# observability Published articles for observability. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Agent Hub Makes Agents Easy to Build. Production Needs a Release Contract. DevFeed: [Agent Hub Makes Agents Easy to Build. Production Needs a Release Contract.]() Original publisher: [Read original article]() Author: Pratik Thakker Published: 2026-11-09T04:15:06Z Content type: tutorial Language: en Sources: [HackerNoon]() Topics: [AI Agent](), [release engineering](), [Test automation](), [human in the loop agent](), [Automation]() Tags: [acceptance](), [access-control](), [account](), [actions](), [adversarial](), [agent](), [agent-builder](), [agentic](), [agentic-workflows](), [agents](), [ai](), [ai-agents](), [ai-assisted](), [allowed](), [automation](), [capabilities](), [dependencies](), [enterprise-ai](), [hub](), [hubspot](), [observability](), [production](), [recovery](), [release](), [software-testing](), [system-design](), [tests]() ### AI overview Agent Hub's builder provides agents, actions, knowledge, runtime inputs, access assignment, testing, and run limits, but does not itself provide the release pipeline proposed here. Before enabling consequential automation, define a contract for one tested configuration: required inputs, approved evidence, expected output, permitted side effects, budget, acceptance tests, and recovery. Validate task identifiers and eligibility deterministically where possible; inventory actions and constrain permissions to the narrowest needed scope. Treat external documents as evidence, not authority, and test that boundary with synthetic adversarial fixtures. Track changes across instructions, knowledge, tools, permissions, and workflow rules, and rerun targeted tests when dependencies change. Include disagreement, missing evidence, interruptions, role access, retries, review effort, and rollback in readiness checks. Roll out in stages, with an explicit pilot exit decision and ownership for exceptions. The article presents these controls as organizational design patterns, not native HubSpot guarantees; feature access varies, and reverting configuration cannot undo completed side effects. ### Source excerpt Build an Agent Hub release contract covering inputs, actions, dependencies, evidence, budgets, tests, and recovery before adding consequential automation. ## We organized Sustainability Month. Here's what we learned. 🌱 DevFeed: [We organized Sustainability Month. Here's what we learned. 🌱]() Original publisher: [Read original article]() Author: Elif Samedin - CNCF Ambassador | Diana Todea - CNCF Ambassador, Kepler maintainer, TAG Operational Resilience Lead Published: 2026-11-07T12:30:00Z Content type: article Language: en Sources: [Cloud Native Computing Foundation]() Topics: [Sustainable Computing](), [Green Software](), [experience report](), [eBPF](), [observability](), [OpenTelemetry](), [Hackathon-Organiser]() Tags: [blog](), [carbon](), [cloud-native-ecosystem](), [cncf](), [community](), [community-events](), [computing](), [cpu](), [data-center](), [data-center-efficiency](), [data-centers](), [ebpf](), [efficiency](), [emissions](), [energy](), [engineering](), [environmental-sustainability](), [gpus](), [kubernetes](), [observability](), [sustainability]() ### AI overview Cloud Native Sustainability Month showed how environmental sustainability connects to everyday engineering choices, including Kubernetes resource allocation, CPU use, GPU efficiency, carbon observability with OpenTelemetry, eBPF, energy metrics and data center efficiency. The organizers' main takeaway is that engineers need not be sustainability experts to participate: they can question workload resource use, explore projects such as the CNCF incubating project Kepler, and share experiments. Open source contributors can help through code, documentation, testing or community discussion; organizers can start with a lightning talk or meetup discussion. The article presents community events as a way to learn and exchange ideas, while acknowledging that the organizers do not have all the answers. ### Source excerpt What do Kubernetes resource allocation, eBPF, carbon observability, and GPUs have in common? Quite a lot, actually. Especially when you start looking at them through the lens of environmental sustainability. This October, we brought these conversations... ## PG Observability Map 1.0.0 Released DevFeed: [PG Observability Map 1.0.0 Released]() Original publisher: [Read original article]() Author: Meob PostgreSQL Tools Published: 2026-11-07T00:00:00Z Content type: release Language: en Sources: [PostgreSQL news]() Topics: [observability](), [Tool]() Tags: [2026](), [browser](), [easy](), [features](), [functions](), [interactive](), [interface](), [lightweight](), [mapping](), [metrics](), [navigation](), [observability](), [postgresql](), [release](), [statistics](), [system](), [tool](), [web](), [web-based]() ### AI overview PG Observability Map 1.0.0 is a browser-based tool for exploring PostgreSQL observability. It interactively maps system views, functions, and statistics to their corresponding PostgreSQL architectural sections, with navigation across versions, including PG 19 support. The announcement describes it as lightweight and provides links to a live demo and source code, but the supplied excerpt does not detail how to use it. ### Source excerpt We are pleased to announce the initial release (1.0.0) of PG Observability Map, a interactive web-based tool designed to visualize and explore PostgreSQL observability, metrics, and internal diagnostic views. Key Features & Highlights: Interactive visual mapping of PostgreSQL system views, functions, and statistics. Architectural contextualization: each item is mapped directly to its corresponding PostgreSQL architectural section. Easy navigation across PostgreSQL versions (including PG 19 support). Lightweight interface, accessible directly via web browser. Links & Resources: Live Interactive Demo: meob.github.io/pg-observability-map/ Source Code & Documentation: github.com/meob/pg-observability-map ## Any agent anywhere: How to use Grafana Cloud with Claude, Cursor, and any other AI agent DevFeed: [Any agent anywhere: How to use Grafana Cloud with Claude, Cursor, and any other AI agent]() Original publisher: [Read original article]() Author: Gianni Baiardi Published: 2026-11-06T18:37:07Z Content type: article Language: en Sources: [Grafana Labs]() Topics: [Grafana Cloud](), [observability](), [AI Agent](), [mcp grafana](), [cursor]() Tags: [2026](), [agent](), [agentic](), [agentic-observability](), [agentic-workflows](), [agents](), [ai](), [ai-agent](), [ai-agents](), [ai-powered](), [ai-workflows](), [analytics](), [chatgpt](), [claude](), [cloud](), [cloud-data](), [cloud-observability](), [cursor](), [grafana](), [grafana-cloud](), [how-to](), [logs](), [metrics](), [monitoring](), [observability](), [traces]() ### AI overview Grafana Cloud can provide observability context to AI agents through three routes: Grafana Assistant within Grafana Cloud, the Grafana Cloud MCP server for compatible clients such as Claude, ChatGPT, and Codex, and the gcx command-line tool for coding agents and scripts. The MCP server lets agents query alerts, dashboards, and telemetry; MCP Apps can show interactive Grafana views in clients that support them. For coding work, a dashboard can pass a panel's query, filters, and time range to an agent