Every version, documented in public.

Versions are semantic. A patch corrects an error without changing published values. A minor version adds capability or data without invalidating existing results. A change marked Value-changing means previously published numbers were superseded, not corrected — cite them against the release they came from.

1. v2.1.0

   2026-11-01

   WI-EV 1.7.0

   WI-SL 1.2.0

   WI-SI 1.0.0

   WI-FD 0.1

   Full-text stance checks (WI-EV 1.7.0)

   Studies whose abstract gave no direction are now read in full where a lawfully reusable open copy exists. Each full-text verdict carries a verbatim quote of at most 40 words and its section, table or figure, and waits in the reviewer queue: it counts in a score only after a staff reviewer approves it. All 14 claims re-run; no score changed and none was set by hand.

   • MethodEvery relevant study marked 'Stance not stated in abstract' or 'Relevant – no direction' (414) was checked for open full text via Unpaywall, with CORE as a second repository source. Only copies under CC-BY, CC-BY-SA, CC0 or public domain are read, and robots.txt and TDM opt-outs are respected.
   • MethodFull-text verdicts are labelled 'Judged from full text' and must quote the paper verbatim (max 40 words) from results or conclusions, with section, table or figure. Verdicts citing the abstract are refused.
   • ChangedFull-text verdicts no longer apply in bulk. Each is approved or rejected individually by a staff reviewer and logged in the public approval log; silicon/silicate claims still require a reviewer independent of the funder.
   • AddedClaim pages show how many pooled studies were judged from full text, from the abstract only, or set by an expert reviewer.
   • DataCoverage of the 414 checks: 15 verdicts staged for review; 166 no open copy; 54 open copy without a qualifying licence; 24 publisher TDM opt-out or robots.txt block; 144 copy could not be retrieved or read; 11 verdicts refused (quote not found, over 40 words or abstract location).
   • DataPer claim, staged for review (abstract → full text): accelerated tests 1 (no direction → no direction); coating maintenance 3 (→ not relevant); detailing vs durability class 2 (1 → not relevant, 1 no direction); mass-timber hygrothermal risk 1 (no direction); moisture threshold 1 (no direction); thermal modification 1 (no direction); biogenic carbon storage 1 (→ contradicts); untreated larch 4 (→ not relevant); carbon substitution 1 (→ supports). No verdicts for acetylation, CLT charring, treated pine UC3, Scheffer index or silicate weathering.
   • DataScores and bands unchanged pending review: thermal modification 52 Low, biogenic carbon storage 58 Moderate, carbon substitution 55 Low, Scheffer index 45 Low, accelerated tests 45 Low; the other nine remain Insufficient evidence (silicate: Not assessed – conflict of interest).

2. v2.0.0

   2026-10-14

   Value-changing

   WI-EV 1.6.0

   WI-SL 1.2.0

   WI-SI 1.0.0

   WI-FD 0.1

   WoodIntel 2.0 – the rebuilt Evidence Engine

   The rebuilt Evidence Engine as one release: relevance gating, expanded recall, evidence tiers by test type, method-claim tiers with a circularity rule, direction-first verdicts, weekly integrity checks, a public approval log and the conflict-of-interest rule. These changes together define how every answer on the platform is scored, so the platform version moves to 2.0.0.

