limingrui679-design/high-stakes-analytics-decision-lab

High-Stakes Analytics & Decision Lab

Evidence-gated analytics for consequential questions. Move from an ambiguous decision to reproducible evidence—and only then, when justified, to bounded action.

Quick start · Architecture · Live explorer · Portfolio · Documentation · Contributing

Why this exists

High-stakes analysis often fails before the model: the question is underspecified, the data contract is implicit, cleaning choices are hidden, uncertainty is treated as independent, or a recommendation is written because the template expects one.

This repository is a platform-neutral Agent Skill and reproducible research portfolio built around a stricter sequence:

The result is not a fixed report generator. It is an evidence-gated orchestration system that can stop, ask for a named decision, or produce a bounded analytical product without upgrading weak evidence into a stronger claim.

Start in three steps

1. Install the Skill

npx skills add limingrui679-design/high-stakes-analytics-decision-lab -g

The Agent Skills installer discovers the compact package under skills/high-stakes-analytics-decision-lab/: 39 files and about 472 KiB, rather than the full research portfolio. Its machine-readable file and hash contract is in bundle-manifest.json.

Use docs/getting-started.md for Codex-specific, no-install, and direct repository options.

2. Ask for the evidence outcome

$high-stakes-analytics-decision-lab
Run the data-readiness gate on this source, preserve the original file, and
select only the analytical routes the evidence supports. Produce an Evidence
Intelligence Report. Add a Decision Intelligence Brief only if the evidence
and decision context justify one.

Start with the decision or evidence question—not a preferred model. A valid result may be a bounded action, a pilot requirement, targeted diligence, an evidence request, negative validation, or do_not_deploy.

3. Review the evidence package

Every complete project keeps the narrative, machine result, visual evidence, and source lineage together:

report.md                    # primary Evidence Intelligence Report
results.json                 # machine-readable analytical result
chart-map.json               # figure-to-question and source contract
figures/*.svg                # accessible analytical visuals

A justified decision layer adds decision-report.md, decision-results.json, and its own figure contract. It never replaces the primary evidence product.

Direct repository entry points

How it works

The fixed evidence spine remains stable while the case-specific analytical layer changes.

The data gate can stop the workflow

Uploaded row-level data do not go directly into a model. The system preserves the original, establishes a contract, checks grain and keys, profiles quality and privacy, and produces a dry-run remediation plan.

Only safe normalization can run without approval. Deletion, imputation, outlier treatment, category merging, unit conversion, target correction, and grain changes require explicit action IDs. The processed copy never overwrites the source.

Four routes, no mandatory recommendation

Routes may compose, but a later route cannot erase the quality and evidence requirements of an earlier stage. Read the full system design in docs/architecture.md.

Two products, one evidence contract

Every material figure is generated from the shared editorial evidence system, includes a title and description, and is paired with its analytical question, supported interpretation, and claim boundary. The visual system is part of the evidence contract, not decorative reporting.

Fifteen complete evidence paths

The public portfolio contains 15 primary reports and 10 conditional briefs—25 intelligence products in total—plus 119 canonical accessible figures: 50 evidence figures and 69 decision figures.

The cases span operational demand, distribution shift, scarce-capacity pilots, temporal model transport, tail-risk decisions, human-in-the-loop triage, commercial diligence, mitigation allocation, filing review, clustered field experiments, survival evidence, policy evaluation, repeated-measures inference, population transportability, and spatial equity.

They intentionally end differently: some support a bounded decision, some require a pilot or targeted review, and some stop at an evidence request or do_not_deploy.

Verification you can reproduce

The current stable release is v1.1.2. Its versioned source package and SHA-256 checksum are published together; the release identity is also recorded in CITATION.cff, CHANGELOG.md, and RELEASE-MANIFEST.json.

make verify

This runs the standalone regression suite and rebuilds all fifteen projects in an isolated verified-file copy. Use make quality for the tracked-secret scan, Ruff, mypy, and codespell; use docs/verification.md for direct commands, static security gates, and no-Git release verification.

The 106 public tests cover data readiness, adaptive routing, numerical behavior, source and artifact identity, package integrity, local links, accessible SVGs, generator idempotence, compact installation, quickstart safety, interactive explorer synchronization, no-Git releases, DNS and SSRF boundaries, and source parser security. The complete 106-test standalone regression suite and the fifteen-project rebuild are separate gates: successful reproduction establishes the reviewed workflow and declared numerical tolerance, not empirical validity, external adoption, or real-world impact.

CI exercises Python 3.11, 3.12, 3.13, and 3.14. Security checks include the tracked-secret gate, Bandit, dependency auditing, hardened source-builder tests, targeted branch coverage, and CodeQL.

Repository design

high-stakes-analytics-decision-lab/
├── skills/high-stakes-analytics-decision-lab/
│   └── SKILL.md                 # compact, installable Agent Skill package
├── demo/                        # dependency-free interactive case explorer
├── docs/                        # onboarding, architecture, layout, verification
├── references/                  # enforceable method and evidence contracts
├── assets/                      # templates and canonical README visuals
├── scripts/                     # profiling, routing, analysis, generation, checks
├── examples/real-data-cases/    # fifteen reproducible evidence projects
├── tests/                       # standalone contract and regression suite
└── .github/                     # CI, security, issue, and PR workflows

The nested Skill package prevents compatible installers from copying the full portfolio. It is generated from the canonical root scripts, references, and templates by scripts/build_skill_bundle.py; the full cases remain in examples/. Human guides live in docs/; precise runtime rules live in references/; generated evidence stays beside the project that produced it. See docs/repository-layout.md before moving or regenerating files.

Documentation

Responsible-use boundary

This is a public, tested research and portfolio prototype. It does not establish production readiness, institutional adoption, medical advice, investment advice, a regulatory finding, an assurance opinion, or achieved real-world impact. A reproducible result can still be decision-inappropriate; domain review remains mandatory before operational use.

Citation and license

Cite the version reviewed using CITATION.cff. Public release history is in CHANGELOG.md, and component-version boundaries are in VERSIONING.md.

Licensed under the MIT License.