Milestones v0.8.0 + v0.9.0 — Ensemble Forecasting + WASM Runtime Verification (Phases 4–8) by sipemu · Pull Request #256 · DataZooDE/anofox-forecast

sipemu · 2026-08-31T19:11:33Z

Milestones v0.8.0 + v0.9.0 — Ensemble Forecasting + WASM Runtime Verification (Phases 4–8)

main currently has neither milestone, so this PR ships both to main in one merge (86 commits): the v0.8.0 ensemble surface (Phases 4–6) and the v0.9.0 WASM runtime verification + CI gating (Phases 7–8). The v0.9.0 summary is below; the original v0.8.0 summary follows unchanged.

Milestone v0.9.0 — WASM Runtime Verification

Proves the built anofox_forecast .wasm actually loads and runs in DuckDB-Wasm — not just that it compiles and links — and gates that in CI so WASM regressions fail the build. No new SQL functions or crate changes; work lives in test/wasm/, .github/workflows/, vcpkg.json, and README.md. Ports the CI/harness pattern from anofox-statistics PR #131.

Phase 7 — WASM Node harness + local green ✅ verified

test/wasm/run.mjs + sqllogic.mjs — boots DuckDB-Wasm (eh bundle, pthreadWorker=null), serves the built .wasm over localhost, FORCE INSTALL + LOADs it, and runs a sqllogictest subset with per-file db.open() catalog isolation and a COLUMNS(*)::VARCHAR wrap so DECIMAL rendering matches native (WASM-01, WASM-02).
DEP-01 — pins @duckdb/duckdb-wasm@1.33.1-dev64.0 (engine v1.5.5) + web-worker@1.2.0, with a documented strings | grep version-verification procedure in test/wasm/README.md.
DEP-02 — vcpkg.json declares openssl as { "platform": "!wasm32" }. Build-confirmed: a fresh local wasm_eh build compiled zero OpenSSL for the wasm target (mbedtls is used instead); native builds still statically link it.
Curated green subset: 396/396 assertions across 8 files pass against the built .wasm.

Phase 8 — CI gating + dedicated workflow + badge ◆ implemented

CI-01 — wasm-runtime-test gating job in MainDistributionPipeline.yml: needs: the wasm build, downloads the anofox_forecast-v1.5.5-extension-wasm_eh artifact, runs the curated harness, fails the build on any WASM load/runtime error.
CI-02 — dedicated .github/workflows/WasmTest.yml (workflow_run-triggered) so WASM status is independently observable.
CI-03 — README WASM status badge wired to WasmTest.yml.
Gate mechanism proven locally: curated run exit 0 → a deliberately broken curated assertion → exit 1 (red) → reverted → exit 0.

Honest scope notes

WASM-03 (full 66-file suite green) is deferred. The full --all run is red (184 failures / 23 files), but these were confirmed against a fresh HEAD build to be pre-existing test/sql test-suite debt — stale references to removed/renamed API (ts_backtest_auto_by, ts_hydrate_features_by, …), DATE + BIGINT bugs, and cascade failures — not WASM or artifact-drift issues. They're masked natively because the unittest runner skips these files on an unsatisfied require json; DuckDB-Wasm auto-loads json and exposes them. CI gates on the curated subset, not --all. Making the full suite green is tracked as separate test-triage.
CI-01 live negative-control (observing green → red → green in GitHub Actions + the badge flip) is a post-merge operator step; the gate's core mechanism is already proven locally.
CI hygiene: excluded windows_amd64 on the v1.4.5 LTS line only — the v1.4-andium ci-tools vcpkg baseline pins an OpenSSL dependency whose upstream download now 404s (not fixable here; same class anofox-statistics documented). The v1.5.5 windows_amd64 build uses a newer baseline, passes, and still ships.

Tag: v0.9.0. Includes the full .planning/ audit trail (repo convention).

Milestone v0.8.0 — Ensemble Forecasting

Exposes the anofox-forecast crate's ensemble surface to SQL: combine multiple models per series (automatically or by explicit member list), with six combination methods, distribution-free conformal prediction intervals, and member/weight introspection — all via the established FFI → C++ → macro → example → docs pattern.

8/8 requirements delivered and verified against the built extension · 3 phases · 6 plans · ~13.9k LOC / 57 files.