   • FixedRelevance fix (WI-EV 1.2.0): retrieval now requires a wood-domain term in title or abstract plus OpenAlex wood-science, forestry and building-materials topic filters. 298 of 450 indexed studies were removed from claim pools as not relevant and kept in a public audit log with reasons. Claims with fewer than 3 relevant directional studies show Insufficient evidence instead of a number.
   • MethodRecall expansion (WI-EV 1.3.0): each claim adds an expanded query set — scientific and trade synonyms, species names, test standards (EN 113, EN 252, EN 330, EN 927, EN 335) and outcome terms — plus citation chaining through OpenAlex references and citing works, and IRG-WP documents as conference papers. Relevant studies rose from 150 to 637 across the 14 claims. Every claim page shows the queries used, records screened and records relevant.
   • MethodEvidence tiers (WI-EV 1.4.0): every relevant study is classified by test type — (a) field test in the claim's exposure, (b) field test elsewhere, (c) laboratory decay or weathering test, (d) mechanism or chemistry study, (e) review — and weighted in that order (1, 0.75, 0.55, 0.35). Reviews never count as directional evidence; near-duplicate records are merged even when DOIs differ; contradictions are explained by treatment, species, method, exposure and duration; field-exposure claims without tier (a) evidence are capped at Low confidence.
   • MethodMethod-claim tiers (WI-EV 1.5.0): claims about a method, index or model are judged by whether studies validate the method against observed outcomes (tier a, weight 1) or compare it with other models (tier b, weight 0.5); applications without validation never take a direction, and the publication that originally proposes a method is shown in a separate 'Original source' section and never counts for its own reliability (Scheffer 1971).
   • MethodDirection-first verdicts (WI-EV 1.6.0): every scored claim shows a verdict — Supported, Contradicted, Mixed or Unclear — computed from unrounded weighted directional evidence, with thresholds documented on /method. Plain-language readings and social previews state direction first, and Evidence strength is labelled 'How much and how good the evidence is, not whether the claim is true.'
   • MethodIntegrity checks (WI-SI 1.0.0): every cited DOI is checked weekly against Crossref (Retraction Watch data) for retraction and correction notices, with durable per-DOI status. Retracted works are excluded from evidence grading and service-life calculations with dated removal notes, and the public Integrity checks panel reports DOIs checked, flagged and the last completed full check from completed-run snapshots.
   • AddedPublic approval log: every human approval — relevance decisions, field-result reviews and stance overrides — is recorded in a publicly readable log with the approver's name, role, affiliation and independence flag. Visitors can read it; it contains no account identifiers.
   • MethodConflict-of-interest rule: silicon- or silicate-based treatment claims can only be verified, classified or approved by a reviewer account flagged 'independent of the funder'; admin and editor accounts are always refused, and the database enforces the same rule directly. The silicate surface-treatment weathering claim is shown as 'Not assessed – conflict of interest' with the funding note, no score and no verdict; its indexed studies, search panel and excluded-studies log remain listed for anyone to read.

3. v1.11.0

   2026-10-14

   WI-EV 1.6.0

   WI-SL 1.2.0

   WI-FD 0.1

   Evidence direction alongside strength (WI-EV 1.6.0)

   Direction is now as prominent as evidence strength: Supported, Contradicted, Mixed or Unclear. Conclusions and link previews lead with direction. No score arithmetic, study classifications, weights or confidence thresholds changed; no score was set by hand.

   • MethodVerdict uses unrounded weighted directional evidence: at least two-thirds support = Supported; at least two-thirds contradict = Contradicted; otherwise Mixed. Fewer than three directional primary studies or no positive weight = Unclear. Existing exclusions, deduplication and tier/grade/peer-review weights are preserved. Rules are documented on /method.
   • ChangedEvidence strength is labelled Evidence strength, with the explanation: How much and how good the evidence is, not whether the claim is true.
   • AddedVerdict badges next to headline strength on claim pages and on every evidence card, plus a verdict filter on the evidence register.
   • ChangedPlain-language readings and working conclusions state direction first, including contradicting counts when evidence leans against a claim. Social-preview descriptions lead with verdict, not score.
   • DataAll 14 claims recomputed from their existing study pools. Four Supported: thermal modification 52, biogenic carbon storage 58, carbon substitution 55 and Scheffer index 45. Accelerated tests: Contradicted, 45 (3 of 4 field validations contradict). The other nine claims: Unclear, Insufficient evidence. Every score and confidence band unchanged.
   • FixedThe overview now pages through the complete public study dataset, avoiding the database's 1,000-record response cap; claim cards and detail pages use the same complete pools.

4. v1.10.0

   2026-10-14

   Value-changing

   WI-EV 1.5.0

   WI-SL 1.2.0

   WI-FD 0.1

   Method-claim tiers and circularity rule (WI-EV 1.5.0)

   Claims about a method, index or model are now judged by whether studies test it against observed outcomes. The publication that proposes a method never counts for its own reliability. Above-ground field studies are tier (a) for above-ground claims. All 14 claims re-run; no score was set by hand.

