Public implementation
FastTD3: inspect the code
Review the open-source robot-learning implementation and its documented engineering approach.
Explore the evidence →Technical evidence & diagnostic engagements
Evidence before the engagement.
Review public work, understand the assessment process, and agree on concrete outputs before committing to a larger implementation.
Where we help
Make capability inspectable.
Public implementations and transparent assessment methods let your team evaluate technical fit without relying on unsupported claims.
Public engineering work
Architecture reference analyses
Defined diagnostic deliverables
Transparent cost assumptions
Written scope and acceptance criteria
Governance questions before access
Inspect the work. Understand the limits.
Reference analyses
Architecture problem briefs
Explore published technical problem analyses. These are reference materials—not evidence of client engagements or endorsements.
Explore the evidence →Downloadable assessment
AI production-readiness checklist
Use the existing checklist to structure a discussion about dependencies, delivery readiness, and operating controls.
Explore the evidence →Proposed engagement structure
A small, defined diagnostic before a larger commitment.
Start with an architecture review, cloud-cost diagnostic, or robot-learning feasibility assessment. Scope, fee, access, timing, and acceptance criteria are agreed in a written statement of work before kickoff; no implementation is implied.
Compare diagnostic scopes and outputs →Proposed two-week assessment protocol
- Days 1–2: Agree objectives and access boundaries; collect the architecture inventory, cost baseline, and operational constraints.
- Days 3–5: Trace dependencies and inspect cost drivers, reliability evidence, and delivery bottlenecks.
- Days 6–8: Validate findings with engineering owners; separate measured facts, estimates, and untested hypotheses.
- Days 9–10: Present options, risks, and a sequenced roadmap. Two working weeks assumes timely access and stakeholder availability.
Deliverables to agree in the SOW
- Executive summary with evidence and limitations.
- Current-state architecture and dependency map.
- Prioritized findings with severity, owner, and effort ranges.
- Assumption-based cost scenarios—not guaranteed savings.
- Implementation roadmap, validation gates, and rollback considerations.
- Readout with engineering and business stakeholders.
Research context · sources reviewed September 17, 2026
Cloud waste is documented. Recoverable savings must be measured.
Flexera’s 2026 State of the Cloud Report reports 29% estimated wasted IaaS/PaaS spend. Its survey covered 753 technical professionals and executive leaders worldwide, collected in winter 2025 through an independent panel. These are respondents’ estimates, not audited recoverable savings or Fitzroy client results.
The FinOps Foundation’s State of FinOps 2026 identifies workload optimization and waste reduction as the leading current priority, while noting diminishing returns in mature practices. Its Usage Optimization framework calls for weighing savings against engineering effort, operational risk, and performance.
Our interpretation: these sources justify investigating cost efficiency. They do not establish a recoverable percentage, implementation fee, or payback period for your estate. No survey waste estimate is used as a guaranteed reduction.
The homepage scenario, step by step
Fitzroy-selected assumptions—not a research finding or quote: $500,000 monthly spend, a 25% achievable reduction, and $250,000 one-time implementation cost.
- Annual gross savings
- $500,000 × 25% × 12 = $1,500,000
- First-year modeled value
- $1,500,000 − $250,000 = $1,250,000
- Simple payback
- $250,000 ÷ $125,000 monthly savings = 2 months
Assumes the full reduction starts immediately and continues for 12 months at constant workload and prices. Excludes additional recurring costs, ramp-up, taxes, financing, and discounting. Delays, contractual commitments, or added operating costs can reduce value and extend payback. The 25% assumption is not validated by the 29% survey estimate.
Change the illustrative planner inputs →Test your cloud-cost assumptions.
Change the inputs to see how a proposed reduction affects net benefit and simple payback. The starting values are illustrative, not a Fitzroy quote or a validated estimate.
Net monthly savings
$2,750
First-year net benefit
$18,000
Simple payback
5.5 months
Scenario only—not a forecast or industry benchmark. Reduction is your assumption, not a measured waste rate. Monthly savings = spend × reduction − added operating cost. First-year benefit = 12 × monthly savings − implementation cost. Payback assumes savings begin immediately; migration delays, taxes, financing, and business risk are excluded. Inputs stay in your browser and are not submitted.
Settle governance before granting access.
Use procurement and kickoff to establish confidentiality and IP terms, least-privilege access, approved tools and AI providers, retention and deletion requirements, incident contacts, and evidence required for vendor approval. AI-provider retention settings must be verified for the specific service and contract—not assumed.
Request applicable insurance documentation, security attestations, and references during due diligence. This page does not assert SOC 2 certification, insurance limits, zero-retention guarantees, named-client outcomes, or unverified founder scale metrics.
Review the engagement governance checklist →Deliverables
Know what a diagnostic will produce.
Evidence-led findings
Separate observed facts, estimates, and hypotheses in a decision-ready assessment.
Prioritized roadmap
Agree owners, dependencies, validation gates, and implementation options.
Explicit scope
Set the fee, boundaries, timeline, and acceptance criteria in the SOW before work starts.
Define a diagnostic
Start with a question worth answering.
Bring the system constraint, cost concern, or deployment risk. We can discuss a bounded assessment before proposing a larger build.