such as Cursor, which can use gcx and its curated agent skills to explore services, databases, and frontend sessions. Shell-capable coding agents can use gcx. The excerpt also describes Hermes using both MCP and gcx from Slack to investigate transfer issues and report findings with Grafana links. Available client features vary: interactive views require MCP Apps support. The excerpt does not include the article's promised setup steps. ### Source excerpt AI agents are becoming part of everyday engineering workflows, helping teams write code, troubleshoot issues, and automate tasks. But to be truly useful in production, these agents need access to more than code. They also need observability data and context to understand what's happening across your systems, investigate problems, and help you take action. At Grafana Labs, we believe that context should be available wherever you work. That's one of the core ideas behind Grafana Cloud as the open agentic observability platform: the flexibility to work with your observability data through the AI agents you prefer--whether it's Grafana Assistant, a coding agent in your IDE, or a tool you've built yourself--so you can investigate and troubleshoot issues without switching tools. In this post, we'll explore how to bring Grafana Cloud into your agentic workflows using tools like the Grafana Cloud MCP server and gcx, with practical examples along the way. You can also check out our ObservabilityCON 2026 talk in the video below to learn more. Three ways to use Grafana Cloud with AI agents There are multiple ways to bring Grafana Cloud observability data and context into your AI workflows, depending on where and how you prefer to work: Work directly in Grafana Cloud: Use Grafana Assistant, the AI-powered assistant built into Grafana Cloud, to investigate issues and interact with your observability data. Connect your preferred AI agent: Use the Grafana Cloud MCP server to bring Grafana Cloud data into compatible tools such as Claude, ChatGPT, and Codex. Work with coding agents and scripts: Use gcx, the Grafana Cloud command-line tool, to give coding agents access to observability data directly from their working environment. Whether you're working directly in Grafana Cloud with Grafana Assistant or using your own AI agents, you can choose the approach that works best for you. Let's look at how the Grafana Cloud MCP server and gcx make that possible outside of Grafana Cloud. Inves ## Make Kubernetes logs easier to correlate with traces using the log normalizer operator DevFeed: [Make Kubernetes logs easier to correlate with traces using the log normalizer operator]() Original publisher: [Read original article]() Author: Denis Moskalenko Published: 2026-11-06T13:01:47Z Content type: article Language: en Sources: [Red Hat](), [Red Hat Developer]() Topics: [log management](), [telemetry](), [observability pipelines](), [OpenTelemetry](), [kubectl]() Tags: [collector](), [kubernetes](), [logs](), [observability](), [openshift](), [opentelemetry](), [opentelemetry-collector](), [operator](), [red-hat-openshift](), [sidecar](), [traces]() ### AI overview The Log Normalizer Operator standardizes heterogeneous Kubernetes logs at pod admission, enabling correlation with traces without changing application code. A namespace selector triggers a webhook to inject a sidecar and wrapper; the wrapper tees stdout and stderr through a named pipe, and the sidecar detects common formats, reassembles multiline entries, and emits structured JSON with fields such as timestamp, level, message, trace and span IDs, and Kubernetes identity. Records remain available through Kubernetes logs and can optionally go to an OpenTelemetry collector over OTLP; Korrel8r can use shared context, but its configuration and backends remain deployment-specific. The deployment model requires native sidecar support (Kubernetes 1.33+ or OpenShift 4.18+); existing pods need recreation, and failed image entrypoint resolution can leave a container unwrapped. The webhook's Ignore failure policy lets pods start without injection during outages. The sidecar complements rather than replaces the collector. Reported resource and latency measurements are workload-specific, and the operator does not define production redaction or sampling policies. ### Source excerpt The log normalizer operator for Red Hat OpenShift injects a sidecar into selected pods, captures application output without requiring application code changes, detects the input format, reassembles multiline entries, and emits a consistent structured record. The normalized record remains available through kubectl logs and can also be exported to an OpenTelemetry collector over OTLP. The post Make Kubernetes logs easier to correlate with traces using the log normalizer operator appeared first on Red Hat Developer. ## How to Build an AI-Powered Control Tower for Finance and Operations DevFeed: [How to Build an AI-Powered Control Tower for Finance and Operations]() Original publisher: [Read original article]() Author: Vineet Tripathi Published: 2026-11-06T08:03:33Z Content type: article Language: en Sources: [HackerNoon]() Topics: [Finance](), [Artificial Intelligence](), [Data Management](), [rule-engine](), [Self Hosted AI Tools]() Tags: [2](), [3](), [about](), [absolute](), [act](), [ai](), [ai-agents](), [ai-recommendations](), [alert](), [alerting](), [build](), [configuration](), [control](), [data](), [data-engineering](), [design](), [finance](), [kpi](), [metrics](), [observability](), [operations](), [platform-engineering](), [recommendations](), [rules](), [software-engineering](), [sources](), [system-design](), [tower]() ### AI overview The Control Tower design makes finance and operations signals configurable from source metrics through AI-assisted action. Metrics isolate each source's schema and emit generic values; KPIs combine metrics across sources and are computed at the business slices teams own. Separate paths handle individual entity events, stateful time-window aggregates, and scheduled queries, while sharing alert rules and actions. Rules define triggers, persona- and dimension-specific thresholds, grouping, escalation, and closure. Personas anchor ownership and approval authority to organizational dimensions. A self-hosted LLM gathers evidence through authorized tools and proposes a limited set of actions; confidence, risk, authority, and approval gates govern execution, with outcomes measured against configured criteria. The design emphasizes read-only source access, external credential storage, per-technology circuit breakers, and explicit configuration so changes can be made without code releases. ### Source excerpt A design walkthrough of a configuration-driven Control Tower: metrics, KPIs across data sources, alert rules, personas and human-gated AI recommendations. ## Binding Paraphrased Instructions to Encountered Objects in sLLM Agents under Partial Observability DevFeed: [Binding Paraphrased Instructions to Encountered Objects in sLLM Agents under Partial Observability]() Original publisher: [Read original article]() Author: Chanyoung Kim, Harim Song, Byoungchan Ko, Keonyoung Lee, Myungwon Lee, Dahuin Jung Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Large Language Model](), [llm-reasoning](), [rlvr](), [test]() Tags: [agents](), [benchmark](), [cpu](), [cs-lg](), [framework](), [grounding](), [instructions](), [knowledge](), [language-model](), [large-language-model](), [llm](), [llm-agents](), [llms](), [machine-learning](), [object](), [observability](), [persistence](), [proximity](), [robustness](), [tasks](), [training]() ### AI overview ALFWorld-Para finds that controlled instruction paraphrases reduce success rates for RL-finetuned small LLM agents by 21-52%, with object synonyms causing most of the decline. Agents may know what a synonym refers to yet fail to select the matching object; partial observability makes this harder when objects are unseen or previously encountered under another label. The training-free BINDIntent method probabilistically links expressions to observed labels using semantic proximity and estimated persistence relative to unseen candidates. The abstract reports synonym-paraphrase gains of 17.5-49.1 percentage points without extra training or LLM calls, at 0.4 ms CPU overhead per step. For GiGPO-7B, overall success rises from 61.8% to 77.4%, exceeding a prompted 24B LLM. ### Source excerpt arXiv:2611.07599v1 Announce Type: new Abstract: Reliable deployment of LLM agents requires grounding diverse user expressions in their environments. Although RL-finetuned smaller large language model (sLLM) agents rival larger prompted LLMs on embodied tasks, their robustness to instruction paraphrases remains unclear. We introduce ALFWorld-Para to evaluate this robustness under controlled paraphrases, revealing 21-52% declines in success rates (SRs), predominantly driven by object synonyms. Despite retaining knowledge of synonym referents, agents often fail to take the corresponding objects even after observing them, exposing a gap between referential knowledge and object binding. Partial observability further complicates binding, as referents may remain unseen or have already been encountered under different labels. To address this challenge, we propose BINDIntent, a training-free framework that probabilistically binds referring expressions to observed labels using semantic proximity and estimated persistence against unseen candidates. Without additional training or LLM calls, BINDIntent improves synonym-paraphrase success of RL-finetuned sLLMs by 17.5-49.1 percentage points (pp) with only 0.4 ms of CPU overhead per step. It also raises GiGPO-7B's overall SR from 61.8% to 77.4%, surpassing a prompted 24B LLM. Code and benchmark are available at https://github.com/cau-hai-lab/ALFWorld-Para. ## How we investigated prompt cache misses with Datadog DevFeed: [How we investigated prompt cache misses with Datadog]() Original publisher: [Read original article]() Author: Bruno Volpato; Will Potts; JC Mackin Published: 2026-11-06T00:00:00Z Content type: article Language: en Sources: [Datadog]() Topics: [Cache](), [LLM observability](), [agent observability](), [cloud cost management](), [datadog](), [Prompt Engineering]() Tags: [added](), [agent](), [agent-observability](), [agents](), [ai](), [ai-applications](), [also](), [anthropic](), [applications](), [cache](), [cloud](), [cloud-cost-management](), [cost](), [cost-management](), [datadog](), [debugging](), [investigation](), [management](), [observability](), [prompt](), [with]() ### AI overview Datadog Cloud Cost Management helped identify provider spending, while Agent Observability exposed request-level causes of prompt cache misses. Reuse depends on a matching request prefix: changing details placed before stable instructions or nondeterministic tool and file ordering can disrupt it. The investigation also found caching options that failed to send required provider controls, and Anthropic compatibility endpoint traffic that did not support caching; native Messages API traffic showed cache reads. The team proposed sorting tools and files, putting stable instructions first, passing cache controls through, and checking outgoing requests and later cache reads. Evaluation should compare the same routes and task types before and after, including task cost, cache reads after tool calls, latency, accuracy, retries, fallback calls, and successful tool use. Token estimates help explain behavior, but provider bills are needed to confirm savings; cache writes can add cost, and longer histories can increase input volume. ### Source excerpt An investigation of prompt cache misses in internal AI applications by using Datadog Cloud Cost Management and Agent Observability. ## What's new with Grafana variables: less clutter, easier reuse across dashboards, and more DevFeed: [What's new with Grafana variables: less clutter, easier reuse across dashboards, and more]() Original publisher: [Read original article]() Author: Haris Rozajac Published: 2026-11-05T19:25:32Z Content type: release Language: en Sources: [Grafana Labs]() Topics: [Grafana](), [dashboards]() Tags: [analytics](), [annotation](), [api](), [api-gateway](), [api-platform](), [dashboards](), [database](), [grafana](), [grafana-cloud](), [logs](), [metrics](), [monitoring](), [observability](), [query](), [reuse](), [template](), [traces](), [variable](), [variables]() ### AI overview Grafana adds section-level variables (generally available) and cross-dashboard variables (public preview) to reduce dashboard clutter and repeated maintenance. Variables scoped to a row or tab apply only to its panels; Grafana resolves them from the nearest section outward to parent sections and then the dashboard. Sections can also have their own Filter and Group by controls, and variables move up a level if a section is ungrouped. Cross-dashboard variables can be global or folder-scoped. Their definition is shared, while each dashboard retains its own selected value; dashboards must opt in to specific variables or all variables in a scope. Local dashboard variables override global ones, and section variables can override both. Cross-dashboard definitions are managed under Dashboards > Variables and are read-only in dashboard edit mode. In public preview, they are separate v2beta1 Variable API resources, manageable through Grafana's API or Terraform's grafana_apps_generic_resource; dashboards can also opt in through configuration. ### Source excerpt Template variables are one of the most used features in Grafana, providing users with a dropdown to turn one dashboard into hundreds: pick a different cluster, region, or service, and every panel updates. For most of Grafana's history, variables had a single home at the top of the dashboard. Each dashboard had one flat list of variables, and that list was shared by every panel on it. But as dashboards grew in size and number, those controls became harder to organize and maintain. The top of a dashboard could get crowded with dropdowns that only applied to certain panels, while shared controls were copied across dashboards and had to be updated individually. To address these challenges, we've rolled out two new features in Grafana: Section-level variables for rows and tabs (generally available): these variables let you place controls near the panels they affect, reducing clutter at the top of the dashboard. Cross-dashboard variables (public preview): these are variables you can define once and reuse across dashboards, keeping configurations consistent and easier to maintain. In this post, we'll take a closer look at each variable type, its use cases, and how to get started. Section-level variables: define and manage variables by row or tab Section-level variables let you define variables and filters