Add Populace pipeline overview brief and figure

[vahid-ahmadi]

Summary

Adds a self-contained brief describing Populace and the calibration pipeline, for use as talk background and a repo reference. Companion to the L0 paper draft.

• paper/populace_overview.md — pipeline overview in README tone:
  • what Populace is + the five-package architecture (frame, data, fit, calibrate, build)
  • the Frame spine (typed weights, strata/provenance, weighted accounting)
  • the six pipeline stages, prior to and including calibration: sources → combine → impute → geography → build targets → calibrate
  • the L0 calibration step (the loss, Hard Concrete gates, λ_L0/λ_L2, target_records budget control)
  • the four-method sampling experiment (informed L0 vs. random→reweight vs. survey-weight sampling vs. combinatorial optimization)
  • a Statistics section covering the points the method has to defend (L0/Hard Concrete relaxation, samplers in survey-statistics terms, out-of-sample validity, correlated evidence, metrics)
• paper/figures/populace_pipeline.png — pipeline overview figure (slide-ready)
• paper/figures/populace_pipeline.py — its matplotlib generator (uv run --with matplotlib python paper/figures/populace_pipeline.py)

Content is grounded in populace/DESIGN.md, paper/sections/data.tex, paper/sections/methodology.tex, the live populace-calibrate solver, and paper/sampling_lit_review.md.

Notes

• Docs/figure only — no changes to paper sections or code.
• Uses Populace naming; the data.tex/methodology.tex drafts still say \microplex.

🤖 Generated with Claude Code