Sold by: Rose & Lotus
Deployed on AWS
Turn injection transcripts into an effectiveness report: block rate, leakage rate, false positive rate and a weighted risk score. No model is called, so scoring is fully reproducible.
Reads any transcript matching the published schema, whether it came from our tester, another vendor's tool, or your own harness. Writes JSON for pipelines and Markdown for auditors, and exits with code 2 when a threshold is breached so it works as a build gate.
Overview
Teams that test their guardrails end up with a folder of transcripts and no answer. Reading a few hundred model replies by hand is not a control, and it does not survive being asked to do it again next quarter.
The obvious shortcut, asking a model to grade the replies, creates worse problems for this particular buyer. It costs money per case, it gives a different answer on Tuesday than it gave on Monday, it sends the exact content the guardrail was protecting to somebody else's model, and when an auditor asks why a given case counted as a leak there is no answer worth giving.
Guardrail Effectiveness Scorer grades transcripts with published rules. Same input, same output, on any machine, with the arithmetic printed in the report. It measures block rate, leakage rate, sanitize rate, false positive rate and a severity weighted risk score with a letter grade, broken out by severity and by payload category. Every number carries its own numerator and denominator, because auditors do not accept a bare percentage. They ask what the seventeen percent is seventeen percent of.
The risk formula is printed in every report along with that run's own working, so anyone can check the headline figure with a calculator. Nothing about the scoring is hidden, and nothing about it is behind a licence.
It reads any transcript that matches the published v1.0 schema. The schema forbids the tool that produced a transcript from recording its own verdict, so all the judgement happens here under one set of rules. Two teams running different testers can still compare numbers, and a team already using another harness can adopt this without replacing it.
The tool is deliberately unwilling to flatter you. A threshold set on something the run never measured fails rather than passes, because passing would be false assurance. Rates that were never measured are reported as not measured rather than as zero. A transcript with no benign controls raises a warning, since a guardrail that refuses everything would otherwise look flawless. And because the risk score measures leakage only, the report prints the over-blocking rate directly under the grade and says plainly when a guardrail is refusing too much legitimate traffic to ship.
It reads, it scores, it writes, it exits. There is no server, no dashboard and no control plane to log into. Reports go to a local directory or your own S3 bucket, and exit code 2 fails a pipeline. No model is called at any point, so a scoring run needs no egress beyond reading your own objects.
Highlights
- Deterministic scoring with no model in the loop. The same transcript always produces the same report, at no per case inference cost, with no content sent anywhere. For a compliance buyer, being able to reproduce the number is the whole product.
- The risk formula is printed in every report along with this run's own arithmetic, so an auditor can recheck the headline figure by hand. Weights, credits, numerators and denominators are all published. Nothing about the scoring is hidden.
- Reads any transcript matching the published v1.0 schema, from our tester, another vendor's tool, or your own harness. Exit code 2 on a breached threshold makes it a build gate, and an unconfigured gate always passes so adding it never breaks a pipeline.
Details
Sold by
Categories
Delivery method
Supported services
Delivery option
Docker container - run with Docker, ECS, or EKS
Latest version
Operating system
Linux
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Pricing is based on the duration and terms of your contract with the vendor. This entitles you to a specified quantity of use for the contract duration. If you choose not to renew or replace your contract before it ends, access to these entitlements will expire.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost/12 months |
|---|---|---|
Growth | up to 100 scoring runs per month | Up to 100 scoring runs per month. Includes the full classification rules, the risk score with letter grade, JSON and Markdown reports, and threshold gating with CI exit codes. Email support with a two business day response. | $800.00 |
Business | up to 1,000 scoring runs per month | Up to 1,000 scoring runs per month. Adds custom refusal corpora for models that refuse in their own style or language, S3 input, and batch scoring across a whole prefix. Email support with a one business day response. | $1,000.00 |
Dimensions summary
You choose between two contract tiers based on how many scoring runs you need each month. Growth covers up to 100 scoring runs per month. Business covers up to 1,000 scoring runs per month and adds capabilities like S3 input and batch scoring across a whole prefix. Both tiers include the core classification rules, risk scoring, and report outputs. The tiers also differ in support response time, with Growth offering a two business day response and Business a one business day response. Pricing scales with your monthly run volume and support needs.
