VAREXIS-INDEX · Free decision guide
Agent Run Cost & Trace Evidence
An agent run has a trace and a cost total, but sampled spans, price mapping, evaluator limits or exporter routing may leave the apparent result incomplete.
For: Agent platform and FinOps teams reviewing declared run costs, trace coverage and export completeness.
Your decision: Decide which declared trace and usage totals are fit for operational review, and which require missing-span, pricing or routing evidence before reconciliation.
Evidence to bring
The trace and observation types; model identifiers and usage buckets; price-table version; sampling and exporter configuration; evaluator spend settings; project routing and flush outcomes. Use redacted declarations rather than credentials or private prompts.
- Identify model, usage and price inputs and separate observed from inferred cost.
- Check sampling, evaluator caps and project-routing scope.
- Reconcile completed export evidence and disclose missing spans or unsettled totals.
What this evidence supports
LangSmith and Langfuse documentation defines where displayed cost comes from, what evaluator limits control, and how trace sampling and routing affect recorded evidence. Existing OpenTelemetry and flush reviews supply complementary completion boundaries.
What remains unestablished
A trace is not a provider invoice, settled charge or proof of successful action. These records do not verify live telemetry, recover dropped spans, enforce a spending limit or guarantee complete cost attribution.
Read the source evidence
These 9 existing records are free. Review the source version and limitations before applying a finding to your decision.
Does a recorded LangSmith LLM trace necessarily have a calculable cost
Require usage, model identity and a matched price definition before relying on computed cost.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGSMITH-COST-REQUIRED-USAGE-MODEL-PRICE-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGSMITH-COST-REQUIRED-USAGE-MODEL-PRICE-01-v1- SHA-256
2e2b9a08781eabe886c07392dcc5f1d913a152c64996f26d06e158bb5efe48bc
Does reaching a LangSmith evaluator budget stop the underlying agent
Distinguish evaluator spend controls from limits on the evaluated application.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGSMITH-EVALUATOR-CAP-NOT-AGENT-STOP-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGSMITH-EVALUATOR-CAP-NOT-AGENT-STOP-01-v1- SHA-256
b7ef13143a90c534ff14e02abd1a911788db3fdb20a9d192345c28a88618aec1
Does adding a thread identifier only to the parent preserve all child cost totals
Check child-run grouping before accepting thread totals.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGSMITH-THREAD-CHILD-METADATA-COST-OMISSION-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGSMITH-THREAD-CHILD-METADATA-COST-OMISSION-01-v1- SHA-256
9f66c7051fb7142cb26f146c4221d18e0b6588b2a3f559362e149a232bf0c1aa
Does changing a Langfuse model price necessarily change an explicitly supplied cost
Determine whether cost was supplied directly or inferred.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGFUSE-INGESTED-COST-PRECEDES-INFERENCE-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGFUSE-INGESTED-COST-PRECEDES-INFERENCE-01-v1- SHA-256
5ea52e67ac1735a6a9706578e13a2a6fbee2015daf3c4a35c4e332515f9c6f33
Can cached input tokens be counted both within input and in a separate Langfuse usage bucket
Check that supplied usage buckets do not overlap.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGFUSE-USAGE-BUCKETS-OVERLAP-DOUBLE-COUNT-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGFUSE-USAGE-BUCKETS-OVERLAP-DOUBLE-COUNT-01-v1- SHA-256
645b494250ac036146dfe89ac54c66cc65dd194ad628f4f5c54fc5ca472f146e
Can a masking error cause Langfuse telemetry to disappear
Account for export loss when a masking hook fails.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGFUSE-MASKING-FAILURE-EXPORT-LOSS-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGFUSE-MASKING-FAILURE-EXPORT-LOSS-01-v1- SHA-256
21349de1e441eadc473597843d44cfca1fa735e46aa3094f014a5d239ac08d24
Can shared background work be attributed to the wrong Langfuse project
Reconcile request-level project routing with shared export queues.
