## Reader references — 25 October 2026 Added a [proof-scope reference](https://vultax.com/datasets/research-proof-library-2026-10-25/articles/ensoforecast-historical-archive.md) to the reader’s Sources list. No Lean proof. Document-based research; its factual and legal claims are not covered by a Lean proof. The data edition remains 0.3.2; its dataset files, citation files, checksums, figures and evaluation protocol are unchanged. # Public release 0.3.2 — 2026-10-25 Published by Christopher Geneth Kurg. Promoted the existing reader edition with public publication metadata and credited exports. Numeric records, original source credits and evaluation protocols are unchanged. Raw source archives and private materials remain excluded. No completed calibration/skill result or DOI is claimed. # Version and corrections ## 0.3.1-research-review — 22 October 2026 Corrections from an independent internal review of 0.3.0, made before any release. **No source value changed.** Documentation, metadata, coverage.json and wording changed, so the affected checksums changed. - **ENSO-008 (method):** Evaluation protocol 1.1: all baselines obey the nominal-start cutoff c = t - L; warm-up rules and matched populations (233 fully matched months; 316 for field agreement) are specified; the convention sensitivity direction is corrected; LDEO2 periods before 1996 are conditional or unresolved rather than retrospective. - **ENSO-009 (new derived metadata):** Verified that one valid-cell mask (the same 1,077 cells) is shared by every month of all nine core files; recorded in coverage.json and checked by the source validator. - **ENSO-010 (clarification):** Evidence ledger statuses restricted to the declared vocabulary, with separate scope and archive-mapping fields; E4's limit made conditional on the unestablished LDEO2 mapping; long quotations replaced by located paraphrases. - **ENSO-011 (documentation):** Proposed corrections contact set to Sophia Kurg ; private copies of cited sources kept with retrieval dates and hashes. - **ENSO-012 (method):** Evaluation protocol 1.2, from a further internal review before any skill evaluation: timing classes are assigned by target month, not nominal start. Only targets up to 1985-12 are retrospective (168 of 316 at every lead, from the model chronology); targets from 1986 are issue time unverified. The bulletins' LDEO2 issuance dates define labelled context phases (114 / 6 / 28 targets), not issuance or production classes; ledger E4 and E5 limits amended. The baseline cutoff, warm-up and 233-month matched population, the displayed map labels and every source value are unchanged. Still not done: review by the original creators, peer review, a repository deposit, a DOI, a licence decision, and any calibration or skill evaluation. ## 0.3.0-research-review — 22 October 2026 Editorial and research-use pass. **No source value changed**: the 18 decoded slices, coverage and manifest values are identical to 0.2.0. Edition labels changed in the data files, so their checksums changed. - Expanded the article into a standalone feature: anomalies and their undocumented baseline, target month versus nominal lead, the six-start composite (Figure 2), why the 1982 fields are hindcasts, what contemporaneous 1997 bulletins say, what a fair test needs, and why the fields are not event probabilities. - Added `data/evidence-ledger.json`: ten claims about what the fields are and when they were made, each with status, source, quotation and limits. - Added credited SVG/PNG export of the current comparison (both maps, equatorial profile, legend) recording source files and hashes, target, nominal lead, units, evidence class, credits and edition. Source receipts now name both files and their evidence classes; model panels carry an evidence badge. - Added a methods and data descriptor, a teaching exercise with script, a prespecified evaluation protocol, a correction path and a share card. - Strengthened `validate.py`: checksum manifest must list exactly the files present; edition labels consistent; no DOI, deposit or baseline claimed; ledger complete with the 1997 issue-time claim still unverified; public data may name only the nine core files; the built page must embed exactly the current slices and receipts. It now also runs under `python3 -I`. Corrections in this version: - **ENSO-005 (new evidence):** evidence ledger with contemporaneous 1996–1997 LDEO bulletin articles documenting the six-start composite, the lead convention in those products and the LDEO2 name; archive correspondence unconfirmed. - **ENSO-006 (interpretation):** the 1997 discussion quotes the September 1997 bulletin as consistent with, not proof of, the archived fields. - **ENSO-007 (clarification):** dated sunset-notice language; credit separates archive teams, model authors, the transfer and Vultax's work. Still not done: creator review, peer review, comparison with bulletin figures, identification of the 40 extra files, a deposit, a DOI, a licence statement, any skill evaluation. ## 0.2.0-research-review — 22 October 2026 Improves scholarly reuse of the existing split. Original auction probabilities, announced values and prepared map values are unchanged. The original combined Opus work and the first split remain preserved separately. Added source receipts and links for selected records/comparisons, research notes, a machine-readable dictionary, reproducible validation and separate source/edition citations. Added complete time-coordinate and mask coverage for the nine core files. This is a private review edition. No public deposit, new DOI, original-author review or peer review is recorded. Functional validation is a separate claim from research conclusions. - **ENSO-001 (interpretation):** Explain retrospective evaluation without alleging hindsight contamination. Training/tuning/post-processing separation has not been established by this inspection. - **ENSO-002 (clarification):** Limit warming comparisons to the supplied equatorial fields. State that observation baseline/revisions remain unverified. - **ENSO-003 (new derived metadata):** Check all nine source time axes and valid-cell masks; distinguish coverage from skill and issue-time evidence. - **Reader flow:** Move dated rescue logistics to preservation history and replace temporary urgency with a dated source link. ## 0.1.0-split-review — 22 October 2026 Separated the completed combined article into two independent projects, preserving its source-specific interaction and adding basic provenance metadata. ## Future corrections Record the affected edition, records/files and fields; what changed and why; the supporting evidence; and whether values, interpretation or citation changed. Never silently overwrite a published edition. A future deposited version should link to its predecessor and retain a readable correction history. A correction does not imply endorsement by the original researchers.