The fifth re-freeze, and the smallest so far in the series and the largest so far in the headline multiple. Four terms were read in the Platsbanken archives before removal rather than inferred from their year profile, and a fifth was narrowed rather than removed. The whole series moves by 1.5%, but nearly all of it lands on the sparse base years, so the published multiple moves by 14%. The full twenty-year series was reprocessed on it on 19 August 2026.

• `klustring` is removed. It is Swedish server clustering, verified in the archives beside Active Directory, DNS and GPO. Bare English `clustering` was removed at v1.4 for the same sense, so the two halves of the lexicon had been out of line since 17 August. It carries 4.77% of the pooled 2006-08 base and 0.185% of 2025.
• `tts` is removed. Its 219 hits are TTS Marine ASA, BorgWarner TTS AB and Technical Sales Specialist, and not one is speech synthesis.
• `moses` is removed. All 87 hits from 2019 to 2025 are Moses Bil & Lack Tollarp AB, a single car body shop, rather than the statistical machine translation system.
• `mapreduce` is removed, and not as a homonym: every hit means MapReduce, and MapReduce is data engineering rather than AI. It always appears inside a Hadoop, Spark or Kafka list. Attribution finds it removed no advertisement at all in 2006, 2008 or 2025, so the removal is correct on the evidence and inert in the series.
• `ml` is GUARDED rather than removed. It now counts only within 150 characters of an AI-specific anchor, or when bound to a technical noun. All 21 base-year matches were false: company names, reference numbers, a query string, a Malmö school programme, SCB's own laboratory and one Standard ML. The guard keeps 91% of 2025 `ml` matches, and the ~9% it drops are genuine. Two alternatives were rejected and are recorded in the matcher with why they lost: a proper-noun discriminator, worse on both axes, and a year gate, rejected on principle because a date inside the definition makes the measure ask whether a term was fashionable yet.
• The series falls 1.5% and the multiple rises 14%, because the corrections clean the base far harder than the endpoint: whole-text 2006 falls 12.9%, 2007 15.6% and 2008 17.9%, against 0.5% for 2025. The rise from a pooled 2006-08 base to 2025 therefore goes from 32-fold to 38-fold, and the floor from 37-fold to 42-fold. The bare-AI band and the generative addendum are unchanged to the advertisement, which is the check that all five interventions stayed inside the core lexicon.
• A second anchor is published beside the headline (Magnus, 25 August 2026). The 2006-08 base is the striking figure and the fragile one: it rests on 190 flagged advertisements and has moved at three consecutive freezes. A 2015-17 base rests on 2,666 and moved 3.5% at this freeze against 17.8% for the pooled early base, so the page now carries both: 38-fold since 2006-08, and 6.2-fold (95% interval 6.0-6.4) since 2015-17. One figure is large and fragile, the other small and stable, and a reader is entitled to see which is which.
• Base-year sensitivity worsened slightly and is reported rather than omitted. Moving the base one step, 2006 to 2007, moves the v1.5 figure by 4.1% against 1.0% under v1.4, because 2007 lost a larger share of its base than 2006 did. Pooling the base over 2006-08 addresses the sampling noise, taking the 95% interval from 29-49 to 33-44, but it cannot address definitional churn: the same false friends sat in all three early years. The multiple is the most fragile number the Monitor publishes, and it rests on 190 flagged advertisements.

Validation. Every removed term was read in the archives first, in context, and the count of its hits established before the decision. The pre-freeze estimate of base contamination covered `klustring` and `ml` at 10.05% of the pooled 2006-08 base; the realised loss is 15.6%, and attribution attributes the difference to `tts`, which carries 8 base-year advertisements in 2006 and 4 in 2008, and `moses` with 1. The estimate was a subset of the interventions rather than a mismeasurement of them.

