Corrections

Two placeholder prices of Bs 2.00 entered the index

What was published

From 29 September 2026 two personal-care listings at one chain posted a price of Bs 2.00, in stock, against their previous prices of Bs 5,762 and Bs 7,340. The index chained both falls, and the paid terminal listed them among its biggest movers over 7 days and over the full history.

What was wrong

A Bs 2.00 price on a Bs 7,340 product is a placeholder, not a shelf price. The existing safeguards missed it: the stock-out rule only removes prices marked out of stock, and the large-move flag sends a move to review rather than removing it.

What we did

A price now counts as a placeholder when it sits 10 times or more away from the product's own usual price, in a product whose price jumped 10 times or more from one observation to the next. Among prices the rule does not flag, the largest move in the panel's history is about 5 times in a day. Placeholders are left out of the index like an out-of-stock price, are never used as either end of a biggest-movers comparison, and are shown marked on the paid basket and product card instead of hidden. Applied to the full history it removes 18 prices across 5 products; three of the five were already excluded by hand since 21 September. The bolívar headline moves on three dates from 29 September 2026, up 0.03% at 1 October. The only daily figure that changes is 29 September, from +0.08% to +0.11%. The dollar headline does not move. No collected price was changed or deleted.

Every product and date treated this way is published with the index data. A placeholder that runs for more than half of a product's history defeats the rule and is still handled by hand.

See the methodology →

A chain was dropped from the index for not raising prices

What was published

Since 10 August 2026 the index excluded any chain that had repriced too few of its products over the previous thirty days. The rule was published as a safeguard against catalogues that had stopped being maintained, and the index history was restated under it.

What was wrong

The rule selects chains on the outcome the index exists to measure, and the bias runs in one direction only. A chain that held its prices is a chain posting zero inflation; removing it leaves a panel of the chains that went up, and the published number reads high. It also could not tell two different things apart: a page nobody is updating, and a live shop that simply is not raising prices, which in a dollarised retail market is one of the more interesting things this index can find.

What we did

The rule is withdrawn. A chain is now dropped only when its catalogue shows no maintenance at all: no price change, no stock change, no change in the discounts it runs. Every chain in the panel passes that test, so nothing is excluded today, and flat prices count as the zeros they are. Applied to the full history: both headline series move on thirteen dates from 5 September 2026, the bolívar headline downward by 0.25% at 17 September and the dollar headline upward by 0.09%. Two category series that the old rule had ended or never started are back. No collected price was changed or deleted. Shipped 17 September 2026.

Per-chain price rigidity is still measured and still reported. It is a finding about Venezuelan retail, not grounds for removing a shop. One separate defect was found while checking this and is logged rather than fixed: one chain's list price is a static manufacturer's price while the price it actually posts moves. It sits outside both headline series.

See the methodology →

Weekly issues Nº1 to Nº11 reissued

What was published

All eleven weekly issues described the pace with a word the report itself chose. Nº11 said “cooling”, and its “annualised monthly pace” was the whole-history average, +9.9%, rather than what the name claims.

What was wrong

The word was decided by the report and not by a rule. The “annualised monthly pace” was not what it claimed to be, and the movers tables were computed outside the issue's own window, with out-of-stock listings and ineligible chains inside them.

What we did

The word is no longer chosen by the report; it is decided by the pre-registered rule in §6.6 of the methodology, and under that rule the pace in Nº11 held rather than cooled. “Annualised monthly pace” is now what it claims to be, the last 7 days projected to 30, +12.7% in Nº11. Movers tables are computed over the issue's own window, in-stock listings only, index-eligible chains only. All eleven issues were reissued; issues 1 to 10 also print the series as restated on 10 August and 1 September 2026. Shipped 15 September 2026.

No index figure in Nº11 changed. The errors were in which number was labelled as what, and in which word was used. The original copies are archived unmodified.

See the weekly issues →Read on the newsletter ↗

The oil production consensus gains a third leg

What was published

The oil terminal published a production consensus built from agency estimates and from press relays. The August 2026 consensus was 1,153 kb/d.

What was wrong

The relays (Kpler, CEIC) are not agency estimates: they repeat third-party figures and were entering the average as if they were a source of their own. And months the EIA's official series has not yet printed were left with no current estimate at all.

What we did

The EIA's STEO forecast was added to the consensus, recorded only for months the EIA's official series has not yet printed, and Kpler and CEIC were removed. The August 2026 consensus moved from 1,153 to 1,162 kb/d, with IEA 1,160, OPEC secondary sources 1,145 and STEO 1,180. Shipped 15 September 2026.

PDVSA's self-reported 1,201 kb/d is still shown separately, never inside the consensus. No VPI figure is involved.

See the oil terminal →Read on the newsletter ↗

Movers tables exclude stock-outs and ineligible chains

What was published

The movers tables on the dashboard, in the weekly and in the paid terminal included listings marked out of stock and chains that are not eligible for the index.

What was wrong

A Farmatodo listing that went from Bs 3.00 to Bs 17,861.00, out of stock all week, had been showing as the “biggest riser”. A price nobody could pay was topping the table.

What we did

Everywhere (dashboard, weekly and paid terminal) the movers tables exclude listings marked out of stock and chains that are not eligible for the index. Shipped 15 September 2026.

