Learn → run → inspect → adapt

Put the research to work.

Start with tools you can run in your own project. Inspect what they actually check, then decide what to adopt.

Download the runnable kit   Hashes and file manifest

Code revision bb105a93280a1b58 · 3 worked entry points · MIT first-party code. No third-party source snapshots or private experiment logs are bundled.

Extract the ZIP and start in fr-starter-kit/. Compare its SHA-256 with the manifest before running it. For existing projects, review and copy selected files; reinstalling is not an upgrade procedure.

Give an AI project gates that can actually fail

Catch an incomplete plan and unsupported evidence labels before work is signed off.

Requires: POSIX shell utilities; Git recommended. Use a new project directory.

sh scripts/init.sh ../my-new-project
cd ../my-new-project
sh scripts/check-readiness.sh

Expected: A pristine project reports NOT READY (exit 1). Complete and review the packet before expecting success.

Limits: Structural completeness is not execution authorization. The installer is not an upgrade command; it reinstalls the live hook.

scripts/init.sh · scripts/check-readiness.sh · CHECKLIST.md

Reproduce the repository check
python3 starter-kit/tests/test_gates.py

Commands above start in the extracted kit. This check starts in an FR checkout. The manifest identifies the exact shipped bytes; test success does not certify the meaning of a research claim.

Run, invalidate and transfer an evidence-backed decision

Learn how to stop an old experiment result silently supporting changed code.

Requires: Python 3.9+, POSIX shell utilities. Give the exercise a destination that does not exist.

python3 examples/decision-cycle/run.py ../decision-exercise

Expected: A real local execution receipt, a stale-input failure, a demoted claim, and a standalone public transfer. See the exercise README for generated paths.

Limits: This deliberately small exercise teaches identity checking; it does not prove research quality or a speedup on your workload.

examples/decision-cycle/run.py · examples/decision-cycle/probe.py · examples/decision-cycle/README.md · scripts/check-decisions.py · scripts/review-decisions.py

Reproduce the repository check
python3 scripts/test_learning_example.py

Commands above start in the extracted kit. This check starts in an FR checkout. The manifest identifies the exact shipped bytes; test success does not certify the meaning of a research claim.

Resume the decisions that need attention

Hand an agent the specific next check and the evidence that made it necessary.

Requires: Python 3.9+ and your own maintained version-1 decision record; paths are relative to your project root.

python3 scripts/review-decisions.py path/to/decisions.json --root /path/to/project --json
python3 scripts/check-decisions.py path/to/decisions.json --root /path/to/project --refresh

Expected: A ranked review queue; refresh demotes stale labels and never promotes claims.

Limits: A zero exit can still mean REVIEW or defer. Read the state. Metadata must already be safe to share; export is not a rights classifier.

scripts/check-decisions.py · scripts/review-decisions.py · DECISIONS.md

Reproduce the repository check
python3 scripts/test_review_decisions.py

Commands above start in the extracted kit. This check starts in an FR checkout. The manifest identifies the exact shipped bytes; test success does not certify the meaning of a research claim.

What needs investigation next?

Last upstream observation: . Check the observation time before relying on freshness.

This queue changes with the daily watch. Changed or unknown upstream facts do not approve a new technique or rewrite an assessment at its recorded commit.

  • asupersync — unknown: unknown: ci now (FR-C.3/no-settled-commit)
  • franken_agent_detection — changed: open crossing franken_agent_detection:ci:C2>C1
  • franken_networkx — unknown: unknown: ci now (FR-C.3/no-settled-commit)
  • franken_remote — changed: open crossing franken_remote:ci:C3>C1, franken_remote:ci:C3>C2, franken_remote:rel:R1>R3
  • franken_snowflake — changed: open crossing franken_snowflake:ci:C3>C5
  • franken_surveillance_system — changed: open crossing franken_surveillance_system:ci:C6>C2, franken_surveillance_system:ci:C6>C3, franken_surveillance_system:rel:R1>R2, franken_surveillance_system:revisit:3
  • franken_threed — changed: open crossing franken_threed:ci:C1>C2
  • franken_whisper — unknown: unknown: ci now (FR-C.3/no-settled-commit)
  • frankengit — changed: open crossing frankengit:rel:R1>R2
  • frankengraphdb — changed: open crossing frankengraphdb:rel:R1>R2
  • frankenjax — changed: open crossing frankenjax:ci:C4>C2
  • frankenmermaid — changed: open crossing frankenmermaid:rel:event:release:v0.3.0, frankenmermaid:rel:event:release:v0.4.0, frankenmermaid:rel:event:release:v0.5.0
  • frankenpandas — changed: open crossing frankenpandas:rel:R2>R3
  • frankenscipy — changed: open crossing frankenscipy:ci:C1>C2
  • frankensearch — unknown: unknown: ci now (FR-C.3/no-settled-commit)
  • frankensim — changed: open crossing frankensim:rel:R1>R2
  • frankensqlite — unknown: unknown: ci now (FR-C.3/no-settled-commit)
  • frankentorch — changed: open crossing frankentorch:rel:R1>R2
  • frankentui — unknown: unknown: ci now (FR-C.3/no-settled-commit)

Daily collection runs · Runnable cloud candidates and review

Candidate generation needs a configured coding-engine credential. A draft candidate is unaccepted code; independent tests and review precede promotion into this kit. Raw candidates are never added to this download automatically.

What counts as better?

Measure a fresh user task: correct application, missed or false claims, transfer failures, and total setup, review and maintenance effort. These tools have bounded regression checks. A 100× improvement in research value has not been demonstrated.

Machine-readable recipes and queue · Complete kit instructions

Help fix it

Each opens a GitHub issue form with this page’s address filled in. You need a GitHub account.

Made by Joshua Nowak (ZestStream) with AI coding agents. Not affiliated with or funded by Jeffrey Emanuel; Joshua uses his public tools and is a paying subscriber to jeffreys-skills.md.

Source on GitHub MIT licensed