Correction policy

What was wrong: The fiscal pace calculation compared one quarter's deficit-to-GDP ratio with the same quarter a year earlier, while its registered convention called for a rolling four-quarter comparison. The quarterly observations themselves were valid, but they did not implement the stated pace convention.

What it reads now: The pace test now divides the sum of net lending or borrowing over four quarters by GDP over those same quarters, then compares it with the preceding four-quarter window. It reads -1.48 per cent against -1.23 per cent. The claim names both windows and the accounting basis.

Effect on published claims: On the refreshed evidence, both comparisons indicate deterioration; the corrected rolling-year percentages differ from the quarterly percentages. The quarter-to-quarter narrowing described in this reading and the revised fiscal-year estimate answer different questions and remain valid.

How it was verified: Recomputed both windows from the same quarterly National Accounts net-lending/borrowing and nominal GDP records. On the refreshed evidence, the ratio of four-quarter sums is -1.481363 per cent for 2025 Q3–2026 Q2 and -1.234929 per cent for 2024 Q3–2025 Q2. The separate single-quarter ratios are -1.686118 per cent for 2026 Q2 and -1.598002 per cent for 2025 Q2.

See the corrected reading on Now →

What was wrong: The second reading's evidence tables were left on the first reading's vintage. The 2026-08-04 pass advanced this Brief's prose and its Scenario inputs to the twelve months ending June 2026, and advanced both trade Resolutions under an explicit supersession chain, but did not move the Brief sections' own pinned value and comparison identifiers off the first reading. Every evidence table on the published page therefore showed the twelve months ending May 2025 and May 2026 -- each label appearing 86 times, in 43 table cells and 43 list items -- under prose reading the June windows, and under a section role sentence describing them as 'the two latest twelve-month windows', which they were not.

What it reads now: The Brief carries 174 pin paths: 157 value pins across its sections and Scenarios, and 17 comparison pins. 143 of them name the two Resolutions this reading advanced, and all 143 now name the current vintage, so the evidence tables show the twelve months ending July 2025 and July 2026 -- the windows this reading actually states. The other 31 name Resolutions this reading did not advance and correctly did not move. Corrected again on 2026-09-04 after independent examination: the first version of this record said 'all 118 section and Scenario pins', which was wrong in its count and wrong in the word 'all' -- seven calendar-year pins were still naming the first reading's publication, under a role sentence reading 'the latest full calendar year'. No figure was wrong, because the issuer has revised no month of 2002, 2024 or 2025 across the three vintages this record holds; those seven pins now name current-vintage values, and the test battery asserts the equality that makes rolling them provenance only. The first and second readings' values remain pinned, unchanged, on their own superseded publications.

Effect on published claims: The central reading did not change.

Canada's exports and the United States: the record break, and where the difference went → · See the Receipt →

What was wrong: Each of the five 2026-08-04 publications states its extent as 'Reference months 1997-01 through 2026-06'. The bytes those publications pin hold 1997-02 through 2026-06, not 1997-01. The capture is taken at a fixed count of the latest published months, so its window advances at its recent end and recedes at its early end with each vintage: the 2026-07-17 capture holds 1997-01 through 2026-05, the 2026-08-04 capture 1997-02 through 2026-06, and the 2026-09-04 capture 1997-03 through 2026-07. Only the first of the three strings was ever true. The overstated string renders 24 times on the published record. Closed 2026-09-06: the capture no longer works this way. From that date it asks for the whole published series instead of a fixed count of the latest months, so its window can no longer recede, and a full-extent capture taken the same day holds 1997-01 through 2026-07. The account above is left as it was published, because it is what was true when this correction was made.

What it reads now: The superseded string is not rewritten: a superseded publication's own text is preserved as it was published. The correction is recorded here and in the errata, the current publications state their extent from their own bytes as 'Reference months 1997-03 through 2026-07', and this entry's subject now states the recession itself in plain words, so a reader is told that the earliest months of 1997 are held on the retained earlier captures and not on the current one. Closed 2026-09-06: the subject now also states that the receding window was fixed on that date, that every capture from then on carries the whole published series, and that a full-extent capture is archived beside these bytes and agrees with them on every month the two share.

Effect on published claims: None. Every value any of these publications pins is a twelve-month window ending in 2025 or 2026, or a calendar-year sum for 2002, 2024 or 2025 -- all wholly inside the captured span on every vintage. The overstated month is January 1997, which no published figure reads.