   • MethodMethod claims (Scheffer index, accelerated weathering tests, CLT charring in Eurocode 5) use their own tiers: (a) validation against observed decay or field performance (weight 1), (b) comparison with another model or method without observed outcomes (weight 0.5), (c) application of the method without validation — always 'Relevant – no direction', (e) review. A method claim with no tier (a) validation is capped at Low confidence.
   • MethodCircularity rule: the publication that originally proposes a method can never count as directional evidence for its reliability. It is shown in a separate 'Original source' section (Scheffer 1971 for the Scheffer index).
   • MethodAbove-ground field studies (EN 330, L-joint, lap-joint, double-layer, in-service surveys) are always tier (a) for claims about above-ground exposure, enforced by a rule after classification. Check: no above-ground field claim had field studies without any tier (a); three individual studies were moved to tier (a) — one each for thermal modification, moisture threshold and detailing.
   • DataScheffer index: 45 Low → 45 Low. Scheffer 1971 moved to Original source; 3 directional studies (2 validations against observed decay, 1 model comparison); 33 applications now count as no direction.
   • DataAccelerated tests: 46 Low → 45 Low. 4 directional validations against natural weathering (1 supports, 3 contradict); 44 applications now count as no direction.
   • DataCLT charring: Insufficient evidence → Insufficient evidence. 3 validation studies found, none takes a side; 21 applications.
   • DataThermal modification: 52 Low → 52 Moderate. Re-classification moved four directional studies from mechanism to laboratory tests (now 3 field in the claim's exposure, 11 laboratory, 1 mechanism), which raised the weighted directional evidence above the Moderate threshold; one non-directional above-ground study was moved to tier (a).
   • DataMoisture threshold and detailing vs durability class: one study each moved to tier (a); both remain Insufficient evidence.
   • DataUnchanged: biogenic carbon storage 58 Moderate; carbon substitution 55 Low; silicate weathering, untreated larch, treated pine use class 3, acetylation, coating maintenance and mass-timber hygrothermal risk remain Insufficient evidence.

5. v1.9.0

   2026-10-14

   Value-changing

   WI-EV 1.4.0

   WI-SL 1.2.0

   WI-FD 0.1

   Evidence tiers, reviews and deduplication (WI-EV 1.4.0)

   Studies are now weighted by the kind of test behind them, reviews no longer count as directional evidence, and duplicate records are merged. All 14 claims were re-run; no score was set by hand.

   • MethodEvery relevant study is classified by test type: (a) field test in the claim's exposure, (b) field test in another exposure, (c) laboratory decay or weathering test, (d) mechanism or chemistry study, (e) review. Tiers (a)–(d) are weighted in that order (1, 0.75, 0.55, 0.35) on top of the evidence grade. Each claim page shows how many directional studies sit in each tier.
   • MethodIf a claim specifies a field exposure and none of its directional studies is a tier (a) field test, confidence is capped at Low and the claim page says so.
   • MethodReviews and syntheses are never counted in consensus or volume; they are listed in a separate 'Reviews and syntheses' section.
   • MethodRecords with the same or near-identical title and authors are merged even when their DOIs differ; the kept record lists the other identifiers.
   • MethodContradictions are explained by treatment, species, test method, exposure and duration extracted from the abstract or full text — never by publication year or venue alone. Where those are not stated, the conflict is marked unexplained.
   • ChangedClaims with a score and at least 5 directional studies show 'Working conclusion (automated, not yet expert-reviewed)' with a one-sentence reading; 'Under review' is kept only for Insufficient evidence. Consensus is shown as 'N of M directional studies' next to the percentage.
   • DataEvidence strength before → after: thermal modification 63 Moderate → 52 Low; biogenic carbon storage 73 Moderate → 58 Moderate; carbon substitution 70 Moderate → 55 Low; accelerated tests 44 → 46 Low; Scheffer index 48 → 45 Low; moisture threshold 46 Low → Insufficient evidence (2 directional primary studies once reviews and duplicates were removed). The other eight claims remain Insufficient evidence.
   • ChangedWording now states the real cadence: each claim is re-searched every 2–3 days.

6. v1.8.0

   2026-10-14

   Value-changing

   WI-EV 1.3.0

   WI-SL 1.2.0

   WI-FD 0.1

   Evidence Engine recall expansion (WI-EV 1.3.0)

   Claims are now searched with expanded query sets, IRG-WP documents and citation chaining, under the same relevance rules. Relevant studies rose from 150 to 637 across the 14 claims; six claims now carry a score, eight still show Insufficient evidence. No score was set by hand.

