Reframe paper for Populace and add gradient-descent calibration background

[juaristi22]

Summary

Migrates the manuscript narrative from the archived Microplex engine to the Populace pipeline, reframes the study as a Populace-backed proof of concept, and strengthens the calibration background. Addresses issue #1 (rename + context accuracy + Populace-centered narrative); experimental-design changes are intentionally deferred to a later PR.

Changes

Microplex → Populace migration

• Renamed \microplex → \populace across all section sources, macros.tex, the LaTeX root (main.tex), and the bibliography; corrected the surrounding context to Populace's actual architecture (one weighted Frame carried through modular package stages, calibration as one swappable component) instead of the old "declarative engine + country pack" model.
• Light, accurate reframe of the data.tex engine subsection only (retitled "Populace as a construction pipeline"); the \fbox figure placeholder and all \tbc markers are kept for the later detailed pipeline pass (companion PR Add Populace pipeline overview brief and figure #5).
• Updated Data-and-code-availability (main.tex) and README.md to point at PolicyEngine/populace as current, with microplex/microplex-us noted as archived migration references. Renamed the uncited microplex2026 bib entry to populace2026 (now cited).

Proof-of-concept reframe

• Abstract and introduction recast as a proof of concept on Populace's current target surface, with subnational/local-area calibration framed as the production motivation and future application rather than completed work. The four-baseline comparison and all experimental-design specifics are unchanged.

Background: gradient-descent calibration

• Added a paragraph to the calibration subsection (retitled "Classical calibration estimators" → "Calibration estimators") presenting calibration as an optimization solved numerically: Deville–Särndal optimization framing (GREG/raking as closed-form cases), global/range-restricted optimization (espuny2018), minimum-distance reweighting in tax microsimulation (creedy2003), and stochastic gradient descent with Adam (kingma2015), with Woodruff & Ghenis (2024) as the microsimulation instance.
• Wired in the previously-orphaned kingma2015 (Adam) citation.
• De-duplicated the "Microsimulation dataset engines" subsection so it no longer restates the gradient-descent mechanics — it now refers back and keeps only its unique content (PUF→CPS imputation, the dropout → L0/Hard Concrete contrast, and the Populace engine description).

README

• Added the IMA 2026 presentation link (presentation July 1, 2026).

Deferred to later PRs (per scope discussion)

• Inserting PR Add Populace pipeline overview brief and figure #5's actual pipeline figure and detailed stage descriptions into data.tex.
• Experimental-design/results changes (budgets, baseline configs, metrics, any narrowing of the comparison, results tables).

Test plan

• cd paper && latexmk -pdf main.tex builds cleanly (exit 0), no undefined citations or control sequences; populace2026, espuny2018, creedy2003, and kingma2015 all resolve in main.bbl.
• grep -rnI '\\microplex' paper/ returns nothing; remaining literal "microplex" strings are only the intentional archived references.

Closes #1

🤖 Generated with Claude Code