 @@ def add_feature_flag_event(span, series_context, detail)
         add_reason_event_details(event_attributes, reason)
       end
 
-      span.add_event(FEATURE_FLAG_EVENT_NAME, attributes: event_attributes)
+      # See build_before_attributes: drop nil values so a missing context id (or
+      # any future optional attribute) cannot emit a nil-attribute OTel error.
+      span.add_event(FEATURE_FLAG_EVENT_NAME, attributes: event_attributes.compact)
     end
 
     def add_reason_event_details(event_attributes, reason)
diff --git a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/opentelemetry_config.rb b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/opentelemetry_config.rb
index 7c0603e2b3..4ad8215e5b 100644
--- a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/opentelemetry_config.rb
+++ b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/opentelemetry_config.rb
@@ -105,9 +105,16 @@ def configure_traces
     # User-provided instrumentation config is merged on top of defaults,
     # so users only need to specify the instrumentations they want to override.
     def configure_instrumentations(config)
+      # Only pass options that the instrumentations actually accept. Unknown
+      # options are not fatal but emit a warning on every boot ("ignored the
+      # following unknown configuration options [...]"):
+      # - `enable_recognize_route` is not an option on the Rails, Rack, or
+      #   ActionPack instrumentations; route-based span naming (http.route) is
+      #   handled automatically by the ActionPack instrumentation.
+      # - ActiveRecord has no `db_statement` option; SQL capture comes from the
+      #   database adapter instrumentations (Mysql2, PG, ...) which default to
+      #   obfuscating statements.
       defaults = {
-        'OpenTelemetry::Instrumentation::Rails' => { enable_recognize_route: true },
-        'OpenTelemetry::Instrumentation::ActiveRecord' => { db_statement: :include },
         'OpenTelemetry::Instrumentation::Net::HTTP' => { untraced_hosts: [] },
         'OpenTelemetry::Instrumentation::Rack' => { untraced_endpoints: ['/health', '/healthz', '/ready'] }
       }
diff --git a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/rails.rb b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/rails.rb
index fa50f2d389..c6238dd549 100644
--- a/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/rails.rb
+++ b/sdk/@launchdarkly/observability-ruby/lib/launchdarkly_observability/rails.rb
@@ -127,8 +127,18 @@ def attach_otel_log_bridge
           warn "[LaunchDarklyObservability] Could not attach log bridge to Rails.logger: #{e.message}"
         end
 
+        # The availability check is inlined here rather than delegating to
+        # LaunchDarklyObservability.otel_logger_provider_available? on purpose.
+        # rails.rb is required from launchdarkly_observability.rb *before* that
+        # file's `class << self` block (which defines the module method) has run.
+        # When the gem is lazily required after Rails has booted, the
+        # `config.after_initialize` hook above executes synchronously while this
+        # file is still loading, so the module method does not exist yet and the
+        # delegation raised "undefined method `otel_logger_provider_available?'".
         def otel_logger_provider_available?
-          LaunchDarklyObservability.send(:otel_logger_provider_available?)
+          defined?(OpenTelemetry::SDK::Logs::LoggerProvider) &&
+            OpenTelemetry.respond_to?(:logger_provider) &&
+            OpenTelemetry.logger_provider.is_a?(OpenTelemetry::SDK::Logs::LoggerProvider)
         end
       end
 
diff --git a/sdk/@launchdarkly/observability-ruby/test/hook_test.rb b/sdk/@launchdarkly/observability-ruby/test/hook_test.rb
index 700006cbd8..3f15f68665 100644
--- a/sdk/@launchdarkly/observability-ruby/test/hook_test.rb
+++ b/sdk/@launchdarkly/observability-ruby/test/hook_test.rb
@@ -192,6 +192,47 @@ def test_context_id_uses_fully_qualified_key_for_multi_kind
     assert_equal 'org:org-1:user:user-1', span.attributes['feature_flag.context.id']
   end
 