for a specific row or tab, so they apply only to the panels inside that section. Placing a control next to the panels it affects makes it clear what a selection will change, and it keeps the top of the dashboard uncluttered. Let's say you want two $instance dropdowns, one for your API gateway panels and one for your database panels. Traditionally, with one flat namespace, you'd end up with names like $api_instance and $db_instance, or you'd split the dashboard into two. Now, with section-level variables, the API gateway section and the database section can each have its own $instance variable with its own query and selected value. Both sections still share the dashboard t ## Partner Summit 26: Building Trusted AI at Scale, Together DevFeed: [Partner Summit 26: Building Trusted AI at Scale, Together]() Original publisher: [Read original article]() Author: Shannon Leininger Published: 2026-11-05T19:00:28Z Content type: article Language: en Sources: [Cisco Blogs]() Topics: [Cisco](), [ai security](), [observability](), [Artificial Intelligence](), [trust](), [Security](), [AI-Ready Data](), [Managed Services]() Tags: [adoption](), [agent](), [agentic](), [agentic-ai](), [ai](), [ambition](), [building](), [business](), [cisco-partner-summit](), [customers](), [era](), [gap](), [observability](), [partner](), [partners](), [roi](), [scale](), [security](), [summit](), [trust](), [trusted-ai]() ### AI overview Cisco's Partner Summit message is that scaling agentic AI requires security, observability, and visibility into costs and returns. Cisco AI Defense protects interactions between AI models and enterprise networks, while the Splunk-powered Cisco Data Fabric provides data context and governance visibility into agent behavior, performance, and compliance. Together, they are presented as a foundation for observing and governing AI at scale, with human oversight. The text also identifies partner growth paths: Blackbelt learning and workshops, co-selling across Cisco's portfolio, advisory and managed services for visibility, cost management, and security governance, and locally hosted solutions for data residency and regulatory needs. The forecast that agents will become mainstream in two to three years is presented as an expectation. ### Source excerpt At Partner Summit 26, partners demonstrated their readiness to turn AI ambition into business outcomes. By prioritizing security, observability, and ROI, we're closing the "trust gap" and helping customers thrive in the agentic AI era. ## Building Data Observability for Banking Pipelines DevFeed: [Building Data Observability for Banking Pipelines]() Original publisher: [Read original article]() Author: Sunilkumar Reddy Eraganeni Published: 2026-11-05T17:39:30Z Content type: article Language: en Sources: [HackerNoon]() Topics: [data observability](), [Data Quality](), [Data pipelines](), [data-engineering](), [observability]() Tags: [3](), [about](), [added](), [alert](), [alerting](), [banking](), [banks](), [building](), [can](), [data](), [data-engineering](), [data-governance](), [data-lineage](), [data-observability](), [data-pipeline](), [data-quality](), [data-reconciliation](), [green](), [here](), [numbers](), [observability](), [pipelines](), [regulatory](), [regulatory-reporting](), [reports](), [ship](), [show]() ### AI overview Banking pipelines can report success while delivering stale, incomplete, or skewed data, so observability should check the data at each processing boundary, not just whether jobs ran. The article recommends tracking freshness by business date, calendar-aware volume, schema changes, distributions, and lineage; reconciling balances to source is a practical way to catch dropped records. It suggests starting with one regulatory report, adding freshness, volume, and reconciliation checks, assigning named owners, and running alerts in warning-only mode for two weeks before making critical checks block publication. Blocking regulatory outputs versus publishing with a warning depends on deadline tradeoffs. These checks cannot detect fresh, consistent data that is wrong because business logic misapplies a rule; that requires code review, testing, and domain expertise. The SQL volume-check example is explicitly a sketch that does not account for holidays. ### Source excerpt Pipelines that show green can still ship wrong numbers. Here is how banks can catch silent data failures before regulatory reports break. ## Groundcover automates Kubernetes sizing. Engineers set the bounds. DevFeed: [Groundcover automates Kubernetes sizing. Engineers set the bounds.]() Original publisher: [Read original article]() Author: Meredith Shubel Published: 2026-11-05T17:14:41Z Content type: news Language: en Sources: [Kubernetes Overview, News and Trends | The New Stack](), [The New Stack]() Topics: [Cloud Native Operations](), [kubectl](), [linkerd]() Tags: [agent](), [ai](), [automate](), [automated](), [automation](), [autonomous-infrastructure](), [autoscalers](), [balance](), [behavior](), [beta](), [capacity](), [challenge](), [change](), [cpu](), [deployments](), [finops](), [groundcover](), [infrastructure](), [k8s](), [kubernetes](), [memory](), [observability](), [post](), [sponsor-groundcover](), [sponsored](), [sponsored-post]() ### AI overview Groundcover's beta uses production telemetry from its eBPF sensor and Wand's optimization engine to recommend and continuously adjust CPU and memory requests for Kubernetes Deployments, weighing workload usage, traffic patterns, and node-level contention. Engineers can inspect recommendations and estimated savings before opting in; automation is off by default, and engineers set minimum and maximum request bounds. The beta changes requests, not limits, leaves HPA- or KEDA-managed resources unchanged, and uses in-place resizing where possible, with a rollout fallback. New Deployments remain in learning until enough data is available. It supports Deployments and sidecars, requires a supported sensor version, and is available at no additional cost to groundcover customers; other workload kinds are on the roadmap. Savings estimates use adjustable default list prices, and actual bill reductions depend on whether freed capacity reduces billed infrastructure. ### Source excerpt On Tuesday, groundcover launched Autonomous Infrastructure in beta to continuously adjust CPU and memory requests for Kubernetes (K8s) Deployments based The post Groundcover automates Kubernetes sizing. Engineers set the bounds. appeared first on The New Stack. ## I broke a working agent seven ways. The metric I was watching caught two of them. DevFeed: [I broke a working agent seven ways. The metric I was watching caught two of them.]