New SQL surface

Function	Purpose	Requirement
`ts_forecast_by(..., 'AutoEnsemble', ..., {top_k, combination_method, seasonal_period})`	Auto-fit ARIMA/ETS/Theta, combine top-K	ENS-01
six `combination_method`s: `mean`,`median`,`weighted_mse`,`inverse_aic`,`stacking`,`horizon_adaptive`	Blend members	COMB-01..04
`ts_forecast_ensemble_by('table', grp, ds, y, members VARCHAR[], ...)`	Explicit user-named member ensemble (+ `build_forecaster` factory: 26-member allowlist, 10 blocked with clear errors)	ENS-02
ensemble conformal intervals via `ts_cv_folds_by` + `ts_conformal_calibrate`/apply	Distribution-free lower/upper bounds (reuses existing path — no new interval code)	EPI-01
`ts_ensemble_inspect_by` / `ts_auto_ensemble_inspect_by`	Selected members + combination weights + MSE score/rank, long format	INSP-01

Verification

Every example run against the built extension (PR docs: fix README ToC + getting-started SQL examples (#224, #225, #226) #230 rule): Mean cross-check diff=0.0, all six methods produce finite forecasts, all error paths raise clear member-naming errors, conformal lower ≤ point ≤ upper per step, all INSP-01 DoD assertions 0 failures.
233 Rust unit tests green. Code review across the milestone fixed 5 warnings (negative top_k clamp, exog param path, NULL-interpolation/min-length guards, SMA window) + info items.
Cross-phase integration checked: shared parse_combination_method + build_forecaster back all three surfaces consistently; no regression on pre-milestone models.

Known tech debt (documented in `milestones/v0.8.0-MILESTONE-AUDIT.md`)

ts_cv_forecast_by('AutoEnsemble') segfaults — pre-existing crate/CV-native bug (ts_cv_forecast_native.cpp:380-388 never parses the ensemble params). EPI-01 ships via a documented manual per-fold _ts_forecast_scalar workaround. Recommended follow-up: wire ensemble-param parsing into the CV native.
AutoEnsemble non-Mean combination weights return NULL — crate 0.15.3 exposes no inner-ensemble weight accessor (documented; upstream enhancement).
build_forecaster SeasonalWindowAverage n_seasons=2 hardcoded (TODO ENS-03).

Tag: v0.8.0. Includes the full .planning/ audit trail (repo convention).

🤖 Generated with Claude Code

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

- Add AutoEnsemble enum variant to ModelType (with doc comment) - Add from_str exact-match ("AutoEnsemble") and lowercase fallback ("autoensemble" | "auto_ensemble") arms - Add name() arm returning "AutoEnsemble" - Add ensemble_top_k: usize and ensemble_method: Option<String> to ForecastOptions and ForecastOptionsExog (additive, after kalman_model) - Thread new fields through ForecastOptionsExog::Default and From<ForecastOptions> for ForecastOptionsExog - Add AutoEnsemble dispatch arm in forecast() and forecast_with_model() - Add AutoEnsemble to both CI-skip guards and calculate_fitted_values guard (point-forecast-only in Phase 4; intervals deferred to EPI-01) - Add parse_combination_method() helper (all 6 methods; custom rejected) - Add forecast_auto_ensemble() helper mirroring forecast_kalman() pattern - Kalman tests still pass (no regression)

… regen header - Append ensemble_top_k: c_int and ensemble_method: [c_char; 32] at END of ForecastOptions (FFI) and ForecastOptionsExog (FFI) in types.rs (ABI-safe; never middle-insert — callers zero-init via memset so existing GARCH/Kalman behavior is unchanged) - Extend Default impls for both structs with ensemble_top_k: 0, ensemble_method: [0; 32] - Thread ensemble_top_k and ensemble_method through all three core-option construction sites in lib.rs: - anofox_ts_forecast (~line 3452): parse ensemble_method from CStr - anofox_ts_forecast_with_exog (~line 3743): same pattern for exog path - build_core_options (~line 4152): same pattern for aggregation path - Regenerate anofox_fcst_ffi.h via `make header` (cbindgen); both structs now expose ensemble_top_k and ensemble_method to C++ callers

