Conversation
…pinning ADR Captures the design decisions reached while grilling issue #17 / PRD #18: - CONTEXT.md adds the 3W language (Well, Instance, Well kind, Event class, Transient label, Observation, Source file), cross-dataset ambiguity entries for Well and Event, and four 3W dataset sections (tables, operating principles, output layout, pre-publish validation). - ADR-0001 records the observations-layout decision (one file per Instance, hive on event_class) with the four rejected alternatives. - ADR-0002 records the upstream-tag-pinning policy (v2.0.0, event-driven refresh) and why tracking main would silently mutate already-published bytes. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Tracer-bullet slice that establishes the transform/export module layout for the Petrobras 3W dataset and emits its smallest deliverable table. Subsequent slices (#20, #21, #22) add the Instance catalog, real-Well master, and Observations hive partition without re-touching this scaffolding. Key decisions: - Source of truth for event_types is upstream `dataset.ini`. The pipeline parses it via configparser (with `optionxform=str` to preserve hyphenated sensor column names like `P-PDG` and `QGL`) rather than hard-coding canonical values — a future upstream rename surfaces as a parse-time change rather than silent drift. - The shallow-clone is idempotent: presence of `<staging>/dataset/dataset.ini` short-circuits the `git clone`, so the smoke test can stand in a fixture upstream tree without network access. - Pinned identity (git tag `v.1.70.0`, dataset version `2.0.0`) is emitted via the export's `petrobras_3w.export` logger before any parquet write, and is also recorded in `schema.json`, `schema.sql`, `README.md`, and the static-site tab. A parsed `dataset_version` that disagrees with the pin aborts publish (ADR-0002 event-driven refresh). - `has_normal_prefix` is materialised equal to `has_transient` per CONTEXT.md (events 0, 3, 4 carry only the steady class — no NORMAL precursor). - DDL identifier quoting in `schema.sql` is general (`_quote_identifier`) so the hyphenated sensor columns added in #22 round-trip without re-touching the writer. - Website integration uses the same sentinel-comment idempotency pattern as the Argentina integrator. Files added: - `scripts/transform/petrobras_3w/` — constants, upstream stager + ini parser, event_types builder, orchestrator. - `scripts/export/petrobras_3w/` — parquet writer, validator (logs the pin + asserts dataset version + event_types row count/PK/transient invariants), schema doc generator (md/json/sql + README + LICENSE), website integrator, orchestrator. - `parquet/petrobras_3w/` — published deliverables: `event_types.parquet` (10 rows), `schema.md` (with 27-sensor glossary mirrored from upstream), `schema.json`, `schema.sql`, `README.md`, `LICENSE-3W-DATA.md` (CC BY 4.0 + attribution). - `tests/petrobras_3w/test_smoke.py` — end-to-end coverage of the scaffolding (event_types contents, doc generation, website idempotency, pin logging, validator abort paths). - `tests/fixtures/petrobras_3w/dataset/dataset.ini` — byte-identical mirror of upstream at the pinned tag, so tests run without a clone. Files modified: - Root `README.md` and `parquet/index.html` patched via the new website integrator (new Petrobras 3W entry / tab parallel to Argentina, Volve, FORCE 2020). - `.gitignore` — exclude the `data/petrobras_3w/` staging dir and the intermediate `database/petrobras_3w.duckdb` file. Notes for next iteration: - #20 (instances.parquet), #21 (wells.parquet), #22 (observations/) all extend `scripts/transform/petrobras_3w/orchestrator.py` and `scripts/export/petrobras_3w/{parquet_writer,validator,schema_doc_generator}.py`. - The validator's structural checks (#19 only covers event_types row-count, PK uniqueness, has_transient/transient_code invariants + pin assertion) need to be extended with the remaining seven rules from CONTEXT.md as later tables land. - Pre-existing test failure in `scripts/export/test_integration.py` (Volve-era, missing `parquet/wells.parquet`) is unrelated to this slice and remains as-is. Closes #19 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Tracer-bullet extension of the #19 scaffolding that emits the full Instance catalog. One row per upstream per-Instance parquet, materialised via pure DuckDB SQL: a single `read_parquet(..., filename=true)` pulls every row, the filename is parsed into `instance_id` / `well_kind` / `well_id`, and the per-file aggregates (`start_ts`, `end_ts`, `duration_s`, `n_rows`, four `n_rows_*` counts, `source_url`) are