() Original publisher: [Read original article]() Author: Kayode Ojo Published: 2026-11-05T15:00:00Z Content type: news Language: en Sources: [The New Stack]() Topics: [AI Agent](), [AI Infrastructure](), [Claude](), [mock api]() Tags: [adversarial](), [agent](), [ai-agents](), [andela](), [anomaly-detection](), [api](), [api-documentation](), [arm](), [asynchronous](), [audit](), [authorization](), [autonomous](), [behavior](), [benchmark](), [billing](), [claude](), [claude-haiku-5-5](), [metric](), [models](), [observability](), [post-contributed](), [software-testing](), [sponsor-andela](), [sponsored-post-contributed](), [support](), [tasks]() ### AI overview In a 19-task support-agent benchmark with two refund traps, completion scores missed most injected failures: only two of seven active regressions lowered task success below threshold. Tool-call volume and scope drift each detected four, while token cost detected one and refund traps detected none. Retries and workarounds masked flaky APIs and missing documentation; stale replica reads left every quantitative metric green, though the agent disclosed the issue in its final summary. The author recommends logging the resolved model ID, tracking tool calls per verified task and retries, parsing uncertainty in agent summaries, and refreshing adversarial tests after model updates. In this setup, baseline tool-call variance was 3% versus 18% for tokens, enabling tighter call-volume alerts. The benchmark is one system and its thresholds derive from three baseline runs. ### Source excerpt In September, I ran two models through the same 20 support tasks. Both scored 20 out of 20. Then I The post I broke a working agent seven ways. The metric I was watching caught two of them. appeared first on The New Stack. ## MASBench: Benchmarking LLM-based Multi-Agent Collaboration under Partial Observability DevFeed: [MASBench: Benchmarking LLM-based Multi-Agent Collaboration under Partial Observability]() Original publisher: [Read original article]() Author: Qizhi Chu, Zekai Yu, Sijie Wen, Yang Liu, Chen Qian, Cheng Yang, Chuan Shi, Zhiyuan Liu Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Multi Agent Systems](), [Large Language Model](), [llm-reasoning](), [test]() Tags: [about](), [abstract](), [agent](), [agents](), [also](), [arxiv](), [autonomous](), [autonomous-agents](), [based](), [benchmark](), [benchmarking](), [benchmarks](), [bridge](), [building](), [but](), [collaboration](), [cs-ai](), [llm](), [multi-agent](), [observability](), [under]() ### AI overview MASBench evaluates LLM-based multi-agent collaboration when agents have only partial access to the environment, addressing a limitation of benchmarks that assume global observability. It organizes evaluation across Reasoning, Scheduling, and Game tasks to assess Protocol, Memory, and Routing mechanisms. Deterministic metrics cover task performance, communication cost, and cost effectiveness, capturing outcomes alongside communication overhead. The abstract says experiments across LLM backbones and mechanism configurations provide empirical guidance, but gives no specific comparative results. ### Source excerpt arXiv:2610.04672v2 Announce Type: replace Abstract: Large language models (LLMs) have progressively evolved into the core of autonomous agents. Building on this progress, LLM-based multi-agent systems (MAS) coordinate multiple agents into a synergistic team to accomplish complex tasks that exceed the capabilities of individual agents. The effectiveness of such systems depends not only on the agents themselves, but also on how collaboration mechanisms are designed and organized. Note that real-world collaboration is typically partially observable, where each agent can only access partial information about the environment due to physical or privacy-related constraints. However, many existing multi-agent benchmarks assume global observability, and leave limited support for systematically evaluating collaboration mechanisms. To bridge this gap, we introduce MASBench, a multi-agent collaboration benchmark designed under partially observable constraints. It is organized into three progressive task categories: Reasoning, Scheduling, and Game. Through this structure, we progressively evaluate three representative collaboration mechanisms: Protocol, Memory, and Routing. MASBench further provides deterministic evaluation metrics, including performance score, communication cost, and cost effectiveness, to characterize both collaboration outcomes and communication overhead. Experiments across diverse LLM backbones and mechanism configurations offer empirical guidance for effective MAS design. Code is available at: https://github.com/BUPT-GAMMA/MASBench ## SysPilot: Cross-Library Performance Autotuning via Library Performance Contracts DevFeed: [SysPilot: Cross-Library Performance Autotuning via Library Performance Contracts]() Original publisher: [Read original article]() Author: Yineng Yan, Haowei Chen, Christopher J. Rossbach Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Library](), [Hyperparameter optimization]() Tags: [application](), [artifacts](), [bandwidth](), [benchmark](), [cpu](), [cross](), [cs-ai](), [cs-pf](), [gpus](), [library](), [memory-bandwidth](), [observability](), [performance](), [performance-metrics](), [presents](), [requirements](), [resource](), [resources](), [results](), [reuse](), [run](), [runtime]() ### AI overview SysPilot addresses cross-library resource contention by representing each library's resource demands, assumptions, controls, and observable signals in a Library Performance Contract written in LPCL. It synthesizes contracts from documentation and code, combines them for an application, proposes resource-conflict hypotheses, profiles to validate them, and selects candidate configurations. The contracts are intended to let performance knowledge transfer across applications using the same libraries. On a benchmark suite of cross-library contention scenarios, SysPilot reportedly matched autotuners using application-specific manual information and retained cross-application reuse. Against OpenTuner and SMAC3, it improved benchmark performance metrics by up to 38% beyond their best-found configurations while using three times fewer runs. The supplied text is an abstract and does not provide further evaluation details. ### Source excerpt arXiv:2611.06078v1 Announce Type: cross Abstract: Modern scientific and machine-learning applications compose many independently developed libraries. When these libraries run together, their independently managed runtimes can oversubscribe shared resources: using the same CPU cores, fighting over GPUs, or contending for memory bandwidth. Avoiding such contention requires careful configuration across library boundaries. Existing autotuners do not directly investigate what resources are shared between libraries, which runtime signals correlate with interference. As a result, they are inefficient for finding optimal configurations to mitigate contention. This paper presents SysPilot, a cross-library performance autotuning system based on Library Performance Contract (LPC). A Library Performance Contract captures a library's resource demands, assumptions, controls, and observability signals in LPCL, a specification language for describing library resource requirements. SysPilot synthesizes contracts from documentation and code artifacts, composes contracts from the libraries used by an application, derives resource-conflict hypotheses, profiles the application to validate those hypotheses, and selects candidate configurations. This design makes the Library Performance Contract a reusable unit of performance knowledge: a contract for a library can be reused across many applications that depend on that library. We evaluate SysPilot on a benchmark suite of cross-library contention scenarios. The results show that SysPilot matches the performance of autotuners that rely on application-specific manual information while retaining cross-application reuse. Compared with general-purpose autotuners such as OpenTuner and SMAC3, SysPilot improves benchmark