Sample deliverable

What the audit actually hands you

This is the AI Compute Audit deliverable, rendered by the same code that renders a paid one, against a synthetic workload for a client that does not exist. Nothing on this page is a real customer's data. The point is that you can read the document before you buy it instead of buying a description of one.

Read this first. The report is a decision-support document, not a service guarantee. Its cost figures are benchmark cloud list prices and a modeled orbital rate — no orbital capacity is contracted, and OrbitRoute's own execution path is simulated today. Every section carries its own data-basis label, and the report withholds any percentage it lacks the sample size to state. That withholding is deliberate and it is visible in the sample below.

1. What each tier includes 2. What you have to send 3. The $499 deliverable, in full 4. What the higher tiers add 5. The numbers we refuse to print 6. Where every figure comes from 7. Check the inputs yourself

1. What each tier includes

Sections are the product. This table is the tier catalog in audit_report.AUDIT_TIERS — the same list that gates what gets assembled — so the page cannot describe a section the renderer does not produce.

Section	Snapshot $499	Deep Dive $1,999	Enterprise $5,000	What it is
executive_summary	included	included	included	Workload profile, the modeled orbital premium or saving at your volume, and whether the report had platform history to work from.
cost_comparison	included	included	included	Per-job and monthly cost across AWS, GCP, Azure and modeled orbital, plus the crossover $/GPU-hr at which orbit wins for your workload.
usage_analysis	—	included	included	Your real API tier, daily limit, limit used today, calls at 1/7/30/all days, job counts, first and last activity. Requires an OrbitRoute key with history.
sla_analysis	—	included	included	The public predicted-vs-actual record, and your own graded jobs if you have any — with the sample size printed next to every figure.
routing_intelligence	—	—	included	The live scoring weights, optimization rounds, orbital win rate, and prediction accuracy — each one either measured with its n, or explicitly withheld.
platform_context	—	—	included	Assets tracked, ground stations modeled, and the standing disclosure that capacity and ground segment are models rather than contracts.
recommendations	included	included	included	Ranked actions. Each one cites the number it came from — no recommendation exists without a figure behind it.
data_provenance	included	included	included	The appendix defining every data basis used in your specific report.

The $1,999 and $5,000 figures here are one-off audit engagements. The same two numbers on /pricing are monthly API tiers — a different product with a different deliverable. Read the tier name, not just the price.

2. What you have to send

Three things, and only the first is mandatory.

Input	Required	What it changes
Workload profile	yes	GPU type, jobs per day, average compute seconds per job. These drive the entire cost model. Send them and they are used verbatim (`derived_from: explicit_overrides`).
An OrbitRoute API key	no	Unlocks `usage_analysis` and your own SLA rows, and lets the workload profile be derived from your real 30-day history instead of what you tell us (`derived_from: customer_history`). Without one the report runs in prospect mode on benchmark and modeled data only, and says so on the first page.
Context	no	Use case, budget, constraints — free text through the intake form on /optimize. It shapes the recommendations narrative; it never becomes a number.

If you send nothing but a workload profile, you get exactly the document in §3. Nothing is padded to fill a tier.

3. The $499 deliverable, in full

Not an excerpt. This is the entire Snapshot report for a prospect with no platform history — generated on 2026-07-30 from the workload H100, 140 jobs/day, 3600 s per job, with the client name redacted. A real engagement changes the numbers; it does not change the shape.

Sample — synthetic workload, redacted client

AI Compute Audit — Snapshot

Prepared for: NORTHWIND ANALYTICS (redacted)

Generated: 2026-07-30T15:09:58Z

Engagement: $499 tier

Executive Summary

Workload profile: H100, 140 jobs/day, 3600s avg compute (source: explicit_overrides)
Modeled orbital premium: 200.0% ($33,432.00/month more than the cheapest cloud at this volume — terrestrial wins on price today)
Client on platform: no — prospect mode, benchmark + modeled data only

Cost Comparison

Data basis: benchmark

Target	Per job (USD)	Monthly × 4200 jobs
AWS	$4.1500	$17,430.00
GCP	$3.9800	$16,716.00
Azure	$4.3000	$18,060.00
Orbital (modeled)	$11.9400	$50,148.00

Modeled premium vs cheapest cloud: 200.0% ($33,432.00/month more). Recommendation: terrestrial.

Orbital pricing is modeled at 3.0x the cheapest terrestrial cloud rate ahead of the 2027 capacity wave — orbital reaches price parity for this workload when its $/GPU-hr crosses $3.98 (H100).

Recommendations

Data basis: measured

Terrestrial cloud wins on price today: modeled orbital routing would cost $33,432.00/month MORE at your workload volume. Orbital pricing is modeled at 3.0x the cheapest terrestrial cloud rate ahead of the 2027 capacity wave — orbital reaches price parity for this workload when its $/GPU-hr crosses $3.98 (H100).

Data Provenance

benchmark — Published terrestrial cloud list prices (AWS/GCP/Azure, March 2026) as configured in the routing engine.
measured — Real rows from the live OrbitRoute database (API usage, jobs, SLA grades).

OrbitRoute — Intelligent Workload Orchestration for Space-Based AI Compute

Yes — the answer this sample gives is "stay terrestrial", and it prints the exact price at which that stops being true. An audit that could only ever recommend the thing being sold would not be worth $499.