computed in the same query. Key decisions: - `n_rows_normal` is defined as `class = 0 AND event_class <> 0` (the NORMAL precursor before an anomaly). Strict literal reading of CONTEXT.md ("rows where `class = 0`") would double-count event 0, where `class = 0` and `class = event_class` overlap. The refined definition makes the four `n_rows_*` columns always partition `n_rows` without overlap, which is the operational requirement stated in acceptance criterion 5. CONTEXT.md, the column doc, and the validator rule are all aligned on this. - `n_rows_transient` is materialised NULL exactly when the event's `has_transient = false` — joined from `event_types` rather than hardcoded to `{0, 3, 4}` so a future upstream toggle surfaces as a validation failure rather than a silent shift. - `well_id` strips leading zeros from `WELL-NNNNN` and is NULL for simulated/drawn — enforced by the validator's well-kind contract. - `source_url` is computed from the ADR-0001 URL pattern at catalog build time, before `observations/` exists. A new `PUBLIC_BASE_URL` constant centralises the base URL. - Test fixtures use DuckDB to materialise five small instance parquets (one per representative event class, spanning real/simulated/drawn). The smoke test exercises the catalog contents, the well-kind/well-id contract, the four-bucket accounting, the source_url shape, and five validator abort paths. Files added: - `scripts/transform/petrobras_3w/instances_builder.py` — single-CTE SQL pipeline (raw → parsed → aggregated → joined-with-event_types). - `tests/petrobras_3w/conftest.py` — `build_instance_parquets` helper that emits the deterministic per-test fixture tree. Files modified: - `scripts/transform/petrobras_3w/{orchestrator,constants}.py` — wire the new builder, add `PUBLIC_BASE_URL`. - `scripts/export/petrobras_3w/parquet_writer.py` — add `write_instances`. - `scripts/export/petrobras_3w/validator.py` — rule 1 (instance_id unique), rule 4 (event_class FK), well-kind contract, transient nullness vs event_types.has_transient, four-bucket accounting. - `scripts/export/petrobras_3w/schema_doc_generator.py` — register `instances` in `TABLE_ORDER` + per-column descriptions; extend the README with corpus-balance and per-event-class-filter examples. - `scripts/export/petrobras_3w/website_integrator.py` — bump tab count to 2 files, surface `instances.parquet` in the download grid, swap the README/page query example for the new catalog-balance one. - `scripts/export/petrobras_3w/orchestrator.py` — call `write_instances` after `write_event_types`. - `tests/petrobras_3w/test_smoke.py` — extend documentation/website tests + add seven `instances`-specific tests (contents, well_kind, row-count accounting, source_url, source_file, plus four validator abort paths). - `CONTEXT.md` — clarify the `n_rows_normal` definition. Notes for next iteration: - The five remaining pre-publish validation rules from CONTEXT.md (rules 2, 3, 5, 6, 7, 8, 9) still need to land. Rule 7's expected real-Well count is **40** (resolved in triage on #21). - Issue #21 (wells.parquet) builds on `instances` via `well_kind = 'real'` aggregation; #22 (observations) extends the validator with the per-Observations-file invariants (rule 5/6). - Pre-existing test failure in `scripts/export/test_integration.py` (Volve-era, missing `parquet/wells.parquet`) remains as-is, same as #19 noted. Closes #20 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Tracer-bullet extension of the #20 instances catalog that derives the real-Well master. One row per distinct `well_id` from instances where `well_kind = 'real'`, with per-Well aggregates (`n_instances`, `first_ts`, `last_ts`, `n_observations`). Built via pure DuckDB SQL against the existing intermediate DB — no additional upstream staging. Key decisions: - Rule 7's expected real-Well count is pinned to **40** in the validator (`EXPECTED_REAL_WELL_COUNT`), matching the locked decision from triage. Upstream's `dataset/README.md` claims "42 real wells covered" but only 40 IDs (`00001..00016`, `00019..00042`) actually appear in instance filenames at `v.1.70.0` / dataset version `2.0.0` — IDs 17 and 18 are absent. Pinning on the observed 40 preserves the fail-loud-on-upstream-drift property. - Validator order is rule 3 (FK from instances to wells) → rule 7 (rowcount) → rule 8 (wells limited to real well_ids). Rule 3 fires first because an orphan instance reference is the more fundamental invariant: a missing wells row breaks every downstream join, whereas rule 8 is the symmetric check that the master does not include synthetic