v1.8
dashboard

The instrumented pipeline.

Anonymized, aggregate usage of opchain's own pipeline — which skills actually run, what tier they run on, and what they cost. Counts and sums only; no prompt content, no identity (see privacy). Metered locally and opt-in by oc-telemetry-ops, costed by oc-cost-ops.

Sample data. These are illustrative seed numbers, not a live aggregate yet — opt-in telemetry is rolling out. The page swaps to real figures once a live export is published. Shape: opchain-usage-aggregate/1.
Pipelines run
1,284
2026-05-01 → 2026-06-25
Skill runs
5,310
across 10 skill buckets
Avg cost / shipped feature
$7.41
attributed by oc-cost-ops

Most-used skills

Buckets with fewer than 5 runs fold into other (k-anonymity).

Model-tier mix

  • sonnet 52%
  • opus 31%
  • haiku 15%
  • fable 2%

Tier, never the full model id — all the distribution needs.

Eval-score trend

W18 W19 W20 W21 W22 W23 W24

Avg score across rubrics (oc-bug-check, oc-code-auditor, oc-prompt-ops), normalized 0–1.

Replays — real runs, with cost overlays

Pipeline runs recorded end-to-end, annotated with attributed spend + eval score.

Ship a RAG answer-bot

$9.12
oc-app-architect → oc-stack-forge → oc-rag-forge → oc-deploy-ops

Discovery → vector-DB pack pick (pgvector) → retrieval eval goldset → staging → prod. Cost overlay shows spec on Opus, build on Sonnet, bulk embedding-eval batched.

~1 week (part-time) eval 8.6/10

Reverse-spec a Django monolith, ship a change

$5.74
oc-reverse-spec → oc-app-architect → oc-bug-check → oc-deploy-ops

40k-line app, no docs → backfilled specs → one scoped feature → pre-commit gate → Render deploy. Most spend was the one-time reverse-spec sweep (Sonnet, high-throughput).

~1 afternoon eval 8.1/10

Build a triage agent on the Agent SDK

$11.38
oc-app-architect → oc-agent-forge → oc-prompt-ops → oc-deploy-ops

Subagent topology + tool budgets + a capped dedupe loop, gated on an agent eval goldset. Cost overlay flags the eval suite as the largest line — batched to halve it.

~3 days eval 8.4/10

How this is measured