How it was verified: Read directly from the archived bytes each publication pins, at the sha256 the publication itself records: the first and last reference periods of every partner series in statcan-12100011-exports-bop-sa-partners--20260717.json, --20260804.json and --20260904.json. No archived byte was refetched and none was altered.

The United States' share of Canada's goods exports: the monthly Balance-of-Payments record, 1997 through July 2026 →

What was wrong: The capture caveat described the item-level counter-tariff schedule as '893 rows in the new 2026-09-08 table alone ... two further, larger tables besides'. One of the two further tables is smaller, not larger: the preserved page carries three tables of 893, 313 and 1,814 rows. The row count this record actually transcribes is unchanged and was never wrong; only the characterization of the two tables it does not transcribe was.

What it reads now: The clause now reads 'beside two further tables of 313 and 1,814 rows', stating both counts rather than characterizing them.

Effect on published claims: None. No pinned value, rate tier, date, scope figure or quotation in this entry changes, and no claim anywhere on the record rested on the relative size of the two untranscribed tables. The disclosure the clause exists to make -- that this record does not transcribe the item-level schedule -- is unchanged and is now more precise.

How it was verified: Recounted programmatically from the preserved copy of the Department of Finance Canada complete-list page this entry already pins -- sha256 6ed901ab58e679a7941f10383b9354aaff23dfdece8e8eaa11f56e7856041b43, 1,702,498 bytes, matching this publication's own archive record exactly and unchanged. Counting the row elements inside each of the page's three tables returns 893, 313 and 1,814. No archived byte was refetched and none was altered.

The 2026 United States Section 338 tariff escalation on Canadian goods: the operative rate, predicates, and dates, from the primary documents →

What was wrong: The reading's reason said construction's pace against CMHC's benchmark 'eased from just over half to just under it'. It eased, but it stayed over half in both quarters: the four-quarter housing-starts total ran at 55.01 per cent of the CMHC benchmark in 2026 Q1 and 53.36 per cent in 2026 Q2. The parallel clause about the Parliamentary Budget Officer's benchmark -- 'from above nine-tenths to just under it', 90.67 to 87.95 per cent -- was correct, and its correctness is what made the CMHC clause read as a matching crossing rather than as loose wording.

What it reads now: The clause now reads 'eased but stayed just over half'. The Parliamentary Budget Officer clause is unchanged.

Effect on published claims: None. No figure moved, the Monitor's housing row still reads behind both independent benchmarks, and the finding this reading carries -- that the gap against each widened on the newest quarter -- is unchanged and was never in question. What was wrong was a threshold word, not a measurement.

How it was verified: Recomputed from this record's own published housing-starts series -- the same series the Monitor's housing row reads. Four-quarter starts were 262,930 at the first quarter of 2026 and 255,061 at the second, against CMHC's benchmark of 478,000 and the Parliamentary Budget Officer's of 290,000: 55.01 per cent and 53.36 per cent against CMHC, 90.67 and 87.95 per cent against the Parliamentary Budget Officer. This record's own published benchmark reading independently states 53 per cent for the latest quarter. Falling to just under half would have required a four-quarter total below 239,000.

See the corrected reading on Now →

What was wrong: The correction above named /big-questions/sell-beyond-the-us/ as an affected route, but the live Question Pulse behind that route was never itself updated when this correction landed 2026-08-04: its own reading text kept stating the two withdrawn superlatives ('the lowest in the twenty-nine-year monthly record, after the fastest one-year fall in that record', attached to the May 2026 window) for a further twenty days, until 2026-08-24. This was not a second, independent error -- it was the same withdrawn claim, in an object the 2026-08-04 pass's own fix did not reach.

What it reads now: The July Question Pulse is superseded by a new August dated reading, whose reading carries the corrected attribution (record low and fastest fall to the near-tied March/April 2026 windows, not May) forward, alongside the same day's tariff-escalation reading. The superseded Pulse's exact prior wording, uncorrected, remains on record in full history, superseded rather than rewritten.

Effect on published claims: The central reading did not change.

Canada's exports and the United States: the record break, and where the difference went → · See the Receipt →

What was wrong: Two false superlatives attached to the wrong month. The Brief's opening summary and its "The answer" section stated that the twelve months through May 2026 were 'the lowest twelve-month share in the twenty-nine-year monthly record', and that this window followed 'the fastest one-year fall in that record: 5.8 percentage points'. Neither held. On the same archived bytes the Brief was published from, the May 2026 window (69.128 per cent) was the THIRD lowest, behind April 2026 (69.023) and March 2026 (69.025); and its one-year fall of 5.78 percentage points was the FOURTH largest, behind March 2026 (-7.14), February 2026 (-6.81) and April 2026 (-6.67). The comparison the prose drew against 2007 ('the largest one-year fall before this episode was 4 points') was correct and is unchanged. The error was editorial, not arithmetical: no pinned value was wrong, and the Brief's own instrument plotted the correct series beside the incorrect sentence.