Top-of-mind questions for buyers
A scoring run processes a batch of transcript JSON and produces one effectiveness report. Each run ingests transcripts matching schema v1.0 and outputs block rate, leakage rate, false positive rate, and a severity-weighted risk score. Reports come as JSON, Markdown, or both. Each report generation counts as one run against your monthly quota.
Current contract dimensions use fixed monthly quotas with no metering. Growth covers up to 100 runs per month; Business covers up to 1,000. The vendor lists per-run metering above quotas as a future roadmap item, not active today. To handle more volume now, choose the tier matching your expected run count.
Growth accepts transcript input for scoring and produces reports with threshold gating and CI exit codes. Business adds S3 input, batch scoring across an entire S3 prefix, and custom refusal corpora for models that refuse in their own style or language. Both run as batch tasks in your VPC.
roselotusllc.com+1
Helpful?
Vendor refund policy
Contact support@roselotusllc.com within 30 days of your first subscription charge for a full refund of that charge. Include your AWS account ID and the reason. We process approved refunds through AWS Marketplace within 5 business days. Renewals after the first 30 days are not refundable, though you can cancel at any time to stop future charges.
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Docker container - run with Docker, ECS, or EKS
Supported services: Learn more
- Amazon ECS
Container image
Containers are lightweight, portable execution environments that wrap server application software in a filesystem that includes everything it needs to run. Container applications run on supported container runtimes and orchestration services, such as Amazon Elastic Container Service (Amazon ECS) or Amazon Elastic Kubernetes Service (Amazon EKS). Both eliminate the need for you to install and operate your own container orchestration software by managing and scheduling containers on a scalable cluster of virtual machines.
Version release notes
This is the first public release of Guardrail Effectiveness Scorer.
What you get
- Deterministic scoring of schema v1.0 injection transcripts (no model is called)
- Block rate, leakage rate, sanitize rate, false-positive rate, and a severity-weighted risk score with letter grade
- Every report prints the risk formula and that run's own arithmetic so an auditor can recompute it
- Works with transcripts from Injection Attack Suite Runner, another vendor tool, or your own harness
- Writes report.json and report.md locally (or reads from S3)
- Exit code 2 when a threshold is breached, so you can use it as a CI / pipeline gate
- Distroless, non-root batch container for Docker, ECS, or EKS
What this version does not do
- It does not call your model or any third-party model
- It does not invent verdicts that were already written into a transcript (producers must record observations only)
Same input always produces the same report.
Additional details
Usage instructions
OPTION A - Score a local transcript folder (fastest start)
-
Put transcripts in a folder, for example: mkdir -p ./transcripts ./reports
copy your *.json transcripts into ./transcripts
-
Run a report with a zero-leak gate: docker run --rm
-v "$PWD/transcripts:/in:ro"
-v "$PWD/reports:/out"
709825985650.dkr.ecr.us-east-1.amazonaws.com/rose-lotus/rl-guardrail:1.0.0
-max-leakage-rate 0 -
Open the files written under ./reports:
- report.json (for pipelines)
- report.md (for people and auditors)
-
Read the console summary. If the gate fails, the container exits 2.
OPTION B - Score one file
docker run --rm
-v "$PWD/transcripts:/in:ro"
-v "$PWD/reports:/out"
709825985650.dkr.ecr.us-east-1.amazonaws.com/rose-lotus/rl-guardrail:1.0.0
-input /in/your-run.json
-output-dir /out
-max-leakage-rate 0
OPTION C - Score transcripts already in S3
- Give the task/pod role s3:GetObject and s3:ListBucket on the prefix.
- Run:
docker run --rm -v "$PWD/reports:/out"
-e RL_SCORER_INPUT=s3://YOUR-BUCKET/injection-transcripts/
-e RL_SCORER_OUTPUT_DIR=/out
-e RL_SCORER_MAX_LEAKAGE_RATE=0
709825985650.dkr.ecr.us-east-1.amazonaws.com/rose-lotus/rl-guardrail:1.0.0
Use a trailing slash on the S3 prefix to score every object under it.