IDs. - Wells master deliberately excludes simulated and drawn instances by filtering at aggregation time. Upstream anonymises every physical attribute (no basin, field, depth, location), so the table is identity-plus-statistics only. - Test fixtures gain 37 single-row event-0 padding instances so the total distinct real well count matches the pinned 40. The IDs leave the upstream gap at 17/18 intact, so happy-path tests exercise rule 7 against a realistic catalog rather than a degenerate one. The existing 5 primary fixtures keep their event-class / well-kind mix for the catalog-shape assertions in #20's tests. Files added: - `scripts/transform/petrobras_3w/wells_builder.py` — pure-SQL aggregation from `instances WHERE well_kind = 'real'`. Files modified: - `scripts/transform/petrobras_3w/orchestrator.py` — wire the wells builder after the instances builder. - `scripts/export/petrobras_3w/{parquet_writer,orchestrator}.py` — add `write_wells` and call it after `write_instances`. - `scripts/export/petrobras_3w/validator.py` — rule 3 (`WellsIdFkError`), rule 7 (`WellsRowCountError`, `EXPECTED_REAL_WELL_COUNT = 40`), rule 8 (`WellsKindError`). - `scripts/export/petrobras_3w/schema_doc_generator.py` — register `wells` in `TABLE_ORDER`, add per-column descriptions, declare the `instances.well_id → wells.well_id` FK, add a wells/instances join query example to README. - `scripts/export/petrobras_3w/website_integrator.py` — bump tab count to 3 files, surface `wells.parquet` in the download grid, refresh the description blurb. - `tests/petrobras_3w/conftest.py` — extend `_INSTANCE_SPECS` with 37 padding real-Well fixtures (IDs 4..16, 19..42). - `tests/petrobras_3w/test_smoke.py` — relax the `test_pipeline_emits_instances_catalog` exact-row-count assertion; add 6 wells-specific tests (contents, exclusion of non-real, aggregates vs instances, rule 3 / 7 / 8 validator abort paths); extend the documentation + website-integration tests to cover the new table. Notes for next iteration: - The remaining pre-publish validation rules from CONTEXT.md (rules 2, 5, 6, 9) cover the Observations files and land with #22. - Pre-existing test failure in `scripts/export/test_integration.py` (Volve-era, missing `parquet/wells.parquet`) remains as-is, same as #19 and #20 noted. Closes #21 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Tracer-bullet extension of #21 that emits the per-Instance Observations time-series, completing the four-table Petrobras 3W shape. Each upstream Instance parquet is republished to `parquet/petrobras_3w/observations/event_class=N/<instance_id>.parquet`, preserving the source columns verbatim (including hyphenated sensor names) and adding three RLE-encoded constants: `instance_id`, `well_id`, `well_kind`. `event_class` lives in the hive partition path only — never in the file body. A `_files.json` manifest at the partition root enumerates every published file. Key decisions: - `observations` is exposed as a DuckDB view over the staged sources rather than a materialised table. The full corpus (~2,228 files × tens of thousands of rows × 30 columns) is too large for memory; the view's columns are derived on read and the validator pivots on them without materialising. `union_by_name=true` keeps the view tolerant of test fixtures that only carry a subset of columns. - The writer iterates the catalog and runs one `COPY` per Instance against the matching staged source file. Going through the view's `WHERE instance_id = ?` would re-scan every staged file on every iteration (O(n²)); per-file COPY is O(n) total. - Rule 9 (50 MB soft-warn) is emitted by `write_observations` after each file is written, not by `validator.validate`. The other Observations rules (2, 5, 6) run pre-write against the view because they catch data-semantic problems before any byte hits disk. - `event_class` is documented as a logical column of `observations` with a new `hive_partition: true` flag in `schema.json`/`schema.md`, surfaced as a "Source" column in the table docs. The reflection passes `hive_partitioning=false` to read parquet metadata so DuckDB doesn't synthesize `event_class` from the parent dir name and double-count. - The export orchestrator now takes a `staging_dir` parameter — the Observations writer needs direct access to the source files. The catalog tables (event_types/instances/wells) remain staging-independent (they live as materialised tables). - Validator rule 5 is split into four pre-existing-style functions (FK, row count, timestamp monotonic, class domain) so each can fire with a precise exception class and