Outputs and manuscript assembly¶
Analysis code reads data beneath /data/<coverage>/ and writes fitted models,
predictions, metrics, figures, and table fragments beneath:
one_percent is the development and formatting coverage. Final manuscript
estimates use full after the required full-sample data products exist.
Selected worker-level pipelines also accept samples or a subset of score
shards for smoke tests; those results are diagnostics, not final estimates.
Artifact lifecycle¶
flowchart LR
direct[(Direct upstream input)]
analysis[(Analysis-ready dataset)]
estimate[[Estimate and format]]
support[(Saved models, predictions,<br/>metrics, and CSV files)]
latex[/LaTeX table fragment/]
direct --> estimate
analysis --> estimate
estimate --> support
support --> latex
classDef upstream fill:#e8eef7,stroke:#52739e,stroke-width:2px
classDef input fill:#eee5f5,stroke:#76528c,stroke-width:2px
classDef code fill:#f3f3f3,stroke:#666,stroke-width:2px
classDef support fill:#fff0cc,stroke:#a66b00,stroke-width:2px
classDef tableOutput fill:#e2f2e7,stroke:#3b7a4b,stroke-width:2px
class direct upstream
class analysis input
class estimate code
class support support
class latex tableOutput
The generated .tex files are tabular fragments. The manuscript or preview
document owns captions, labels, notes, placement, and surrounding table
environments. A local preview document may aggregate fragments using paths
relative to the coverage artifact directory, but it is not a generated
canonical result.
Saved estimation state¶
Long-running hazard, performance, and wage pipelines persist enough state to
resume or reformat without repeating every fit. Their analysis pages identify
the exact saved models, covariance matrices, predictions, and metrics. A normal
rerun reuses valid state; --rerun requests refitting where supported.
Every completed analysis should preserve:
- numeric results in CSV or RDS form;
- metadata identifying inputs, sample choices, and outputs;
- a standalone LaTeX fragment when the analysis is manuscript-ready;
- expensive fitted state or predictions when needed for reliable resumption.
Run commands from /hct-tower/analyses in the analyses container. See the
Slurm guide for submission, resource overrides, and
log locations.