…eck example ts_forecast_scalar.cpp — 6 additive sites mirroring kalman_model pattern: - TsForecastScalarBindData: add ensemble_top_k (int64_t) and ensemble_method (string) - Copy(): forward both new fields - ValidateParams valid_keys: add "top_k", "combination_method" - Error message: append top_k, combination_method to valid-params list - Local var decls in TsForecastScalarExecute: read from bind_data - MAP/STRUCT params parse: ParseInt64Param("top_k"), ParseStringParam("combination_method") - opts building: ensemble_top_k cast + guarded strncpy for ensemble_method ts_forecast_native.cpp — 4 additive sites (same field names): - TsForecastNativeBindData: add ensemble_top_k and ensemble_method - ValidateParamKeys valid_keys + error message: same additions - Bind params parse: ParseInt64FromParams, ParseStringFromParams - opts building in Finalize: same guarded strncpy pattern examples/forecasting/autoensemble.sql — new runnable example: - Synthetic 60-obs linear series (all three Auto* members converge reliably) - Step 1: AutoEnsemble(mean, top_k=3) → ae_mean - Steps 2–3: independent AutoARIMA/AutoETS/AutoTheta → manual arithmetic mean - Step 4: cross-check — every row shows abs(ensemble_yhat - manual_mean) < 1e-6 - Assertion query fails loudly if any row mismatches (DoD, PR#230 rule) - Section 3: six-method smoke test (all six combination_method values return finite non-NULL yhat) Build: extension rebuilt via ninja; example ran against built extension — cross-check all 5 steps pass (diff=0.0, match=true); yhat_lower/upper=NULL.

…y to autoensemble.sql - Section 3 upgraded from display-only to UNION-ALL assertion: all six combination_method strings (mean, median, weighted_mse, inverse_aic, stacking, horizon_adaptive) tagged with ok=isfinite(yhat) AND IS NOT NULL; failure query returns 0 rows on success (COMB-01..04) - Section 4 added: skewed series (exp growth + spikes at obs 10/30/50) runs Mean and Median combinations separately; comparison shows delta 1.45-2.69 per step; assertion confirms at least one step has delta > 1e-6 (COMB-01 demonstrability) - Verified against built extension: all assertions return 0-row failure sets

Add docs/reference/models/ensemble/autoensemble.md (method-string dispatch, top_k/combination_method/seasonal_period params, six combination methods, Mean cross-check, NULL-interval note, fewer-than-top_k behavior) and an AutoEnsemble section in docs/api/07-forecasting.md. Every SQL snippet verified against the built extension (PR #230). Reconciled by orchestrator: the executor reported a commit (a37f8c9) and SUMMARY that were never actually written; the on-disk work was verified correct (example + doc snippets run clean against the built extension) and is committed here. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

… mis-report) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

Mirror the .max(0) pattern already used for the window field at all three ensemble_top_k cast sites. Negative c_int values like -1 previously wrapped to usize::MAX; they now clamp to 0, which the core's top_k==0 guard converts to the default of 3. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

forecast_with_exog's else-branch routed AutoEnsemble through forecast_with_model, which hard-codes top_k=3 and method=None. Dispatch AutoEnsemble separately in that branch, reading the caller's ensemble_top_k and ensemble_method from ForecastOptionsExog. All other models continue to go through forecast_with_model unchanged. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

…utoensemble.md (IN-01) Add a note to the Rust doc comment on parse_combination_method that 'stacking' maps to CombinationMethod::Stacking { folds: 2 } with a fixed 2-fold second-half in-sample holdout not user-configurable in v1. Mirror the same note in the combination_method SQL parameter table. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

…e (IN-02) Adds five inline tests in forecast::tests: - parse_combination_method_rejects_unknown: 'custom' and unknown strings -> Err - parse_combination_method_accepts_canonical_strings: all six methods -> Ok - parse_combination_method_accepts_aliases: representative aliases -> Ok - parse_combination_method_none_and_empty_give_mean: None/''/'' -> Ok - forecast_auto_ensemble_basic: 60-pt series, top_k=0, horizon=5 -> Ok Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