No index figure changed. Out-of-stock listings never entered the chaining; they entered the table printed next to it.

See the terminal →Read on the newsletter ↗

Multimax is not eligible for the index

What was published

The “current state” paragraph of §3.4 of the methodology still described Multimax as part of the panel, and six weekly issues showed Multimax rows in their movers tables.

What was wrong

Multimax has been out of the index since 3 September 2026, through the 1 September shelf-price restatement: on that basis its catalogue reprices far below the breadth floor. The methodology said otherwise.

What we did

The “current state” paragraph of §3.4 was corrected and the six weekly issues still citing Multimax rows were regenerated. Shipped 15 September 2026.

No index figure changed. The eligibility gate applies to the whole history, so the series already excluded Multimax while the tables were still citing it.

See the methodology →Read on the newsletter ↗

One chain's catalogue was its Colombian storefront

What was published

Between 3 July and 4 September 2026, one of the panel's chains was collected from its Colombian online store rather than its Venezuelan one. Published counts of products and of chains included those rows.

What was wrong

We queried the wrong catalogue index at the chain's search provider, which runs a single application for all of its markets. Most of the barcodes we collected were registered in Colombia and the prices were in pesos, a stable currency, while the bolívar index was rising.

What we did

Those rows never entered the index calculation: the source was excluded from the chain by the eligibility gate, which drops a source from every day's rows and not only from the judged window. Verified before and after removal: the bolívar and dollar series are identical. What was overstated is the count of products and of chains. The rows were archived, not deleted, and the source restarted on 4 September on its Venezuelan catalogue.

No collected price row was modified. The error was ours, not the chain's: their Colombian storefront was working correctly and we queried it.

Index restatement: shelf price instead of price paid

What was published

From 3 July to 31 August 2026 the index chained the price actually paid, which is the promotional price whenever a promotion is running. The headline moved with how many promotions each day's collection happened to catch.

What was wrong

Two things. The first, measured: on a fixed 267-product basket from one chain, 15 to 18 August, the shelf price rose smoothly while the paid price whipsawed, tracking promotion incidence. The second, procedural: the restatement notice §8 requires was not published when the change took effect. The restatements of 10 August and 4 September both have one. For eight days, the largest restatement this series has made was the only one without a notice.

What we did

From 1 September 2026 the index chains the posted shelf price, applied to the full history. On closed windows recomputed on 8 September: the 30-day bolívar change to 22 August moves from +4.67% to +5.82%, and the annualised pace at 24 August from 7.95% to 6.59%. Eight live posts quote figures from the old construction. Each one gets a correction in its own thread, posted from 8 September onward and spaced by the minimum interval between posts; none is deleted. This entry is updated when the batch finishes. The restatement notice was published on 8 September, dated the 1st and marked late.

The price paid is not discarded: it is a companion measure built by the same chaining machinery. The restatement did not flatter the index: it revised the 30-day bolívar change upward and widened the distance between the bolívar and dollar series.

The launch thread's “clean test” was not reproducible

What was published

The launch thread of 11 August 2026 presented as its core evidence: “the clean test is products with the same barcode sold by two chains, over the same window. Farmatodo: 0.00% / Locatel: 14.81%.” It was published as a test with n=10 and complete coverage.

What was wrong

The two chains share 5 barcodes, not 10, and only one of them has complete coverage of the window. Four of the five Locatel listings only exist from 4 August, so “the same days” is not true of them, and three of them fell. No construction of the test produces +14.81%. The figure was never supported by the database, not on the day it was published, and not since.

What we did

The figure was publicly retracted on 23 August 2026. All 54 live posts on the account were audited by recomputing every numeric claim; 12 posts were deleted, the thread was corrected and pinned, and the retracted figures were struck through in the methodology rather than removed. +14.81% must not be cited in any context.

What this retraction does NOT touch: the Farmatodo catalogue freeze is solid and independently verified (675 of 699 SKUs at exactly one price, checked live on 2 August 2026). The 10 August restatement stands on its own evidence; what failed was the test used to narrate it.

Index restatement: the source-eligibility gate

What was published

Between 3 July and 10 August 2026 the index was published while averaging into the panel a chain whose catalogue was effectively frozen.

What was wrong

The index understated inflation by roughly a factor of four. Every health check the pipeline had was passing: the defect was in the panel, not in the code. It was caught by an audit, not by the alarms.

What we did

A source-eligibility gate (§3.4) and a native-USD headline (§4.1) were introduced, both applied to the full history. The VES headline for 3 July → 2 August moved from +3.31% to +13.39%; the USD headline as of 10 August moved from −8.38% to −1.44%. Nothing was deleted and no price row was modified. Every figure published before that date under the old construction is superseded.

Lesson kept: audit the panel, not just the pipeline. It happened before any money changed hands.

An unmeasured day that cannot be recovered

What was published

There was no collection on 28 July 2026. The published series carries the last observed level forward for that day.

What was wrong

We publicly claimed to have measured “every day, without skipping one.” That was false for as long as the claim was live.

What we did

The claim was corrected. The gap is declared in the series and on this page; an unmeasured day is not recoverable and is not filled by interpolation.

Open in the sense that the data does not exist and never will.