performance metrics by up to an additional 38% over the best configuration found by these autotuners, while using 3x fewer runs. ## When Does a Cross-Device Run Justify PASS? Obligation-Level Evidence Qualification under Partial Observability DevFeed: [When Does a Cross-Device Run Justify PASS? Obligation-Level Evidence Qualification under Partial Observability]() Original publisher: [Read original article]() Author: Zimu Liu, Xiaoxi Guo, Xiaolu Tang, Liang Yu Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Software Engineering]() Topics: [test](), [Test automation](), [audit]() Tags: [association](), [atomic](), [business](), [characterization](), [closure](), [cross](), [cs-se](), [device](), [evaluation](), [field](), [flat](), [graph](), [input](), [isolates](), [matching](), [observability](), [over](), [pass](), [policy](), [post](), [run]() ### AI overview The evidence contract qualifies support before evaluating business predicates by binding each obligation to admissible subjects, episodes, objects, and observation scopes, and recording unresolved support. In a 12-seed controlled study of 2,304 claims with identical records and policy, Atomic-Facts Flat produced 160 false PASS judgments and 76.30% PASS recall. Adding computed associations reduced false PASS to five and raised recall to 99.09%; Deterministic Closure achieved zero false PASS and 100% recall, outperforming Atomic Flat in 11 seeds with one tie. The authors attribute most benefit to explicit association rather than graph notation. A separate characterization across four workflows, 40 physical episodes, and 80 claims found 95% accuracy and 100% PASS recall (27/27), with three false PASS judgments. These field results are descriptive: a policy-input mismatch blocked the planned matched-policy comparison, and post-hoc sensitivity did not establish superiority over a policy-informed textual baseline. ### Source excerpt arXiv:2611.06013v1 Announce Type: new Abstract: A cross-device test can complete every command yet judge a business claim from evidence belonging to another device or attempt. We introduce an executable evidence contract that makes support qualification an explicit stage before predicate evaluation. It binds each obligation to admissible subjects, episodes, objects and observation scopes, and records unresolved support rather than treating an apparently successful observation as proof. We evaluate two implemented parts of the contract. A prospectively specified, 12-seed controlled study isolates association under reused identifiers across 2,304 claims. Given identical records and business policy, Atomic-Facts Flat produces 160 false PASS and 76.30% PASS recall; supplying computed associations reduces false PASS to five and increases PASS recall to 99.09%. Deterministic Closure produces zero false PASS and 100.00% PASS recall, improving on Atomic Flat in 11 seeds with one tie (two-sided sign p=0.00098). Most of the benefit comes from explicit association rather than graph notation. A separate prospective field characterization spans four workflows, 40 physical episodes and 80 claims. Against frozen reference labels, the physical-evidence verifier attains 95.00% accuracy and 100.00% PASS recall (27/27), with three false PASS judgments. The field results are descriptive: a policy-input mismatch prevents the planned matched-policy comparison, and post-hoc sensitivity does not establish superiority over a policy-informed textual baseline. Together, the studies characterize qualification on physical evidence and isolate the value of explicit association under a controlled matching contract. ## Manage Kubernetes observability with DatadogInstrumentation DevFeed: [Manage Kubernetes observability with DatadogInstrumentation]() Original publisher: [Read original article]() Author: Mathew Estafanous; Wassim Dhif; David Lentz Published: 2026-11-05T00:00:00Z Content type: tutorial Language: en Sources: [Datadog]() Topics: [observability](), [tracing]() Tags: [agent](), [apm](), [collection](), [configuration-management](), [configure](), [containers](), [datadog-agent](), [gitops](), [how-to](), [http](), [kubernetes](), [log-management](), [logs](), [monitoring-as-code](), [observability](), [profiling](), [tracing]() ### AI overview DatadogInstrumentation lets Kubernetes workload owners manage checks, logs, and APM settings in one resource while platform teams retain control of Agent deployments. The Cluster Agent validates configuration at apply time and reports readiness conditions. In the checkout example, an HTTP check, log settings, and Java tracing are moved out of pod annotations and Agent targeting rules; check and log updates take effect without an application rollout or Agent restart. The team later adds OpenMetrics metrics and email masking the same way. Tracing changes, including enabling profiling through the Java tracer, apply to newly created pods and therefore require a rollout to replace existing pods. The described setup requires Agent and Cluster Agent v7.85 or later, plus Operator v1.31 or later or Helm chart v3.255.0 or later. ### Source excerpt Configure Kubernetes checks, logs, tracing, and profiling with DatadogInstrumentation while keeping workload changes separate from Agent deployments. ## Cyber ups and downs: Datadog falls 14% on outlook tied to one customer as Rapid7 jumps and Fastly slides DevFeed: [Cyber ups and downs: Datadog falls 14% on outlook tied to one customer as Rapid7 jumps and Fastly slides]() Original publisher: [Read original article]() Author: Duncan Riley Published: 2026-11-04T23:07:11Z Content type: news Language: en Sources: [SiliconANGLE]() Topics: [datadog](), [AI in cybersecurity]() Tags: [agentic-soc](), [ai](), [ai-firewall](), [annualized-recurring-revenue](), [artificial-intelligence](), [black-hat-usa](), [cloud](), [cyber](), [cybersecurity](), [datadog](), [david-obstler](), [earnings](), [edge-cloud](), [fastly](), [guidance](), [net-retention](), [news](), [observability](), [olivier-pomel](), [outlook](), [prompt-injection](), [q3-2026-earnings](), [rapid7](), [remaining-performance-obligations](), [restructuring](), [stock-market](), [the-latest](), [third-quarter](), [wael-mohamed]() ### AI overview Datadog, Rapid7 and Fastly all beat third-quarter analyst estimates, but their shares diverged. Datadog reported 35% revenue growth and raised its full-year guidance, yet its shares fell about 14% after it forecast fourth-quarter revenue near the prior quarter's level, assuming its largest, unnamed AI customer spends only its contractual minimum. Rapid7's revenue declined 2.5%, but earnings beat estimates, annualized recurring revenue exceeded its prior guidance, and the company raised its earnings outlook; shares rose about 18% after hours. Fastly posted 20% revenue growth, led by security sales growing twice as fast, and raised full-year revenue guidance, but shares fell about 14% after the bell. The excerpt attributes no specific cause to Fastly's share decline. ### Source excerpt Datadog Inc., Rapid7 Inc. and Fastly Inc. all beat Wall Street's third-quarter forecasts today, but Rapid7 was the only one of the three to see its shares rise. A cautious outlook tied to one large customer knocked about 14% off Datadog's stock in regular trading. Rapid7 jumped 18% after hours on results that came despite [...] The post Cyber ups and downs: Datadog falls 14% on outlook tied to one