What it reads now: The Brief now names the record low as a two-window near-tie rather than a single month: the twelve-month windows ending March and April 2026 both read 69.01 per cent, 0.004 percentage points apart, a gap smaller than the revision each of those two windows underwent between the July and August 2026 captures (0.014 and 0.016 percentage points) -- and the record's fastest one-year fall (7.16 percentage points) is attributed to the window ending March 2026, not to May. The current reading, the twelve months through June 2026, is stated as what it is: 69.07 per cent, just above that low, after a one-year fall of 5.41 percentage points. The flagship conclusion is unchanged, and the episode is if anything more severe than published: the true fastest fall (-7.16pp) is larger than the -5.8pp the Brief claimed as the record.

Effect on published claims: The central reading did not change.

Canada's exports and the United States: the record break, and where the difference went → · See the Receipt →

What was wrong: The published evidence gap 'What would Ring of Fire mining loads require of Ontario's northern grid?' rested its account of the public record on a single document from the previous regional planning cycle — Hydro One's Northwest Ontario Needs Assessment of 2020-07-17, which recorded that the need for new wires to accommodate the Ring of Fire 'has not yet materialized' and required 'no actions required at this time other than to continue to monitor the Ring of Fire development' — and characterised the resulting absence as 'an institutional decision on the record, not an oversight'. The quotation was accurate and remains accurate for its date. The framing was not: it presented a 2020 monitor-only posture as the standing institutional position, when the position had since changed.

What it reads now: The gap now states the current record: a 230 kV line toward the Ring of Fire gateway is a designated provincial priority with a named builder and an expected 2032 in-service date, the current regional planning cycle does not name the Ring of Fire at all while calling the region's mining and data-centre demand urgent, and no public document states how much electricity Ring of Fire development would draw. The finding is strengthened rather than withdrawn: the missing quantity now sits behind a construction commitment instead of behind an absence of interest.

Effect on published claims: The central reading did not change.

Claims on the grid: what Ontario's connection record proves about its largest new loads — and why the register they sit in cannot be added up → · See the Receipt →

What was wrong: The Brief stated a limitation more absolute than the public record supports. Its executive answer said that of the withdrawn 431 MW Alliston application 'nothing public preserves how that state came about'; its key uncertainty said that with one captured register state 'every status is a snapshot and no entry, exit, or status change is yet observable', and that the record of change would begin only with Countability's second monthly capture. The register's own behaviour was described correctly -- the issuer does overwrite the file in place and publishes no history -- but the conclusion drawn from it was too strong, because a third party archives the same file.

What it reads now: The Brief now shows the transition it previously said it could not: two Alliston applications recorded live in a prior vintage and withdrawn in the current one, with the issuer's own committed-flag rule quoted against the rows that break it. The limitation is restated accurately -- the web archive is captured irregularly by a third party, so it supports specific comparisons rather than a continuous record, and a continuous record still begins with Countability's successive captures.

Effect on published claims: The central reading did not change.

Claims on the grid: what Ontario's connection record proves about its largest new loads — and why the register they sit in cannot be added up → · See the Receipt →

What was wrong: Two overbroad public claims. (1) The Brief's opening summary, its "The answer" section, and "What this means for the decision" stated or implied an unqualified absence of precedent ('a pace with no precedent in the record', 'a growth regime the record has not yet shown') using only the full-period 2005-2023 endpoint-average as the comparator; the wording did not name that the comparison was a full-period average, and did not address shorter historical intervals. (2) The same three places stated 'measured generation has not yet turned' / 'delivered supply has not moved' without qualification, omitting that calendar generation increased from 2023 through 2025 before its freshest trailing-twelve-month window declined.

What it reads now: Every scenario requires a sustained 2023-2035 pace above every comparable twelve-year stretch in the CER's 2005-2023 record (generation: roughly 1.7 to 3.1 times the fastest such stretch, 2010-2022; capacity: roughly 1.6 to 2.6 times, 2011-2023), and several times the full-period average. Generation increased from 2023 through 2025, but its freshest trailing-twelve-month window did not sustain that increase, and no durable, scenario-consistent acceleration is established yet.

Effect on published claims: The central reading did not change.

Canada's electricity expansion: the projected pace and the delivered record → · See the Receipt →