…tate tracking

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

…semble to core - build_forecaster: compiler-exhaustive match over all 36 ModelType variants; 10 blocked variants (GARCH, Laplace, ARIMA, MFLES, AutoMFLES, MSTL, AutoMSTL, TBATS, AutoTBATS, AutoEnsemble) return Err(InvalidParameter) naming the member - forecast_explicit_ensemble: builds member Vec, calls Ensemble::new(members) .with_method(combination_method); reuses parse_combination_method from Phase 4; returns model_name = 'Ensemble' (point-only, intervals Phase 6 EPI-01) - Ensemble implements Forecaster (model.rs:575); extract_forecast works directly

…header - anofox_ts_forecast_ensemble: null-delimited member buffer + explicit members_buf_len (no over-read); catch_unwind panic safety; InvalidParameter → ErrorCode::InvalidInput path (same as forecast); NULL lower/upper bounds (point-only, EPI-01 deferred) - Export forecast_explicit_ensemble as pub from anofox-fcst-core lib.rs - make header: anofox_ts_forecast_ensemble + members_buf_len in generated anofox_fcst_ffi.h (verified: 1 fn + 2 param occurrences)

…fied tracer - _ts_forecast_ensemble_native: ScalarFunction (per _ts_forecast_scalar precedent); takes LIST(VARCHAR) members, builds null-delimited buffer, calls anofox_ts_forecast_ensemble; yhat_lower/upper emit NULL (EPI-01 deferred) - CMakeLists.txt: src/table_functions/ts_forecast_ensemble_native.cpp added to EXTENSION_SOURCES after ts_forecast_var_native.cpp (Phase 5: ENS-02) - extension.cpp: #include ts_forecast_ensemble_native.hpp + Register* call - ts_macros.cpp: ts_forecast_ensemble_by macro (per-series, GROUP BY, unnest shape) - ensemble_explicit_tracer.sql: mismatch_count=0, non_null_intervals=0, model_name='Ensemble'; diff=0.0 on all 5 steps (exact match, no 1e-6 tolerance needed)

Archive ROADMAP/REQUIREMENTS/AUDIT + phase dirs to milestones/v0.8.0-*; reorganize ROADMAP.md (v0.8.0 shipped, collapsed); evolve PROJECT.md (5 requirements → Validated, v0.8.0 decisions, current state); append v0.8.0 retrospective. MILESTONES.md + STATE.md updated by milestone.complete. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

…ate) The ensemble build_forecaster/forecast_explicit_ensemble (Phase 5) and the introspection FFI exports (Phase 6) were not rustfmt-clean; CI's Rust Tests job gates on cargo fmt --check. Formatting-only — cargo test still 240+12 green, no logic change. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

…g safety doc) The Rust Tests CI job runs fmt -> clippy -D warnings -> test. Two clippy lints in the Phase 5 ensemble code: field_reassign_with_default in build_forecaster (AutoARIMAConfig) and missing_safety_doc on the anofox_ts_forecast_ensemble FFI export. Both fixed; fmt + clippy + 240/12 tests green locally. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

- Add test/wasm/run.mjs: main harness (eh bundle, pthreadWorker=null, web-worker@1.2.0, version-agnostic localhost server, per-file db.open() catalog isolation, FORCE INSTALL + LOAD) - Add test/wasm/sqllogic.mjs: minimal sqllogictest-subset parser/runner (statement ok/error, query, mode skip; SELECT COLUMNS(*)::VARCHAR DECIMAL fix; float tolerance; bool normalization) - Add test/wasm/package.json: exact pins @duckdb/duckdb-wasm@1.33.1-dev64.0 (engine v1.5.5) and web-worker@1.2.0 - Add test/wasm/package-lock.json: committed lockfile for reproducible installs - Update .gitignore: add test/wasm/node_modules/ - Tracer: ts_forecast_by.test passes 90/90 assertions in DuckDB-Wasm Note: pre-built local artifact (July 10) was stale and missing the duckdb_signature metadata section; replaced with CI artifact from run #33554081155 (build/wasm_eh). Emscripten not available locally.

Convert bare 'openssl' dependency string to object with platform: '!wasm32' so Emscripten does not compile/link OpenSSL for the WASM target. Native builds continue to statically link OpenSSL (issues #211/#215 requirements unchanged). WASM rebuild verification: Emscripten not available locally; vcpkg guard verified by node check (node -e 'vcpkg.json parser'); harness re-run against CI artifact (run #33554081155) still green (90 passed, 0 failed). Full rebuild + OpenSSL-drop confirmation deferred to CI.