+  # Regression: an invalid context (e.g. a non-string kind) is a non-nil
+  # object whose fully_qualified_key is nil. Setting a nil OTel attribute
+  # raised "OpenTelemetry error: invalid attribute value type NilClass for
+  # key 'feature_flag.context.id'" on every evaluation. The attribute must be
+  # omitted instead.
+  def test_invalid_context_omits_context_id_attribute
+    invalid_context = LaunchDarkly::LDContext.create({ key: 'k', kind: 123 })
+    refute invalid_context.valid?
+    assert_nil invalid_context.fully_qualified_key
+
+    series_context = LaunchDarkly::Interfaces::Hooks::EvaluationSeriesContext.new(
+      'my-flag', invalid_context, false, :variation
+    )
+
+    # Capture OpenTelemetry's error channel. Setting an attribute to nil does not
+    # raise and does not set the attribute (OTel drops it) — the only observable
+    # symptom is an "invalid attribute value type NilClass" error logged here,
+    # which is exactly the noise the customer saw on every evaluation.
+    captured = StringIO.new
+    original_logger = OpenTelemetry.logger
+    OpenTelemetry.logger = ::Logger.new(captured)
+    begin
+      data = @hook.before_evaluation(series_context, {})
+      @hook.after_evaluation(series_context, data, create_evaluation_detail)
+    ensure
+      OpenTelemetry.logger = original_logger
+    end
+
+    refute_match(/invalid .*attribute value type NilClass/, captured.string,
+                 'must not emit an OTel nil-attribute error for an invalid context')
+    refute_match(/feature_flag\.context\.id/, captured.string)
+
+    span = @exporter.finished_spans.first
+    refute_nil span
+    refute span.attributes.key?('feature_flag.context.id')
+
+    flag_event = span.events.find { |e| e.name == 'feature_flag' }
+    refute_nil flag_event
+    refute flag_event.attributes.key?('feature_flag.context.id')
+  end
+
   def test_nil_context_does_not_raise_and_still_records_event
     series_context = LaunchDarkly::Interfaces::Hooks::EvaluationSeriesContext.new(
       'my-flag', nil, false, :variation
diff --git a/sdk/@launchdarkly/observability-ruby/test/rails_railtie_test.rb b/sdk/@launchdarkly/observability-ruby/test/rails_railtie_test.rb
index c8a163f766..da4c7072e0 100644
--- a/sdk/@launchdarkly/observability-ruby/test/rails_railtie_test.rb
+++ b/sdk/@launchdarkly/observability-ruby/test/rails_railtie_test.rb
@@ -114,5 +114,30 @@ def test_rails_file_loads_when_after_initialize_runs_immediately
            'ViewHelpers should be defined after loading rails.rb'
     assert defined?(LaunchDarklyObservability::Railtie),
            'Railtie should be defined after loading rails.rb'
+
+    # Regression: the Railtie's `attach_otel_log_bridge` runs from the synchronous
+    # `config.after_initialize` block *during* the require of launchdarkly_observability.rb,
+    # before that file's `class << self` block (which defines the module-level
+    # `otel_logger_provider_available?`) has executed. The Railtie used to delegate to
+    # that not-yet-defined method, so the bridge attach failed with:
+    #
+    #   Could not attach log bridge to Rails.logger: undefined method
+    #   `otel_logger_provider_available?' for module LaunchDarklyObservability
+    #
+    # The Railtie's check must be self-contained. Simulate the load-order state by
+    # removing the module method, then assert the Railtie check still works.
+    sc = LaunchDarklyObservability.singleton_class
+    saved = sc.instance_method(:otel_logger_provider_available?)
+    sc.send(:remove_method, :otel_logger_provider_available?)
+    begin
+      result = nil
+      assert_silent do
+        result = LaunchDarklyObservability::Railtie.send(:otel_logger_provider_available?)
+      end
+      assert_includes [true, false], result
+    ensure
+      sc.send(:define_method, :otel_logger_provider_available?, saved)
+      sc.send(:private, :otel_logger_provider_available?)
+    end
   end
 end