Scenario-based decision support for comparing alternatives across multiple criteria and multiple future contexts.
baldrick-seer is designed for decisions where one answer must survive more than one situation.
Instead of asking "what is the best option overall?", it lets you ask:
- which option is best for a startup?
- which option is best for a regulated environment?
- which option still looks good when the context changes?
The v1 design is centered on a CLI that reads a structured CUE configuration, validates it, computes weights with AHP, ranks alternatives with TOPSIS, and emits reports for both humans and tools.
- Define a decision problem with named alternatives, criteria, scenarios, and evaluations.
- Reuse the same alternatives across several scenarios instead of rebuilding the model each time.
- Keep scenario setup separate from scenario evaluations, so re-evaluating a model is easy.
- Validate the input config before any ranking happens.
- Derive criteria weights from pairwise comparisons with AHP.
- Rank alternatives with TOPSIS.
- Aggregate scenario results with practical v1 strategies such as equal or weighted averaging.
- CUE is the configuration source of truth.
- The model is designed to stay readable for both humans and AI.
- Criteria, scenarios, alternatives, reports, and references use meaningful names instead of opaque ids.
- Scenarios can define:
- active criteria
- pairwise preferences
- hard constraints
- their own alternative evaluations
- Report definitions can carry CLI-style
key=valuearguments for later customization.
- Human-readable markdown reports.
- Machine-readable JSON output.
- CSV output for spreadsheet or analytics workflows.
- Deterministic output ordering, so repeated runs are stable and reviewable.
- Explainable and traceable results rather than bare rankings only.
The v1 design expects the CLI to catch problems early, including:
- invalid config structure
- unknown references
- broken pairwise comparisons
- missing evaluation coverage
- invalid constraints
- invalid report definitions
- technology platform selection
- infrastructure strategy planning
- system design selection
- supplier or service-provider comparison
- product feature prioritization
- scenario-based ranking of strategic options
These are intentionally treated as later extensions in the current design:
- robustness-focused post-analysis
- sensitivity analysis as part of the ranking workflow
- ELECTRE as an alternative ranking method
- more general method extensibility beyond the initial AHP + TOPSIS pipeline
The current design examples use a scenario-based MCDA model with:
- one problem definition
- a criteria catalog
- a list of alternatives
- a list of scenarios
- a separate evaluation section keyed by scenario name
- a list of requested reports
See:
The repository now includes the full v1 CLI pipeline with:
- Go CLI implementation
- Cobra command wiring
- CUE loading and validation
- AHP weighting
- TOPSIS scenario ranking
- cross-scenario aggregation
- markdown, JSON, and CSV rendering
- Bun + TypeScript end-to-end tests
Recommended install method:
brew install flarebyte/tap/baldrick-seerCheck that the CLI is available:
seer -v
seermake buildThe local development CLI binary is written to:
.e2e-bin/seer
If you build from source and want to run seer directly, add .e2e-bin/ to your shell PATH.
Release binaries for supported operating systems are written to:
build/
You can embed build metadata when needed:
make build VERSION=v1.0.0 COMMIT=$(git rev-parse --short HEAD) BUILD_DATE=2026-03-12T00:00:00Zseer validate --config testdata/config/minimal.cue
seer report generate --config testdata/config/minimal.cue
seer validate --config testdata/config_split--config accepts either a single .cue file or a directory containing a CUE package.
If you want a working model to copy and adapt, start with:
- examples/hello-world.cue for the smallest single-file markdown example
- examples/hello-world-json.cue for a minimal JSON report example
- examples/hello-world-package for a split CUE package loaded from a directory
Try them with:
seer validate --config examples/hello-world.cue
seer report generate --config examples/hello-world.cue
seer report generate --config examples/hello-world-json.cue
seer validate --config examples/hello-world-package
seer report generate --config examples/hello-world-packageThese examples are intentionally small:
- one criterion
- one alternative
- one scenario
- one report
They are meant as a quick-start baseline, not a full demonstration of the v1 feature set.
seer report generate currently renders:
- Markdown for human-readable summaries
- JSON for tool-friendly structured output
- CSV for spreadsheet and analytics workflows
The output is deterministic across repeated runs for the same input.
seer validate checks the loaded model before any ranking or report generation. Current validation covers:
- top-level structure
- duplicate names
- reference resolution
- AHP pairwise comparison completeness and shape
- evaluation coverage and value types
- scenario constraint definitions
- report definition and argument validation
Validation failures return deterministic error output with concise remediation guidance.
See:
Contributor workflow, development commands, coverage, and release-preparation notes are documented in CONTRIBUTING.md.
Many decision tools are either too simple for real trade-offs or too heavy for day-to-day use.
The goal here is a middle ground:
- explicit enough to be reliable
- readable enough to edit
- structured enough to automate
- small enough to fit in a human or AI head