DSPy Program Evaluation Telemetry

Implementation evidence

DSPy Program Evaluation Telemetry: from boundary to verified row

Use DSPy Program Evaluation Telemetry at a controlled application boundary, keep the event contract small, and verify a known outcome before building aggregate views.

1
Choose the outcome
DSPy program reliability
2
Define the contract
operation_id, program_name, program_version, model_alias, optimizer_name, and release
3
Instrument the boundary
Wrap the public DSPy module call for production outcomes and emit evaluator results separately after dspy.Evaluate completes.
4
Verify the evidence
Exercise a known fixture, then inspect dspy_program_completed for one correctly typed terminal row.

Before you start

Prerequisites and boundaries

DSPy and telemetry-sh initialized in a trusted Python process
Stable program, model, optimizer, dataset, metric, and release versions
Separate event grains for live predictions and offline evaluation results

Delivery setup

Install and initialize server-side

Initialize Telemetry for synchronous code or TelemetryAsync for asyncio code once per process with a server-side key. Keep ingestion credentials out of browser bundles, client-visible environment variables, source control, logs, and exception messages.

dspy-install

pip installation

bash

python -m pip install telemetry-sh

1Prepare one reusable server-side delivery client with bounded network behavior.
2Add the outcome event at the success, failure, retry, or timeout boundary.
3Send controlled fixtures and inspect the stored rows before enabling an alert.

Snippet

Start with one structured event

Add this shape where the workflow completes, fails, or retries. Then build the dashboard from real fields.

dspy

DSPy Program Evaluation Telemetry event

python

from time import perf_counter

def run_support_program(program, question: str, operation_id: str):
    started_at = perf_counter()
    status = "success"
    error_type = None

    try:
        return program(question=question)
    except Exception as error:
        status = "failed"
        error_type = classify_program_error(error)
        raise
    finally:
        telemetry.log("dspy_program_completed", {
            "operation_id": operation_id,
            "program_name": "support_answer",
            "program_version": PROGRAM_VERSION,
            "model_alias": MODEL_ALIAS,
            "status": status,
            "error_type": error_type,
            "duration_ms": round((perf_counter() - started_at) * 1000),
            "release": APP_RELEASE,
        })

Event contract

operation_id, program_name, program_version, model_alias, optimizer_name, and release

status, duration_ms, estimated_cost_usd, accepted, and error_type

dataset_version, metric_name, metric_version, evaluation_score, passed, and evaluated_at for evaluation events

Implementation checkpoints

Checkpoint 1

Wrap the public DSPy module call for production outcomes and emit evaluator results separately after dspy.Evaluate completes.

Checkpoint 2

Do not compare optimizer candidates that used different datasets, metrics, judge models, thresholds, or evaluation coverage without separating those versions.

Checkpoint 3

Keep examples, predictions, traces, metric feedback, prompts, and model history in the approved DSPy or observability workflow rather than copying them into Telemetry.

Verification

Prove the event arrived

Run this after exercising known success and failure cases. Replace the fallback table name if your final event contract differs from the snippet.

dspy-verification

DSPy Program Evaluation Telemetry verification query

sql

SELECT *
FROM dspy_program_completed
ORDER BY timestamp_utc DESC
LIMIT 20;

Confirm one terminal row per logical outcome, with the expected status, identifiers, units, and UTC time.

Inspect the inferred schema and verify that retries do not change field types or generate a new logical event ID.

Search the stored fields for credentials, raw payloads, prompts, private content, and unbounded error messages.

Exercise a provider timeout, ingestion rejection, and process shutdown before treating the dashboard as complete.

Implementation references

Review the event contract, data-safety guidance, and upstream primary documentation before enabling a new production path.

Evaluate AI agents with SQL AI quality regression SQL Accepted output per dollar SQL DSPy metrics and evaluation DSPy debugging and observability DSPy tools with ReAct

Production boundary

Keep the outcome event small and recoverable

This pattern provides

A bounded, SQL-ready outcome beside the upstream workflow.
Stable fields for dashboards, alerts, and cross-event correlation.
A fixture-driven path for validating success, failure, retry, and timeout behavior.

This pattern does not provide

An OTLP exporter, automatic collection pipeline, or replacement for detailed traces and diagnostic logs.
Exactly-once delivery merely because the payload contains an event ID.
Permission to collect raw provider payloads, user content, credentials, or regulated data.

Delivery and idempotency Batching and shutdown Troubleshoot ingestion

Event schema starting points

Review the row grain, emit boundary, required types, privacy classes, example payload, and validation checklist before adapting a query or snippet to production.

AI operations10 fields