   • MethodEach claim keeps its primary query across all registries and adds an expanded query set: scientific and generic trade synonyms (for silicates: sodium silicate, water glass, TEOS, silica sol, siloxane), species names, test standards (EN 113, EN 252, EN 330, EN 927, EN 335) and outcome terms (decay, mass loss, weathering, colour change, service life, dimensional stability). The query sets are listed on each claim page.
   • MethodCitation chaining: for every study judged relevant, its references and citing works are fetched from OpenAlex and screened by the same wood-domain rule and relevance classifier.
   • AddedIRG-WP (International Research Group on Wood Protection) documents as a source, using the open metadata in the University of Melbourne repository. They are graded as conference papers, not journal peer review. The open registry list grows from 16 to 17.
   • AddedEvery claim page shows how it was searched: queries used, records screened, records relevant and relevant records added via citation chaining.
   • DataEvidence strength before → after (relevant studies): thermal modification 55 → 63, Low → Moderate (12 → 42); biogenic carbon storage 60 → 73, Low → Moderate (26 → 70); carbon substitution 57 → 70, Low → Moderate (24 → 82); accelerated tests Insufficient → 44 Low (13 → 65); moisture threshold Insufficient → 46 Low (6 → 61); Scheffer index Insufficient → 48 Low (16 → 50).
   • DataStill Insufficient evidence (relevant studies before → after, directional): silicate weathering 0 → 7 (1); untreated larch 5 → 23 (0); treated pine use class 3 1 → 13 (0); acetylation 7 → 26 (0); CLT charring 7 → 28 (0); coating maintenance 2 → 64 (0); detailing vs durability class 9 → 72 (0); mass-timber hygrothermal risk 22 → 54 (2). More relevant studies do not by themselves produce a score: at least 3 must state a direction.

7. v1.7.0

   2026-10-12

   Value-changing

   WI-EV 1.2.0

   WI-SL 1.2.0

   WI-FD 0.1

   Evidence Engine relevance fix (WI-EV 1.2.0)

   Studies that do not bear on a claim no longer count. 298 of 450 indexed studies were removed from claim pools and scores; 11 of 14 claims now show Insufficient evidence instead of a number.

   • FixedIrrelevant studies (different material, property or field) were previously counted as Neutral in claim pools, study counts and every score component. They are now excluded and kept only in a public audit log with the reason.
   • MethodRetrieval requires a wood-domain term (wood, timber, lumber, softwood, hardwood, cladding or a species name) in title or abstract, and adds an OpenAlex search filtered to wood-science, forestry and building-materials topics. Geology, geochemistry, medicine, ecology and soil-science records are excluded unless they also match wood terms.
   • MethodThe Neutral label is split into Not relevant and Relevant – no direction. All 450 indexed studies were re-classified for all 14 claims.
   • MethodA claim with fewer than 3 relevant directional studies shows Insufficient evidence and its relevant-study count instead of a numeric Evidence strength.
   • AddedReviewers can mark any study not relevant, with a required reason; it leaves scoring at once and the decision is logged.
   • DataStudies removed per claim: accelerated tests 30 of 43; acetylation 33 of 41; CLT charring 30 of 37; coating maintenance 22 of 24; detailing vs durability class 18 of 27; mass-timber hygrothermal risk 3 of 25; moisture threshold 16 of 23; treated pine use class 3 16 of 17; Scheffer index 14 of 30; silicate weathering 29 of 29; thermal modification 28 of 40; biogenic carbon storage 13 of 39; untreated larch 35 of 40; carbon substitution 11 of 35.

8. v1.6.0

   2026-08-04

   WI-METH 1.0

   WI-SL 1.2.0

   WI-EV 1.0.0

   WI-API 1.0.0

   Methodology 1.0 and the public release history

   The full methodology is published as a single citable document, and every platform version is now documented in public.

   • AddedWoodIntel Methodology 1.0 — the complete reference document covering models, data sources, the evidence model, confidence and consensus scoring, quality grading and the API. Web version and downloadable PDF.
   • AddedPublic release history with per-release component versions and change classification.
   • AddedCitation block on the methodology document in APA, MLA, Chicago and BibTeX.

