Sample dataThis is the real console UI over a fabricated tenant, “Ironwood Energy Systems,” a grid-storage builder that writes its own fleet software and ships an assistant to the utilities it serves. Every component below is the one customers use. The trains and runs are invented, so the ids here don’t open anything; in your console they do.

The console, exactly as your team would see it.

Four surfaces answer the four questions that matter: is anything broken, what went wrong, is quality holding, and what does it cost. The trains below are the work where being wrong is expensive: auditing your own code, vetting what customers send you, catching schedule risk before it lands, and grading the assistant you have already put in front of customers.

The departure board

Is anything broken right now?

Every train on one board, attention first, so a fault can never sit below the fold. The red eval alert means a quality check is failing even though runs are succeeding: that's the 'silent regression' the platform exists to catch.

Departure board· This month · so far

5 trains on the board

Aug 1 – Aug 18, 2026

Running

Attention

Eval alerts

Runs

103

Failure rate

Spend

$82.57

Status

Train

Last run

Recent

Runs

Failed

Median

Fault
ironwood-agent-review
agent_qaagent-answer-review@v0.8
15d28 runs2 failed
15d
28
7%
3m
Running
ironwood-doc-intake
doc_intakedoc-intake-review@v1.6
15d31 runs
15d
31
0%
6m
On time
ironwood-security-auditevals 1
security_auditsecurity-audit@v2.1
15d24 runs
15d
24
0%
11m
On time
ironwood-schedule-risk
schedule_riskschedule-risk@v1.3
15d20 runs
15d
20
0%
2m
No service
ironwood-commissioning
commissioningcommissioning-pack@v0.4
37d—0 runs
37d
—
—
—
—

Aug 1 – Aug 18, 2026 · trains run on demand, so the board shows the last departure, not a timetable

Run diagnosis

What went wrong overnight?

A failed run opens on the cause, not a log dump: the step timeline below plays back exactly which railroad car called which tool, and where it died. One click re-runs it once the upstream recovers.

Run

0199…0038

failedtrain ironwood-agent-reviewAug 17, 6:48 PM47scost $0.09

by schedule:nightly · scheduled

DocsGatewayError: manual lookup failed for 14 of 96 graded answers; upstream /v3/articles returned 502 three times (circuit opened after retry budget)

DocsGatewayError: manual lookup failed for 14 of 96 graded answers; upstream /v3/articles returned 502 three times (circuit opened after retry budget)

Step timeline

8 events

Phase · load-config+0.0s
Transcript Reader · started+2.0s
Transcript Reader · fetch_support_transcripts3.8s+4.0s
Transcript Reader · execute_select2.1s+9.0s
Manual Grader · started+14.0s
Manual Grader · docs_article_lookup9.4s+16.0s
Manual Grader · docs_article_lookup12.6sHTTP 502 from /v3/articles (attempt 3 of 3)+29.0s
ErrorDocsGatewayError: manual lookup failed for 14 of 96 graded answers (in docs_article_lookup)+43.0s

No silent regressions

Is quality holding, version over version?

Every run is scored against a test suite built for this business, and history is kept per check, so a regression shows up as a red column, not a support ticket. The amber flag marks a flickering check under review.

Regression history

suite v9 · last 16 evaluated runs

85%latest pass rate1 regressed·0 improvedvs previous evaluated run

Pass rate	100	100	100	92	100	100	100	100	92	100	100	100	92	85	92	85
L1No secret value in the report
L1Report matches declared JSON shape
L1Response is non-empty
L2Dependency manifest parsed
L2Every changed file was read
L2No write tool called, ever
L3Every finding cites file and line
L3Every finding is reproducible
L3Flags hardcoded credentials
L3No duplicate findings
L3Severity matches the CVSS band
L4Escalates exposure to a human
L4Summary reads plainlytracked

pass fail skipped not evaluated· each column is one evaluated run, oldest → newest

Eval scorecard

0199…0115 fail

L1 Structural3/3100%

L2 Behavioral3/3100%

L3 Outcome3/560%

L4 Quality2/2100%

Costed, not estimated

What does it cost, per run, in dollars?

Every model call and tool call is metered and attributed. Not credits, not tokens: dollars, per run, per train, with the pricing provenance to back it.

Cost

This month

Total spend

$82.57

≈ $0.80 / run · 103 runs

LLM

$75.20

MCP

$4.02

By workload

train-run$71.05eval$4.15connector$4.02infrastructure$3.35

Train	Share	Cost
ironwood-security-audit	58%	$48.20
ironwood-doc-intake	23%	$19.30
ironwood-agent-review	13%	$10.60
ironwood-schedule-risk	5%	$4.47

Priced via sample-2026-08 · authoritative

Spend by daysample month

Runs

recent runs · sample

Run	Train	Status	Who	Duration	Cost	When
0199…0041	ironwood-doc-intake	running	—	—	$0.00	Aug 18, 1:58 PM
0199…0040	ironwood-security-audit	succeeded	ci	11m 48s	$2.14	Aug 18, 1:02 PM
0199…0039	ironwood-schedule-risk	succeeded	—	2m 41s	$0.22	Aug 17, 11:36 PM
0199…0038	ironwood-agent-review	failed	sre	47s	$0.09	Aug 17, 6:48 PM
0199…0037	ironwood-security-audit	succeeded	ci	10m 21s	$1.98	Aug 17, 1:02 PM
0199…0036	ironwood-doc-intake	succeeded	—	6m 5s	$0.63	Aug 17, 6:48 AM
0199…0035	ironwood-schedule-risk	succeeded	—	2m 33s	$0.21	Aug 16, 11:36 PM
0199…0034	ironwood-agent-review	succeeded	—	4m 2s	$0.40	Aug 16, 1:02 PM

See it run on your own work.

Tell us what your team keeps redoing, and we’ll show you a train doing it on your data, scored against checks built for your business.

Talk to us →