4. What the higher tiers add

The same run at the $5,000 tier, same synthetic workload. Executive summary, cost comparison, recommendations and provenance are identical to §3; these are the sections layered on top.

Sample — additional sections, synthetic workload

Usage Analysis measured

Deep Dive and above, and only with an API key that has history — omitted entirely in prospect mode rather than filled with defaults.

API tier, daily limit, limit used today, calls today / 7d / 30d / total, jobs total / active / 30d, customer since, last active — read from your own rows.

SLA Analysis measured

Data basis: measured

Platform scoreboard (all graded jobs): 9 jobs measured, 88.9% deadline hit rate, 115.24% avg latency prediction error — a dated reading from 2026-07-30; the live figure is at /api/v1/jobs-stats/sla and on /sla.

Calibration phase — n = 9 (< 50). These are calibration measurements of the predictor, not a service guarantee; sample size is disclosed next to every figure, misses included.

Routing Intelligence (Learning Loop) measured

Data basis: measured

Scoring weights: latency 0.25, cost 0.30, availability 0.20, queue_depth 0.15, power 0.10
Optimization rounds completed: 0
Orbital win rate: not yet measured (0 routing decisions analyzed)
Avg cost savings across decisions: not yet measured (0 routing decisions analyzed)
Prediction accuracy: withheld below 30 graded jobs
Learning cycles run: 0

Calibration phase — this report does not quote a prediction-accuracy percentage below 30 graded jobs. Nothing is hidden: every graded job is published from job one and no record is ever excluded, and the running figure — misses included — is live at orbitroute.ai/sla. It is withheld HERE because at low n a handful of misses dominates the mean, so a percentage at this stage measures sample size rather than steady-state accuracy, and this document should not hand you a number that will move by tens of points.

Platform Context modeled

Data basis: modeled

Orbital assets tracked live: 24 core assets (the full catalog under propagation is larger — see /api/v1/fleet/live)
Ground stations in network model: 37

OrbitRoute tracks live satellite ephemerides (CelesTrak + SGP4) and grades every managed job on a public SLA scoreboard. Orbital compute capacity and ground stations are currently modeled — no operator uplink or ground-segment contract is in place. Every routing decision discloses this via capacity_class and source labels.

5. The numbers we refuse to print

A paid document is the worst possible place to publish a percentage that has no sample behind it, so the renderer has hard floors. They are the reason two lines in §4 say "withheld" instead of showing a number that would look better.

Figure	Floor	Below the floor
Prediction accuracy	30 graded jobs	Withheld, with the reason stated in the report. At low n a couple of misses dominate the mean, so the number would measure sample size rather than accuracy.
Deadline hit rate / latency error	1 graded job	Withheld. The platform's own summary returns a 100.0% display default at zero samples; the report refuses to print it, because a default is not a measurement.
Deadline hit rate / latency error	50 graded jobs	Printed, but every figure carries its n and an explicit "calibration measurements, not a service guarantee" caveat — the same threshold /sla uses.
Orbital win rate, cost savings	1 analyzed decision	"Not yet measured (0 routing decisions analyzed)" — never a seeded 0.0% dressed as a result.

6. Where every figure comes from

Three data bases appear in a report, and the appendix in your copy defines the ones your tier actually used. The full vocabulary, with the physics behind it, is on /methodology.

Basis	Means	Used for
benchmark	Published terrestrial cloud list prices (AWS/GCP/Azure, March 2026) as configured in the routing engine.	The AWS/GCP/Azure rows of the cost table. Same rate card the public calculator publishes.
measured	Real rows from the live OrbitRoute database — API usage, jobs, SLA grades.	Usage analysis, SLA analysis, routing intelligence, and the figures the recommendations cite.
modeled	Simulated or projected. Orbital pricing is a multiple of terrestrial cost; fleet and ground-segment capacity are modeled, not contracted. No real orbital uplink exists yet.	The orbital row of the cost table, the crossover price, and platform context.

What the audit is not. It is not a security or compliance audit, it does not access your cloud accounts, and it does not benchmark your models. It prices a described workload against published cloud rates and a modeled orbital rate, reads whatever OrbitRoute history you have, and tells you what to do about it — including, as in §3, doing nothing.

7. Check the inputs yourself

Every input to the sample above is reproducible from a public endpoint before you spend anything.

Endpoint	Checks
`/api/v1/audit/pricing`	The tier catalog and prices in §1, straight from the source the report is assembled from.
`/api/v1/cost-compare`	The §3 cost table for your own workload, including the crossover line, before you buy anything.
`/api/v1/jobs-stats/sla`	The graded-job record behind §4's SLA section and the 30/50-job floors in §5.
`/api/v1/ground-stations/coverage`	The 37 modeled stations in §4's platform context, and their real contact duty cycle.

curl https://www.orbitroute.ai/api/v1/audit/pricing
curl https://www.orbitroute.ai/api/v1/jobs-stats/sla

Reproduce the cost table curl -X POST https://www.orbitroute.ai/api/v1/cost-compare \ -H "Content-Type: application/json" -H "X-API-Key: $KEY" \ -d '{"model_name":"audit-workload-profile","gpu_required":"H100","input_size_mb":1,"estimated_compute_seconds":3600}'

Ready, or want the shape changed for your case? Start on /optimize — or read the methodology first, which is the honest order.