The ceiling correction is NOT re-derived on this freeze. v1.5 leaves the bare-AI band unchanged to the advertisement, so the v1.4 derivation of 12.8% applies unaltered. The vocabulary module lags this freeze: its term composition is rebuilt from a candidate extract, and that extract was still being rebuilt under v1.5 when this version shipped.

The fourth re-freeze, and the first whose net effect is to LOWER the series and STEEPEN the trend. It adds the ML abbreviation on the recall side and removes three terms that are genuine AI vocabulary elsewhere and false-positive generators in Swedish advertisements. The full twenty-year series was reprocessed on it on 17-18 August 2026.

• The ML family joins the core: bare `ML`, the `ML-`/`ML/` compound, and the glued spelling `machinelearning`. It enters as a case-sensitive pattern rather than a term in the list, because the matcher is case-insensitive and lowercase `ml` is millilitres in a corpus carrying a great deal of care, food and laboratory work. Two larger candidates measured at the same time, `data scientist` and `datavetare`, were REJECTED: such a post is often statistics or business intelligence rather than AI. A screen of the 555 advertisements `data scientist` would have added in 2022 puts their AI share at 5.0%, against about 13% in the bare-AI band those advertisements sit beside, so the candidate would have bought volume rather than recall.
• Bare `språkbehandling` becomes `naturlig språkbehandling`. In ordinary Swedish the bare word means command of language, and 14 of its 15 occurrences across 854 advertisements are lawyers, HR officers, telephonists and journalists. Every one of the 13 advertisements it dropped from the early floor stratum was hand-labelled as not AI, and floor precision on that stratum rises from 32.5% to 48.1%.
• Bare `clustering` and `image processing` are removed as corpus false friends, joining `torch`. Clustering of standard errors and server clustering are what the bare word means in most Swedish advertisements; image processing is classical signal processing that predates machine learning and is still practised without it. The specific senses survive in both cases: hierarchical, spectral and single-linkage clustering, cluster analysis, k-means, computer vision, image recognition and object detection. Headline core 279 to 277.
• Both lines FALL and the trend STEEPENS, which is the opposite of what the freeze was expected to do. The whole-text measure loses 4.5% of its advertisements over the series and the floor 3.6%, but almost all of it is at the sparse left edge: 2006 loses a third of its whole-text count while 2025's floor RISES by 20 advertisements on the ML additions. The rise from a pooled 2006-08 base to 2025 therefore goes from 22-fold to 32-fold on the whole-text measure and from 27-fold to 37-fold on the floor. Attribution on 2006 and 2015 puts `språkbehandling` at three-quarters of the 2006 loss and three-fifths of the 2015 loss, so the steeper trend is the correction of an understated one rather than an artefact: the old figure was measured against a 2006 base inflated with advertisements for people who write well.
• The steeper number is also the more stable one. Moving the base year one step, 2006 to 2007, moved the v1.3 figure by 24 per cent; the same move under v1.4 changes it by 1 per cent. The published base is nonetheless POOLED over 2006-08, because a single year of 70 flagged advertisements is a thin estimate: pooling tightens the 95% interval on the multiple from 26-41 to 28-37 while moving the answer by 0.1.
• The bare-AI dot pattern now requires balanced dots. `a\.?i\.?` made the two dots optional independently and so also matched `a.i`, which is what a lost sentence space looks like after text extraction. It moves no published series, the bare band being a diagnostic, but it does move the BAND, and the band carries the ceiling correction.

Validation. The definition changes are validated on hand-labelled advertisements rather than on the gold set alone, because three of the four are removals whose evidence is the labelled bare-AI band. Of the advertisements the ML additions pull out of that band, 62% are hand-labelled AI against 13% in the band as a whole; of the 19 the balanced-dots fix removes, none is AI. The gold set itself is unchanged by this freeze and still speaks only for 2024.