Original source · Reviewed 2026-10-30. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp/catalogue- Tool
get_publication_record- Arguments
{"record_id":"VXI-EXT-ICE08-LANGFUSE-SHARED-QUEUE-REQUEST-PROJECT-MISATTRIBUTION-01-v1"}- Record identifier
VXI-EXT-ICE08-LANGFUSE-SHARED-QUEUE-REQUEST-PROJECT-MISATTRIBUTION-01-v1- SHA-256
55bf8978823da8209e961a77cc11b1e6003f9d2f1b0e609f3ec5795f88f004a5
LangSmith: Function completion can precede trace delivery
Use the existing serverless flush review to qualify export completion.
Original source · Reviewed 2026-10-18. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp- Tool
preview_evidence_answer- Arguments
{"route_id":"agentic_reviewed_langsmith_serverless_trace_flush_completion"}- Record identifier
VXI-EXT-PROCUREMENT-LANGSMITH-SERVERLESS-TRACE-FLUSH-COMPLETION-01-v1- SHA-256
63e30be64d4dfca30fd9bffba25ebdb3d6330e0dc5fedbf385f1efd9e8519ccd
OpenTelemetry ForceFlush timeout and outcome
Use the existing timeout review to distinguish a flush request from acknowledged completion.
Original source · Reviewed 2026-10-24. Full provenance and limitations are retained in the evidence record JSON.
Exact agent access and integrity
- Connector
https://varexis.tech/mcp- Tool
preview_evidence_answer- Arguments
{"route_id":"agentic_reviewed_otel_forceflush_timeout_outcome_boundary"}- Record identifier
VXI-EXT-PROCUREMENT-OTEL-FORCEFLUSH-TIMEOUT-OUTCOME-BOUNDARY-01-v1- SHA-256
746e905c3da2b6b43dfffffdf36c8fbdcbafac093a59ca25c283d4eb702c5c8d
Connect an evidence client
Configure the two existing MCP connectors, then use the endpoint and arguments shown for each record. The original /mcp provides its frozen search and previews; /mcp/catalogue provides bounded browsing and exact access to additional published records. They are separate tool contracts, not interchangeable connections.
Canonical decision-guide JSON: records, provenance, hashes and checker selection. Enhanced natural-language search remains on HOLD. Keep credentials and private declarations local; public MCP supplies evidence and does not run licensed analysis.
When a local checker helps
Audit Coverage can check supported supplied event-chain and completeness declarations. It does not import LangSmith or Langfuse exports, calculate provider bills, validate token prices or enforce evaluator budgets.
The guide helps identify missing evidence before relying on a trace or cost report. The existing checker provides repeatable local analysis only for its supported declarations. Optional saved evidence preserves an eligible existing source answer; it does not save this entire guide or execute a cost analysis.
Agent Audit & Evidence Coverage
£79 + VAT · One-time internal-use licence. Separate from saved evidence and subscriptions. VAT is confirmed at checkout.
Supplied inputs: JSON declarations of identity, authorisation, action and outcome events, invocation references, and capture, retention, export and recovery coverage for supported source profiles.
Local results: Structured event-chain gaps, conflicting references and review requirements, with source and content hashes.
Synthetic demonstration: An outcome cannot be connected to its declared parent event. Resolve the missing reference before relying on the chain.
Assesses supplied declarations only; no independent live-system testing, runtime enforcement, certification, verified deletion or settlement.
Read the free demonstration and licence terms · Review purchase — £79 + VAT · Existing product details
A free record supplies documented evidence. A licensed checker performs reusable local analysis of supported customer-supplied declarations. Full operational schemas, runnable examples and executable archives accompany the existing licensed package. These examples are concise synthetic excerpts, not complete operational inputs or analyses. No runtime enforcement, live-system verification or certification is supplied. Compare existing prices · Checkout and recovery terms.
Original evidence-pack section · Choose another decision guide