OPTION D - CI / build gate (recommended thresholds)
docker run --rm
-v "$PWD/transcripts:/in:ro"
-v "$PWD/reports:/out"
709825985650.dkr.ecr.us-east-1.amazonaws.com/rose-lotus/rl-guardrail:1.0.0
-min-block-rate 0.95
-max-leakage-rate 0
-max-false-positive-rate 5%
-require-benign-controls
In GitHub Actions / CodeBuild, fail the job when the exit code is 2.
OPTION E - Report only (no gate)
Omit all threshold flags. The tool still writes the report and exits 0 after a successful score:
docker run --rm
-v "$PWD/transcripts:/in:ro"
-v "$PWD/reports:/out"
709825985650.dkr.ecr.us-east-1.amazonaws.com/rose-lotus/rl-guardrail:1.0.0
USEFUL OUTPUT CONTROLS
Summary on stdout (default): -e RL_SCORER_STDOUT=summary
Full JSON on stdout: -e RL_SCORER_STDOUT=json
Markdown on stdout: -e RL_SCORER_STDOUT=markdown
Quiet stdout (files only): -e RL_SCORER_STDOUT=none
IF YOUR MODEL REFUSES IN A HOUSE STYLE
Create a text file with one refusal phrase per line, then:
docker run --rm
-v "$PWD/transcripts:/in:ro"
-v "$PWD/reports:/out"
-v "$PWD/our-refusals.txt:/phrases.txt:ro"
709825985650.dkr.ecr.us-east-1.amazonaws.com/rose-lotus/rl-guardrail:1.0.0
-refusal-phrases /phrases.txt
-max-leakage-rate 0
EXIT CODES
0 = scored successfully and all set thresholds passed 1 = config or runtime error 2 = a threshold was breached (fail the build) 3 = input is not a readable schema v1.0 transcript
Unset thresholds mean "do not gate on that metric." Setting -max-leakage-rate 0 means "fail on any leak."
Typical handoff from Injection Attack Suite Runner: tester writes -> s3://bucket/injection-transcripts/ scorer reads -> s3://bucket/injection-transcripts/
Docs: https://roselotusllc.com/docs/guardrail-scorer Support: support@roselotusllc.com
Support
Vendor support
Support is by email at support@roselotusllc.com , Monday to Friday, 9am to 6pm Eastern, excluding US public holidays.
Growth: response within two business days. Business: response within one business day. Enterprise: response within four business hours, with a named contact and a scheduled onboarding call.
Every tier includes the full product documentation at https://roselotusllc.com/docs/guardrail-scorer , ready to use deployment templates for Docker Compose, ECS RunTask, EKS Jobs and CloudFormation, and the complete scoring rules including the refusal phrase corpus.
Please include your AWS account ID, the product version, and the run ID shown at the top of the report. If a case was classified in a way you disagree with, quote the case ID and the rationale line from the report's case detail table, since that names the exact evidence the classifier used. Do not send transcripts or reports unless we ask for them, because they contain your prompts and your model's replies.
If your model refuses in a house style or a language the built in phrase corpus reads poorly, you can supply your own with the refusal-phrases flag. Ask us and we will help you build one from a sample of your own refusals.
For private offers, bundle pricing with Injection Attack Suite Runner, or procurement questions, write to sales@roselotusllc.com .
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
Turbot Guardrails is the leading platform for policy-based control and automatic remediation of enterprise clouds.
Turbot Guardrails is the leading platform for policy-based control and automatic remediation of enterprise clouds.
Secure AI applications, models, and connected data with runtime protections and continuous testing. Detect, prevent, and outpace evolving risks across the AI attack surface.
Tumeryk AI Trust Score™ Guardrails provide real-time security and governance for GenAI agents and applications by enforcing jailbreak detection, bias monitoring, and content moderation. Integrated with tools like Wiz, Tumeryk also discovers Shadow AI and scores its risk, ensuring safe and compliant AI adoption.
A cutting-edge solution designed to support secure coding practices with generative AI, ensuring legal compliance and adherence to organizational policies.
Customer reviews
No customer reviews yet
Be the first to review this product . We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.