Skip to content

Version History

Release status

Status: Current through the v0.7 evidence-backed guidance release Last updated: 2026-06-17 Scope: Shipped milestones and near-term cleanup work for decision-ready PET guidance.

This page tracks what each site version is meant to change. It is not a changelog for every line edit; it is the editorial map readers and contributors can use to understand the maturity of the guide.

v0.7 — Evidence-Backed Guidance

v0.7 moves deployment and benchmark guidance from strong editorial advice toward sourced field-guide evidence. The goal is to help readers distinguish measured deployments, pilots, vendor case studies, proposed use cases, and illustrative examples before they make decisions.

Area Shipped improvement
Evidence policy Add source-quality labels and clearer deployment maturity labels.
Deployments Separate production/recurring deployments, pilots, vendor case studies, and proposed use cases on domain pages.
Deep dives Add source-backed studies for EXAM COVID-19 federated learning and Japanese bank fraud-detection FL.
Existing deep dives Add source-quality fields to the snapshot tables.
Benchmarks Add benchmark source-quality labels and mark illustrative values as Unsourced / illustrative.
Claim register Refresh the claim snapshot around deployment maturity, benchmark evidence, and remaining unresolved claims.

v0.6 — Consistency And Evidence Hygiene

v0.6 is a cleanup release after the larger v0.3-v0.5 content expansion. The goal is to make the site easier to trust: no hidden sections, no stale release language, and no illustrative benchmark values that look like measured results.

Area Shipped improvement
Navigation Add Tool Reviews to the main nav and keep every Markdown page reachable.
Homepage Route readers to Worked Decisions, Tool Reviews, and Benchmark Example Runs.
Evidence hygiene Reframe benchmark examples as hypothetical scorecard exercises unless a value is measured or literature-backed.
Claim register Update the snapshot scope to the current evidence backlog.
Reader paths Remove stale wording that described the decision tree as a single diagram.
CI Add navigation coverage checking and build with MkDocs strict mode so warnings fail early.

v0.5 — Tool Reviews And Worked Evaluation

v0.5 added practical evaluation artifacts around PET tools and decision workflows.

Area Shipped improvement
Tool reviews Added worked examples for Flower, OpenDP, Microsoft SEAL, and confidential-computing platforms.
Tool evaluation Added a framework and template for reviewing tools by architecture fit, threat model, evidence, operations, and first benchmark.
Worked decisions Added concrete scenario pages for hospital model training, cross-bank fraud detection, private RAG, synthetic data release, private model inference, and public-sector statistics.
Benchmarks Added example scorecard runs to show how benchmark templates can be applied to realistic workloads.

v0.4 — Decision Kit Foundations

v0.4 made I ❤️ PETs more usable as a decision kit: clearer use cases, reusable benchmark scorecards, stronger architecture review, and better evidence discipline.

Area Shipped improvement
Use cases Turned domain summaries into playbooks with scenarios, PET stacks, non-use guidance, failure modes, and measurements.
Benchmarks Added scorecards for private RAG, private inference, cross-silo FL, synthetic data release, and federated/MPC analytics.
Architectures Added trust-boundary reviews, assumptions, non-protections, and evaluation checklists.
Evidence Strengthened claim states, dated review workflow, and claim-register backlog.
Tools Added a framework for evaluating PET tools by fit, threat model, evidence, operations, and first benchmark.
Start Here Improved onboarding with wrong-assumption checks, routing, and practical glossary examples.
Site quality Added internal-link checking and kept navigation aligned with new guidance.

Release Checklist

  • mkdocs build --strict --clean passes.
  • Internal-link check passes.
  • Navigation coverage check passes.
  • New navigation entries point to existing pages and every docs page is intentionally reachable.
  • Decision pages include at least one concrete table, checklist, worked example, or scorecard.
  • No page claims a PET solves privacy without naming assumptions and outputs.
  • Illustrative examples cannot be mistaken for measured production evidence.
  • Deployment and benchmark claims carry source-quality labels when they affect a decision.

Known Follow-Ups

  • Replace hypothetical benchmark examples with measured or literature-backed runs where possible.
  • Add source-backed claim notes for high-impact claims in the claim register.
  • Revisit fast-moving tool version and performance claims before major releases.
  • Add a lightweight stale-content check for pages with dated review metadata.