the test suite can target them individually. The "single (instance_id, well_id, well_kind) triple per file" sub-clause is satisfied by construction (constants derived per filename) and is therefore not re-checked at runtime. - Test fixtures gain a hyphenated `P-PDG` sensor column + `state` so the column-name fidelity policy is exercised end-to-end on the written parquets, not just in the docs. - The validator-rejection tests for rules 5 (timestamp / class domain) and 6 (class) mutate the staged Instance file *after* transform builds the catalog, preserving the catalog's bucket accounting (rule from #20) so the targeted Observation-side rule is the failing one rather than the upstream catalog check. Files added: - `scripts/transform/petrobras_3w/observations_builder.py` — view with derived `event_class` / `instance_id` / `well_id` / `well_kind`. Files modified: - `scripts/transform/petrobras_3w/orchestrator.py` — wire the new builder after `wells_builder`. - `scripts/export/petrobras_3w/parquet_writer.py` — add `write_observations` (per-Instance COPY + `_files.json` manifest + rule 9 soft-warn). - `scripts/export/petrobras_3w/orchestrator.py` — pass `staging_dir` through; call the new writer after `write_wells`. - `scripts/export/petrobras_3w/validator.py` — new exceptions (`ObservationsInstanceFkError`, `ObservationsRowCountError`, `ObservationsTimestampError`, `ObservationsClassDomainError`, `ObservationsNonTransientClassError`) and the five rule checks (rules 2, 5×3, 6). - `scripts/export/petrobras_3w/schema_doc_generator.py` — register `observations` in `TABLE_ORDER`, add description / column docs / PK / FKs, prepend `event_class` as a hive-partition column, surface the four new query examples in `README.md` (event-class scan, single-Instance fetch, manifest enumeration, hive glob). - `scripts/export/petrobras_3w/website_integrator.py` — bump tab count, surface `observations/_files.json`, swap the placeholder follow-up paragraph for the hive-glob query example. - `tests/petrobras_3w/conftest.py` — add `state` and `P-PDG` columns to the per-Instance fixtures. - `tests/petrobras_3w/test_smoke.py` — add 9 observations-specific tests (layout, hyphen fidelity, NULL well_id for non-real, manifest, README query, schema.sql round-trip, rules 2 / 5 / 5 / 6 abort paths) + thread `staging_dir` through every export call. Notes for next iteration: - Issue #23 (parity test suite vs upstream) extends the validator with the nine parity queries described in PRD #18; the Observations view defined here is the SQL surface those parity queries will run against. - Issue #24 (final docs + website polish) refines the README/schema examples and the index page beyond this slice's tracer-bullet level. - Pre-existing test failure in `scripts/export/test_integration.py` (Volve-era, missing `parquet/wells.parquet`) remains as-is, same as #19/#20/#21 noted. Closes #22 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds the nine-check parity suite from PRD #18, wired in as the post-write correctness gate. Where the existing pre-write `validator` asserts structural invariants of the intermediate DB, the new `parity.check(staging_dir, output_dir)` proves the published bytes round-trip the upstream bytes 1:1 — class, state, timestamp, and every sensor column are preserved verbatim, so any divergence is a writer bug rather than legitimate transformation. Checks (PRD #18 order, all hard-fail on divergence): 1. Per-event-class row count — upstream / catalog / pub. 2. Per-instance row count — upstream / catalog / pub. 3. Global `class` distribution — upstream vs pub. 4. Global `state` distribution — upstream vs pub. 5. Per-sensor SUM/AVG/MIN/MAX/COUNT/NULL — upstream vs pub. 6. Per-sensor aggregates grouped by `event_class` — same metrics. 7. Per-instance (start_ts, end_ts) — upstream / catalog / pub. 8. Distinct real-Well count — upstream / wells.parquet / pinned 40. 