9. v1.5.0

   2026-08-03

   WI-SL 1.2.0

   WI-EV 1.0.0

   WI-FC 1.0.0

   WI-TWIN 1.0.0

   Knowledge infrastructure completed

   Knowledge graph, structured AI debate, field observatory, climate forecasting, digital twin, carbon intelligence and specification review.

   • AddedKnowledge graph of species, standards, mechanisms and claims with typed relations.
   • AddedStructured AI debate: both sides argued from graded evidence, then adjudicated against the corpus.
   • AddedReal-world observatory — moderated public register of in-service field observations, with open CSV and JSON export.
   • AddedClimate forecast model (WI-FC) applying IPCC AR6 SSP deltas to modelled decay load.
   • AddedDigital twin (WI-TWIN) simulating per-facade hazard accumulation for a building brief.
   • AddedCarbon intelligence following the EN 15804+A2 module structure.
   • AddedAI specification review combining deterministic rule checks with a retrieval-grounded model pass.
   • MethodEN 15804:2012+A2:2019 added to the source registry.

10. v1.4.0

    2026-08-02

    WI-SL 1.2.0

    WI-EV 1.0.0

    Plain-language layer

    Every decision surface now leads with the answer in plain language and folds the technical machinery underneath it.

    • ChangedAnswer-first pattern applied across evidence claims, the match engine, scores, gaps, compare, observatory, map and the home page.
    • AddedCollapsible technical detail component so Monte Carlo tables, source registries and formulas stay available without dominating the page.
    • ChangedUncertainty is now stated in words before it is stated in percentiles.

11. v1.3.0

    2026-08-02

    WI-EV 1.0.0

    WI-GAP 1.0.0

    WI-SL 1.2.0

    Evidence Engine

    The living evidence synthesis: evidence scoring, consensus, contradictions, timelines, the research agenda and a public audit trail.

    • AddedEvidence Engine (WI-EV) — Evidence strength, confidence band, weighted consensus and detected contradictions per tracked claim.
    • AddedNightly automated literature sweep with per-study stance classification and published reasons.
    • AddedResearch Agenda (WI-GAP) scoring open gaps 0–100 with a per-factor breakdown.
    • AddedPublic evidence changelog and RSS feed: every revision, with the reason the score moved.
    • AddedClaim-level citation export and Schema.org structured data.

12. v1.2.0

    2026-08-01

    WI-SL 1.2.0

    WI-INDEX 1.0.0

    WI-API 1.0.0

    The Observatory: scores, map and open API

    Species scoring, the live climate map and a free, unauthenticated CC BY 4.0 read API.

    • AddedWoodIntel Score — six normalised sub-scores and an unweighted composite, each with its formula published.
    • AddedLive world wood map built on ERA5 reanalysis climate data.
    • AddedOpen read API v1 with a licence and attribution envelope on every response.
    • AddedPrecomputed prediction index (WI-INDEX) across species, zones, orientations and treatment states.
    • AddedFact-check portal, research briefs, open data downloads and the press portal.

13. v1.1.0

    2026-08-01

    Value-changing

    WI-SL 1.2.0

    WI-CAL 1.1.0

    WI-VAL 1.1.0

    Validation and calibration

    Predictions are measured against published field evidence, and the resulting bias is corrected per climate zone.

    • AddedValidation benchmark set (WI-VAL) of field trials and in-service inspections, with mean error, bias and interval coverage reported in public.
    • MethodPer-zone empirical calibration layer (WI-CAL 1.1.0) introduced. Previously published service lives changed value in calibrated zones.
    • MethodService-life method raised to WI-SL 1.2.0 to incorporate the calibration term.
    • AddedDownloadable specification report and machine-readable calculation record.

14. v1.0.0

    2026-07-31

    WI-SL 1.1.0

    First public release

    The reproducible service-life method, Monte Carlo uncertainty and the retrieval-grounded standards corpus.

    • AddedService-life screening method (WI-SL) published in full: every parameter, equation and default.
    • AddedMonte Carlo uncertainty propagation with seeded reproducibility, reporting P10/P50/P90 and per-input variance shares.
    • AddedRetrieval-grounded question answering over an indexed corpus of European standards, with inline citations.
    • AddedProvenance layer: every published value carries its source, derivation and confidence class.
    • AddedSeven-language interface with right-to-left support.