Changelog

2026-09-04 — v2-weighted (expenditure-weighted headline)

The headline index is now expenditure-weighted instead of equal-weighted. The chained-Jevons index is computed within each subcategory (meat, bakery, dairy, produce, pantry, beverages) and the subcategories are combined by expenditure weight (a Lowe/Laspeyres aggregation), so the index reflects how households actually spend rather than how many SKUs happen to be in each group. Weights (share of the index): meat 29.4%, bakery 19.6%, dairy 13.4%, produce 13.4%, pantry 13.4%, non-alcoholic beverages 11.0%. Sources & caveats: the food vs non-alcoholic-beverage split (≈0.89 / 0.11) is from INDEC's ENGHo 2017-2018 (Alimentos 20.2% / Bebidas no alcohólicas 2.5% of total household spending); the split of the food share across food subgroups is an interim estimate reflecting Argentine consumption structure, pending INDEC's class-level IPC ponderadores, and will be revised when those are incorporated. Alcoholic drinks were removed (they belong to a different CPI division). Because some heavily-weighted groups (notably meat) are thinly sampled at a single online retailer, per-subcategory coverage is now published alongside the index. The equal-weighted index remains available for comparison via ?weighted=0. Methodology version: v2-weighted.

2026-08-27 — v1-chained estimator (index construction change)

The index is now built as a chained elementary index. Previously the base was frozen to the SKUs in stock on the series' first day, so any item that later went out of stock, was substituted, or was re-pinned could never re-enter the coverage count — coverage could only decay. Under the chained construction, each day is a link: for every item in stock that day we form its price relative to that item's most recent prior in-stock price (a standard “last-price” treatment for temporarily missing items), aggregate the relatives under the chosen estimator (Jevons / Carli / trimmed), and chain the daily links from a base of 100. New and substituted items now enter at their first observation and contribute change from the next period. The full series is re-derived from the existing raw observations — no data was deleted and no rebaseline was performed; because the construction changed, published index levels shift accordingly. Also this release: coverage floor is now proportional to the active basket (≥60%, absolute minimum 25), a lower-noise “accepted with warning” alert tier was added below the quarantine threshold, and the basket was expanded with stable packaged staples for margin. Methodology version: v1-chained.

2026-08-13 — pinned basket & rebaseline (data-quality hardening)

Item identity is now pinned to fixed product IDs: each basket item is resolved once and fetched by that ID every day, so the index composition no longer drifts as search rankings change. Added per-item outlier rejection (a single item moving >±40% in a day is dropped) alongside the existing daily coverage and index-move quarantine gates. To establish a clean, stable base with the pinned basket, the series was rebaselined to 100 on 2026-08-13; the two early development days were removed. Base ≈ 38 in-stock items.

2026-08-12 — live collector active

Real data now flowing. A daily collector pulls prices from Jumbo Argentina's public catalog (VTEX) into the index; the synthetic seed used during development has been removed. The series restarts at base = 100 on 2026-08-12; the 30-day rate populates as more days are collected.

2026-08 — v0-simple-avg (initial)

Initial release. Estimator: unweighted mean of daily price relatives, base = series start. Coverage: Argentina (Buenos Aires), food & beverages, ~40 SKUs. API and MCP endpoints published at api.oscar-lopez.com.

Planned

• Additional retailers (multi-store average) beyond Jumbo.
• v1 estimator: geometric mean within category + Laspeyres across categories.
• Citable methodology note with DOI (SSRN/Zenodo).
• Coverage expansion (additional categories/cities).