The ceiling correction is re-derived on the v1.4 band and is 12.8% (26 of 203 surviving hand labels), against 13.9% on the band the sample was actually drawn from. That sample came from a candidate extract built under v1, three freezes earlier, which is a defect in the sampling frame rather than in any published number: every re-reading sits inside the published interval. The extracts now record the definition that produced them and the extractor refuses to run across a definition change.

The third re-freeze. It closes the two lexicon defects v1.2 carried openly, and widens the fingerprint again to cover the pipeline source. The full twenty-year series was reprocessed on it on 7–8 August 2026.

• Plural forms are now matched. Coverage in v1.2 was an accident of which plurals the three source lists happened to duplicate: neural networks matched because the plural is a separate literal entry, while 12 of 24 English multi-word core terms missed theirs. Plurals are generated for every multi-word term plus the named singles llm and chatbot. Multi-word phrases are safe to pluralise mechanically, since a phrase that does not occur in the plural simply never matches; single tokens are not, so they are enumerated instead.
• Bare boosting and bare torch are gone. A parenthetical in the source lists is now split only when it is an acronym: X (machine learning) is a disambiguator, and splitting it had made bare boosting a term matching boosting traffic and boosting sales. Separately, torch is Alekseeva et al.'s entry for the Lua machine-learning framework and matches welding equipment in Swedish advertisements, so it is removed as a corpus false friend. PyTorch is unaffected.
• Both lines move, unlike v1.2 which moved the floor alone. The floor rises from 17,243 to 17,659 advertisements (+2.4%) and the whole-text measure from 37,283 to 38,281 (+2.7%); the raw bare band falls by 136, because an advertisement asking for LLMs was sitting in it purely because the lexicon could not see the inflection. The advertisement count, the adjacent band and the entry-level series move by exactly zero.
• 2006 to 2009 do not move at all, in any measure, which is why the growth figures rise only at their recent end: the whole-text rise since 2006 goes from 21.4-fold to 21.8-fold and the floor from 26.8 to 27.5. This is a change of SHAPE as well as level: 2020 loses 84 whole-text advertisements, the largest single-year fall, while 2022 gains 258, almost all autonomous vehicles.
• The fingerprint now also covers the pipeline source, which holds the deduplication key, the negative-context window and the entry-level regex — all printed as rules the reader is told they can reproduce the number from. A reviewer had changed two of them by mutation with the fingerprint staying green. The pipeline also carries the fingerprint stamp itself, so the hash deliberately excludes those two lines: it covers the rules, not the record of what the rules hash to.

Validation. 88.6% precision [74.0, 95.5] and 37.3% recall [27.7, 48.1] against all AI positives on the 369-advertisement hand-labelled set, with Wilson 95% intervals. The pre-registered gate — recall up, precision not below the incumbent — is met, but both moves sit well inside their intervals, so we claim only that the freeze has not degraded the measure, never that it improved it.

The hand-labelled set is still drawn entirely from 2024 advertisements, so every validation figure describes one year of a twenty-year series. An early-period batch of 120 advertisements from 2006–2013 is drawn across three strata and awaiting labelling; until it is done, the left edge is unvalidated and said to be.

The second re-freeze, correcting a filter fault that suppressed the floor, and widening the fingerprint to cover the matcher source as well as the configuration. The full twenty-year series was reprocessed on it on 5 August 2026.

• The teaching-about-AI filter tested a hard-coded set of four patterns and, when an advertisement had matched an AI term outside that set, fell through to discarding it. A doctoral post requiring PyTorch, TensorFlow and computer vision was thrown out because its benefits section contained the word kurser. The fault suppressed the floor, so every correction to it adds.
• Exactly one measure moved. Summed over all 22 periods the change is zero, not merely small, for the advertisement count, the whole-text measure and its components, the raw bare band, the adjacent band and the entry-level series: the fault sat only on the role-scoped path. The floor rose in every period, from 16,650 to 17,243 advertisements, +3.6%. The identical denominator in each year confirms the fix reclassifies advertisements and never drops one.
• Because the proportional gain is larger in the sparse early years, the floor's own growth from 2006 to 2025 falls slightly, from 53.9-fold to 52.6-fold on counts. The headline whole-text figure, a 21-fold rise, is untouched.
• The fingerprint now covers the configuration minus the hard-negative evaluation list, plus the matcher source. It previously hashed the configuration alone, which was wrong both ways: editing a marker in the matcher changed every published number while the fingerprint stayed identical, and editing the evaluation list forced re-freezes that changed no number.

