Skip to content

Scope decision pointers (Terms of Reference → DDSL proposals)

This page is a link target for the presentations in docs/presentations/. It records the scope decision points (DPs) we are treating as “in view” when writing Phase‑1 proposals, so the decks can become a finalized scope/proposal set without rewriting prose yet.

Source: Econ‑ARK DML Terms of Reference (ToR) v0.1.0a — especially Scope (Track A) + Phase 1.1 deliverables.

YAML‑first / file conventions (DP‑2.*)

  • DP‑2.1 (canonical text format): YAML‑first.
  • DP‑2.2 (file/layout conventions): imports/includes/versioning/comments.
  • DP‑2.4 (relationship between model files and “problem chunks”): what is authored per‑stage vs per‑period vs per‑model.

“Problem chunks”: stage/period/events + composition (DP‑3.4, DP‑3.6)

  • DP‑3.4 (problem chunk structure): decide whether the primitive modular unit is a stage/period/events object; whether graphs are involved.
  • DP‑3.6 (problem chunk composition): how chunks link (connectors/twisters; graph/category composition).

Mathematical problem statement vs solution statement (DP‑3.5)

  • DP‑3.5: explicitly distinguish “problem statement” vs “solution statement” and how each appears in the language.

Model representation boundaries (DP‑4.*)

  • DP‑4.2: clear boundaries between symbolic model / IR vs numerical instantiation.

Numerics are separated (DP‑5.*)

  • DP‑5.2: how computational methods are specified (grids/quadrature/interpolation) while keeping the economics object clean.

Factorizations in scope (DP‑7.*)

  • DP‑7.1: formal operator statements (Bellman / expectation / etc.).
  • DP‑7.2 (allowable factorizations): ADC vs primal vs Q vs other—this is an explicit scope decision point.

Information structure restriction (presentation scope guardrail)

This is the restriction used in the decks:

  • Observable sufficient statistics (not full POMDP belief‑state): we represent within‑stage partial observability via perches as information sets, while avoiding general belief‑state POMDPs.

This is the “tractability constraint” that keeps policies as \(x \mapsto a\) rather than beliefs \(\mathcal P(S)\mapsto a\).

Phase‑1.1 exemplar models (proposal scope)

The ToR frames Phase‑1.1 proposals around concrete exemplar classes:

  • Buffer‑stock model (ADC factorization prototype).
  • Finite‑horizon / lifecycle / non‑stationary model.
  • Nested structure (e.g., finite‑horizon with stopping times inside an infinite‑horizon structure).
  • Portfolio choice appears explicitly as a benchmark for backend support / minimal Dolo changes.