- SKIP_FILES: 4 structurally WASM-infeasible files (heap overflow on TSFresh, UNNEST engine limit, Emscripten ___trap); each entry carries a one-line reason that is logged at run time - CURATED: updated to the 8 files verified green against the current WASM artifact (ts_forecast_by, ts_metrics, ts_decomposition, ts_diagnostics, ts_conformal, ts_cv_folds, ts_aggregate_hierarchy, ts_periods) — 396/396 assertions pass - --all baseline: 2255 passed, 188 failed (23 files); 39 of 66 files fully green; 23 failing files track artifact-API drift and pre-existing test bugs, documented in SUMMARY.md

- Purpose, requirements, quick-start, running modes table - DEP-01: strings-based procedure to read embedded DuckDB version from the duckdb_signature WASM custom section; version-mapping table; bump checklist - Full-suite status table (66 files, 39 pass, 23 fail, 4 skipped) - Skip-list rationale and known-failure categories for the 23 failing files (artifact-API drift vs. pre-existing test bugs) - Architecture notes: pthreadWorker=null, COLUMNS(*)::VARCHAR wrap, per-file db.open() catalog isolation

…ifact drift); DEP-02 build-confirmed Local HEAD wasm_eh build (emsdk + wasm32-unknown-emscripten + binaryen 123) reproduced the same 23 failing files, disproving the stale-artifact theory. Failures are pre-existing test-suite debt (removed/renamed API refs, DATE+BIGINT, cascades) masked natively by the unittest require-json skip. DEP-02 openssl !wasm32 guard confirmed at build time (zero openssl compiled for wasm). Phase 7 harness accepted; full-suite green deferred as out-of-scope test triage. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

…ase 8 framing Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

- Ports the dedicated WASM workflow from anofox-statistics PR #131 - Triggers via workflow_run on "Main Extension Distribution Pipeline" (completed) - Downloads anofox_forecast-v1.5.5-extension-wasm_eh artifact via cross-run github-token download (run-id: github.event.workflow_run.id) - Runs curated harness: node test/wasm/run.mjs --ext "$EXT" (no --all) - Permissions: actions read + contents read (read-only, T-08-01 mitigated) - Powers the README WASM badge; badge file name WasmTest.yml matches exactly

…line.yml (CI-01) - New job wasm-runtime-test needs duckdb-latest-build; if-guarded on success - Downloads anofox_forecast-v1.5.5-extension-wasm_eh (same-run, no run-id/github-token) - Runs curated harness: node test/wasm/run.mjs --ext "$EXT" (no --all, no --file) - No continue-on-error: harness exit non-zero -> step fails -> pipeline turns red - checkout step omits submodules (harness only reads test/ + downloaded .wasm) - No existing job modified or removed (duckdb-lts-build/deploy, latest-build/deploy, smoke-tests, build-and-test-rust all unchanged)

- Inserts WASM badge into the existing header badge block (p align=center) - Badge href: https://github.com/DataZooDE/anofox-forecast/actions/workflows/WasmTest.yml - Badge img src: .../WasmTest.yml/badge.svg?branch=main (branch=main keeps dev branches from flipping the visible badge; file name WasmTest.yml matches Task 1 exactly) - Existing License, DuckDB, Build, Tests badges unchanged

…en locally (green→red→green); live-CI push remains (human) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

…ferred_human) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

…line) The v1.4-andium ci-tools vcpkg baseline pins an OpenSSL dependency whose upstream download URL now 404s ('Download failed, halting portfile'), failing the LTS windows_amd64 build + deploy. Not fixable here (vcpkg comes from extension-ci-tools); same class of issue anofox-statistics documented and excluded. Scoped to the LTS line only — the v1.5.5 (variegata) windows_amd64 build uses a newer, working baseline and still ships, honoring the cross-platform Windows requirement on current DuckDB. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

sipemu and others added 30 commits August 30, 2026 21:51

docs: start milestone v0.8.0 Ensemble Forecasting

d8a5285

docs: define milestone v0.8.0 requirements

a6e14bf

docs: create milestone v0.8.0 roadmap (3 phases)