customer as Rapid7 jumps and Fastly slides appeared first on SiliconANGLE. ## Fortra says AI could save file transfer admins up to 160 hours a year DevFeed: [Fortra says AI could save file transfer admins up to 160 hours a year]() Original publisher: [Read original article]() Author: Matthew Burns Published: 2026-11-04T17:53:47Z Content type: news Language: en Sources: [The New Stack]() Topics: [Artificial Intelligence](), [Data transmission]() Tags: [ai](), [ai-operations](), [devops](), [diagnostics](), [file-transfer](), [fortra](), [observability](), [operations](), [sponsor-fortra](), [sponsored-webinar](), [webinar]() ### AI overview Fortra estimates that troubleshooting accounts for up to 160 of the 571 hours per user per year its AI features could save on routine managed file transfer tasks. In one example, an admin spent 40 of 50 minutes investigating a failed settlement-file transfer caused by a trading partner rotating SSH keys; the repair took 5 to 10 minutes. Guided Diagnostics, available in GoAnywhere MFT, reads a failed job and ranks likely causes and next steps, leaving diagnosis and any changes to the admin. Fortra expects the most help with recurring errors that leave clear messages. The feature is less useful for intermittent network drops, partner-side problems, complex workflow failures, performance bottlenecks, and custom integrations. The estimate excludes time waiting for a partner response, and the article does not provide the underlying customer survey details. ### Source excerpt Fortra makes software that moves files between companies on a schedule and says AI could give admins back up to The post Fortra says AI could save file transfer admins up to 160 hours a year appeared first on The New Stack. ## Datadog vs. New Relic Pricing: What Actually Drives Your Bill DevFeed: [Datadog vs. New Relic Pricing: What Actually Drives Your Bill]() Original publisher: [Read original article]() Author: John Blust Published: 2026-11-04T12:00:00Z Content type: comparison Language: en Sources: [New Relic]() Topics: [observability](), [datadog](), [OpenTelemetry](), [digital experience monitoring]() Tags: [17](), [2](), [2025](), [2026](), [about](), [account](), [add-ons](), [bill](), [can](), [commit](), [compare](), [cost](), [data](), [datadog](), [devops](), [new-relic](), [observability](), [pricing](), [size](), [users](), [vs]() ### AI overview Datadog bills observability through separate product units: infrastructure and APM per host, logs by both ingestion and indexing, and additional charges for features such as synthetic monitoring, real user monitoring, and containers above plan limits. New Relic centers billing on data ingest and user type; it includes 100 GB of monthly ingest and unlimited Basic users at no charge, while deeper platform access is billed by user on paid editions. Capabilities generally do not add separate SKUs, though Data Plus and Advanced Compute can add costs. The article says data growth affects both models, while host and container growth directly affect Datadog and user counts matter more in New Relic. It recommends modeling likely changes in data, infrastructure, users, retention, and add-ons, and asking vendors how spikes, overages, and bundled features are billed. Rates change, so use current vendor pricing for estimates. ### Source excerpt Compare how Datadog and New Relic actually price observability -- data, hosts, and users -- so you can size your real cost before you commit. ## Reconstruction-Based Diagnosis through G\\"odel-Regularized Virtual Observability DevFeed: [Reconstruction-Based Diagnosis through G\\"odel-Regularized Virtual Observability]() Original publisher: [Read original article]() Author: Oliver Niggemann, Jonas Ehrhardt, Alexander Diedrich Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [observability](), [AI Foundation Models]() Tags: [cs-ai](), [data](), [diagnosis](), [isolation](), [logic](), [networks](), [neural](), [observability](), [observer](), [plants](), [production](), [root](), [sensors](), [tennessee](), [training](), [virtual]() ### AI overview The approach isolates faults that sensors detect only after their effects propagate by reconstructing unmeasured, fault-sensitive variables as virtual sensors. A neural observer trained only on normal data uses available measurements to predict normal behavior; deviations yield local symptoms, whose onset times are compared to identify root causes. Gödel logic encodes physical knowledge to guide training. The authors evaluate the method on synthetic networks, the Tennessee Eastman Process, and datasets from two modular production plants. The supplied text is an abstract, so it does not report comparative results or detailed limitations. ### Source excerpt arXiv:2611.04487v1 Announce Type: new Abstract: Not every component in a technical system can be equipped with sensors. Hence, when a fault occurs, available sensors often detect it only after its effects have propagated. This makes the isolation of faults difficult. We approach this problem by introducing virtual sensors. For each combination of component and fault type, we let external knowledge describe the fault-sensitive variables. We introduce a novel neural observer that is only trained on normal data. For fault isolation, it uses available sensor measurements where it is possible and reconstructs missing variables as virtual sensors to predict normal behavior. It detects deviations between current values and nominal predictions and computes local symptoms, which are then compared by onset time. The result is a set of root causes describing the faulty behavior. We use G\"odel logic to encode physical knowledge and guide training. We evaluate our approach on synthetic networks, the Tennessee Eastman Process, and datasets of two modular production plants. ## Flow Equivariant World Models: Structured Memory for Dynamic Environments DevFeed: [Flow Equivariant World Models: Structured Memory for Dynamic Environments]() Original publisher: [Read original article]() Author: Hansen Jin Lillemark, Benhao Huang, Fangneng Zhan, Yilun Du, Thomas Anderson Keller Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [World models](), [Representation learning]() Tags: [3d](), [architectures](), [benchmarks](), [cs-ai](), [cs-cv](), [cs-lg](), [diffusion](), [environments](), [flow](), [framework](), [memory](), [modeling](), [models](), [motion](), [observability](), [prediction](), [recurrent](), [streams](), [structure](), [structured](), [temporal](), [world](), [world-models]() ### AI overview Flow Equivariant World Modeling uses latent memory that shifts and transforms with self-motion and inferred motion of external objects, keeping representations of out-of-view regions aligned as the environment evolves. The authors present this structure as a way to address partial observability and improve stable, long-horizon dynamics prediction. They report advantages over diffusion, memory-augmented, and recurrent world-model architectures on 2D and 3D partially observed video benchmarks. The supplied abstract does not specify benchmark metrics or the size of the reported gains. ### Source excerpt arXiv:2601.01075v3 Announce Type: replace Abstract: Embodied systems experience the world as 'a symphony of flows': a combination of many continuous streams of sensory input coupled to self-motion, interwoven with the dynamics of external objects. These