Validation. 88.2% precision [73.4, 95.3] and 36.1% recall [26.6, 46.9] against all AI positives on the 369-advertisement hand-labelled set, with Wilson 95% intervals. No difference between v1, v1.1 and v1.2 survives its interval, so we claim only that successive freezes have not degraded the measure, never that they improved it.

A known recall defect is carried openly rather than patched: plural forms are missed on many entries: llm and chatbot match in the singular but not the plural, and of 24 English multi-word core terms tested, 12 miss their plural. Measured cost on the 2025 archive is 35 distinct advertisements, about 1.6% of that year's floor. Its direction is the same as the fault above, inflating the bare band and depressing the floor. Fixing it would have changed the fingerprint and invalidated this reprocess, so it was deferred; v1.3 fixes it.

The first re-freeze. Folds three corrections into one release, because all three biased the published series in the same direction and shipping them separately would have moved every number twice within a fortnight, once down and once up.

• The published counting unit becomes the distinct advertisement: headline, employer name and the first 400 characters of the body identical, deduplicated within year. Raw records, v1's unit, are retained in every output file as a robustness line. Repeat postings run from 13.1% of records in 2008 to 48.9% in 2023.
• Four bare product names are date-gated to the date the product came into existence: copilot from 29 June 2021, claude from 14 March 2023, llama from 24 February 2023, gemini from 6 December 2023. Before those dates the tokens match an aviation co-pilot, a given name, an animal and a Swedish company. Multiword forms such as github copilot are unambiguous and ungated, and a gated term is dropped only where the advertisement rests on nothing else.
• Four polysemous terms (prompting, finjustering, bare gpt, bare llm) establish an advertisement on their own only from a stated convention date, and only where no negative-context marker appears within 120 characters. Unlike the product names these words existed throughout and merely acquired an AI sense, so the date is an explicit convention rather than a fact. A strict variant, in which those terms never establish an advertisement alone, is computed in the same pass and published as a robustness bound.

Validation. 90% precision, 34% recall against all AI positives on the 369-advertisement hand-labelled set. The matcher code is byte-identical to v1, so v1.1 is a post-filter over v1's own match sets; every term it touches lies in the generative-AI increment rather than the citable core, leaving core-only results comparable across the freeze.

The gold set is entirely 2024 advertisements, so every product gate is open by construction and the anachronism fix cannot be tested on it. Its evidence is the pre-2009 concentration of the affected advertisements: 301 advertisements between 2006 and 2022 had been classified as AI on no other evidence, a fifth of the 2006 series and under one per cent from 2017.

The first frozen definition. Held core plus the generative-AI addendum, the governance cluster, and three matcher repairs.

• Generative-AI addendum and governance cluster admitted through the admission sequence.
• Matcher repairs: AI/ML folded into the core, multiword plurals, and the glued form generativ-ai.
• Role scoping introduced, restricting the floor to the part of the advertisement describing the role's own tasks and requirements.
• Staged rather than merged: the AI-doing phrase family, the user-tier competence phrases, fleragentssystem, semantic search, edge ai. These remain on the watch list pending validation on a grown gold set.

Validation. 88% precision and 34% floor recall against the 370-advertisement hand-labelled set, against 83% and 4% for the pre-freeze baseline on the same set. The gate was that recall must rise and precision must not fall: met.

Predecessor: held snapshot 23f23bbc32baa0fd.