668993c

docs(04): smart discuss context

606532d

docs(04): phase research

1d1798e

docs(04): pattern map

92b5603

docs(04): phase plans (2 plans, ENS-01 + COMB-01..04)

d8ca048

docs(04): create AutoEnsemble surface phase plan (2 plans)

f3ee140

docs(04-01): complete AutoEnsemble tracer slice plan summary and state

92d11e7

docs(04-02): add plan summary (orchestrator-reconciled after executor…

57d6ec4

… mis-report) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

docs(04): code review (2 warnings, 2 info)

fb34aa5

docs(04-02): complete AutoEnsemble combination-methods plan; update s…

f48870e

…tate tracking

docs(04): phase verification passed (6/6; tracking gap closed)

5b79e3d

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

docs(04): finalize phase 4 completion tracking (phase.complete)

adc6e4a

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

docs(05): smart discuss context

0d896ad

docs(05): phase research

2d5157a

docs(05): pattern map

4316bc7

docs(05): create phase plan

1801c3a

sipemu and others added 28 commits August 31, 2026 20:41

chore: remove REQUIREMENTS.md for v0.8.0 milestone (fresh for next)

19a9c13

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

docs(state): clear current-focus after v0.8.0 close

b35d80c

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

chore: remove audit from planning root (archived to milestones/v0.8.0-)

212a44b

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KS9bB2RQf1JFLM8oRoVV8a

docs: start milestone v0.9.0 WASM Runtime Verification

da55a20

docs: define milestone v0.9.0 requirements

bab1e3e

docs: create milestone v0.9.0 roadmap (2 phases)

e611332

docs(07): smart discuss context (infrastructure phase)

cee47d9

docs(07): research phase 7 WASM node harness domain

e738f49

docs(07): create phase plan (WASM Node harness + local green)

6c7c6ae

docs(07-01): complete wasm-node-harness plan — SUMMARY, STATE, ROADMAP

d377243

docs(07): mark Phase 7 complete; carry WASM test-debt finding into Ph…

71ce038

…ase 8 framing Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

docs(08): smart discuss context (infrastructure phase)

b599e0c

docs(08): create phase plan (CI WASM gating + workflow + badge)

b60455c

docs(08-01): complete CI gating wasm badge plan

9cc2f05

docs(08): verification — CI-01/02/03 structural + gate mechanism prov…

a9c6265

…en locally (green→red→green); live-CI push remains (human) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

docs(08): defer live-CI negative-control to operator (verification_de…

c295388

…ferred_human) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LkRFReMJRg33mDS4zo55Tt

sipemu changed the title ~~Milestone v0.8.0 — Ensemble Forecasting (AutoEnsemble + explicit-member ensembles, intervals, introspection)~~ Milestones v0.8.0 + v0.9.0 — Ensemble Forecasting + WASM Runtime Verification (Phases 4–8) Sep 2, 2026

sipemu merged commit a9328e1 into main Sep 2, 2026
91 checks passed

Navigation Menu

Uh oh!

Milestones v0.8.0 + v0.9.0 — Ensemble Forecasting + WASM Runtime Verification (Phases 4–8) - #256

Milestones v0.8.0 + v0.9.0 — Ensemble Forecasting + WASM Runtime Verification (Phases 4–8)#256
sipemu merged 86 commits into
mainfrom
milestone/v0.8.0-ensemble-forecasting

sipemu commented Aug 31, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

sipemu commented Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Milestones v0.8.0 + v0.9.0 — Ensemble Forecasting + WASM Runtime Verification (Phases 4–8)

Milestone v0.9.0 — WASM Runtime Verification

Phase 7 — WASM Node harness + local green ✅ verified

Phase 8 — CI gating + dedicated workflow + badge ◆ implemented

Honest scope notes

Milestone v0.8.0 — Ensemble Forecasting

New SQL surface

Verification

Known tech debt (documented in milestones/v0.8.0-MILESTONE-AUDIT.md)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

sipemu commented Aug 31, 2026 •

edited

Loading

Known tech debt (documented in `milestones/v0.8.0-MILESTONE-AUDIT.md`)