sensory streams and the underlying dynamics of the world obey smooth, time-parameterized symmetries which existing world models ignore. Without a memory that respects this structure, partial observability presents a major obstacle to existing methods: each observation reveals only a fraction of the world, while unobserved regions continue to evolve. In this work, we introduce Flow Equivariant World Modeling, a framework that leverages time-parameterized symmetries within a latent memory for stable and accurate dynamics prediction over long horizons. The latent memory shifts and transforms equivariantly with self-motion and inferred external object motion, keeping information about out-of-view regions aligned as time progresses. We demonstrate the advantage of this framework over state-of-the-art diffusion, memory-augmented, and recurrent world model architectures on 2D and 3D partially observed video world modeling benchmarks. More broadly, our results suggest that predictive representations become more powerful when they are organized in line with the temporal and dynamical structure of the world they model. Project page: https://flowequivariantworldmodels.github.io/ ## RTSGameBench: Evaluating VLMs for Strategic Reasoning in Game Environments DevFeed: [RTSGameBench: Evaluating VLMs for Strategic Reasoning in Game Environments]() Original publisher: [Read original article]() Author: San Kim, Daechul Ahn, Reokyoung Kim, Hyeonbeom Choi, Seungyeon Jwa, Jonghyun Choi Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [llm-reasoning](), [Artificial Intelligence](), [Agent Harness]() Tags: [adapting](), [agent](), [agent-harness](), [agents](), [benchmarks](), [coverage](), [cs-ai](), [environments](), [evaluation](), [experience](), [framework](), [game](), [games](), [harness](), [memory](), [models](), [multi-agent](), [observability](), [planning](), [reasoning](), [vlms]() ### AI overview RTSGameBench evaluates strategic reasoning in multi-agent real-time strategy settings using Beyond All Reason. It combines full-game matchups, diagnostic mini-games focused on distinct strategic demands, and a self-evolving framework that converts free-form queries into new mini-games. The authors also introduce RTSGameAgent, which combines group-level trajectory planning with short- and long-term memory and outperforms existing agents. They report that state-of-the-art VLMs struggle particularly with ally coordination and multiple opponents, where humans succeed. The evidence provided is an abstract, so it does not specify evaluation details or quantitative results. ### Source excerpt arXiv:2606.18950v4 Announce Type: replace Abstract: Vision-Language Models (VLMs) often struggle with strategic reasoning in a multi-agent setting, i.e., deciding under uncertainty while coordinating with and adapting to other agents. Real-time strategy (RTS) games can be a natural testbed for this capability, requiring long-horizon planning, coordination with allies, and adaptation to opponents under partial observability. But existing RTS benchmarks cover limited agent configurations, lack controlled evaluation across distinct strategic demands, and remain confined to fixed scenarios. To address these gaps, we propose RTSGAMEBENCH, built on Beyond All Reason, an RTS game whose larger scale broadens the strategic space beyond prior testbeds. In RTSGAMEBENCH, we offer holistic evaluation across diverse full-game matchups, diagnostic mini-games each emphasizing a focal strategic demand, and extensible coverage via a self-evolving framework that turns free-form queries into new mini-games, improving with accumulated experience. We further propose RTSGAMEAGENT, an agent harness that combines scalable group-level trajectory planning with short- and long-term memory and outperforms existing agents. We find that state-of-the-art VLMs struggle where humans succeed, particularly when they must coordinate with allies or contend with multiple opponents. ## Recovering Geodesic Distances from Noisy Observations DevFeed: [Recovering Geodesic Distances from Noisy Observations]() Original publisher: [Read original article]() Author: Charles Fefferman, Jonathan Marty, Kevin Ren Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [graph theory]() Tags: [boundary](), [bounded](), [cs-lg](), [data](), [distribution](), [error](), [graph](), [math-dg](), [math-pr](), [missing-data](), [noise](), [novel](), [observability](), [observer](), [probability](), [random](), [recover](), [runtime](), [sample](), [space](), [split](), [stat-ml](), [visibility](), [volume]() ### AI overview The paper gives quantitative guarantees for recovering geodesic distances from noisy observations in two settings. With an i.i.d. sample of size about ε^(−2d−2)|log ε|, it recovers all sampled pairwise distances to additive error O(ε|log ε|) with high probability and runtime O(N^(2.5+o_d(1))). The noise may vary heterogeneously, provided it is bounded in subgaussian norm and has mean F of the true distance, where F is unknown, increasing, and bilipschitz; a "ruler-building" procedure learns F. The result also extends to randomly missing distances. For partial observability, where observed points lie in a visibility-constrained observer subset and target points are sampled from the space, it recovers target-to-target distances to O(ε) using |Z| about ε^(−2d−2)|log ε|, |Y| about ε^(−max{dν,4}), and runtime O(|Z|³). ### Source excerpt arXiv:2511.13025v2 Announce Type: replace-cross Abstract: Let $(M,d,\mu)$ be a nice geodesic probability space (such as a manifold) and $X=\{x_i\}_{i\in [N]} \subset M$ a finite $\mu$-i.i.d. sample. Suppose we observe noisy distance random variables $d'(x_i,x_j)$. When can the true distances be recovered to small additive error? This question has attracted interest in two disparate fields: latent-space graph models and boundary-value inverse problems. We first show that, taking $N \asymp \epsilon^{-2d-2}|\log\epsilon|$ and using runtime $O(N^{2.5+o_d (1)})$, we can recover all distances in $X$ to additive error $O(\epsilon|\log\epsilon|)$ with high probability, assuming $\mathbb{E} d'(x_i, x_j) = F(d(x_i,x_j))$ for an unknown, increasing bilipschitz $F$ and with the $d'(x_i,x_j)$ allowed to be arbitrarily heterogeneous in their distribution as long as they are bounded in subgaussian norm. This generalizes existing work, which required $F$ to be known and made strict assumptions on the distribution of $d'(x_i, x_j)$. Achieving this generalization required a novel ``ruler-building'' procedure to learn $F$. We extend our results to a missing data case where we randomly lose access to noisy distances. Our second result uses ideas from the first to tackle a partial observability case: $X = Y \cup Z$ with $Z$ sampled from $M$ and $Y$ sampled from an observer subset satisfying visibility conditions that model the assumptions made on the boundary in boundary-value inverse problems. In this setting, we recover all pairwise distances in $Z$ to additive error $O(\epsilon)$ with $|Z| \asymp \epsilon^{-2d-2}|\log\epsilon|$, $|Y| \asymp \epsilon^{-\max\{d_\nu,4\}}$ with $d_\nu$ determined by the volume-growth properties of the probability measure on the observer set, and runtime $O(|Z|^3)$. This is, to our knowledge, the first quantitative guarantee for this observer/observed split under noise, which complements existing stability results for boundary data. [Next page]()