9. Per-event-class instance count — upstream vs catalog. Key decisions: - Parity runs *after* the parquets are written but *before* the static-site tab is patched, so a divergence keeps the existing published site intact. Implementation-wise, this means closing the validator's read-only DuckDB connection, calling `parity.check` with its own in-memory connection over the written parquets, then re-opening a read-only connection for the schema-doc generator. - Sensor columns are discovered at runtime from the published `observations` schema (everything that is not an identifier/label/time/hive column). Keeps the check agnostic to upstream's exact 27-column set — a future column rename surfaces as a parity match on the renamed column instead of a hard-coded reference to a missing column name. - NULL `class` and `state` values are legitimate on the warmup prefix of real-Well anomaly Instances, so the global-distribution checks (3 / 4) use `IS NOT DISTINCT FROM` for their join predicate. Plain `USING (class)` would split the NULL bucket into two phantom mismatches. - Sensor columns are double-quoted via a `_quote` helper so hyphenated identifiers (`P-PDG`, `ESTADO-SDV-GL`) round-trip safely. - The bit-for-bit comparison policy (no epsilon) is from PRD #18: every sensor float is copied unchanged, so identical SUM/AVG on identical multisets is the contract. If full-corpus runs ever surface non-associativity drift, we can switch to per-instance row-level EXCEPT — for now the simpler aggregate compare suffices. - The real-Well count (check 8) pins on `EXPECTED_REAL_WELL_COUNT` imported from `validator` so the parity suite and the pre-write validator stay in lockstep on the pinned 40. - The orchestrator-level integration test patches `parity.check` to raise, then asserts the website integrator never runs (the stub README + index.html stay unchanged). This proves the wiring without needing to artificially break the published bytes. - The mutation-detection tests use a `_mutate_published_parquet` helper that re-reads a written file with `hive_partitioning=false`, applies a SQL mutation, and overwrites in place. Parallel pattern to `_rewrite_staged_instance` from #22, but operating against the published tree rather than the staged sources. Files added: - `scripts/export/petrobras_3w/parity.py` — module with `check()` entry point and nine `Parity*Error` subclasses, one per PRD check. Files modified: - `scripts/export/petrobras_3w/orchestrator.py` — split the existing single-connection block in two so `parity.check` can run between the writes and the schema-doc / website-integrator steps; updated the module docstring to describe the two correctness gates. - `tests/petrobras_3w/test_smoke.py` — five parity tests: happy-path, sensor-value mutation, partition-internal per-instance drift, row drop (per-event-class), class-label flip, and orchestrator-aborts- on-parity-failure. Notes for next iteration: - Issue #24 (final docs + website polish) is now unblocked. - The full-corpus parity run is part of the production publish path via the orchestrator hook; the smoke suite runs the 42-instance subset version under 20 s on the fixture, well within the 30 s cap. - Pre-existing test failure in `scripts/export/test_integration.py` (Volve-era, missing `parquet/wells.parquet`) is unchanged. Closes #23 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Refreshes the published documentation surface for the Petrobras 3W dataset to the final four-table state. The committed `parquet/petrobras_3w/` docs and the static-site entry were last touched at the #19 skeleton slice and only described `event_types`; this PR brings them in lockstep with the writers landed by #20 (`instances`), #21 (`wells`), #22 (`observations`), and #23 (parity). Generator improvements: - `schema.md` sensor-glossary intro drops the stale "once issue #22 lands" forward-reference — those bytes have shipped, so the prose now describes the observations columns as present. - The observations table's per-column docs now inline upstream's sensor descriptions from `dataset.ini` (e.g. `P-PDG` → "Downhole pressure at the PDG (permanent downhole gauge) [Pa]") instead of leaving the Description cell blank with the glossary as the only reference. - README adds the per-Well cross-validation split query (acceptance criterion c from issue #24) — leave-one-Well-out via `well_id % n_folds`, with a follow-up note on handling simulated/drawn Instances. The four canonical examples (a load-by-event-class, b fetch-by-URL, c per-Well CV split, d corpus-balance from catalog only) are now all surfaced. Test fixture upgrade: - `tests/petrobras_3w/conftest.py` reads upstream's `PARQUET_FILE_PROPERTIES` and writes all 27 sensor columns into the per-Instance fixtures (placeholder values for the 26 non-P-PDG columns; `P-PDG` keeps its row-varying values so parity sensor aggregates still differentiate). This makes the fixture-derived schema docs structurally match production, which is what lets us commit them. The lone hyphenated `P-PDG` column was enough to exercise column-name fidelity end-to-end at #22's level, but it understated the column inventory in the reflected schema. - `test_parity_detects_per_instance_only_drift`'s INSERT now uses `INSERT BY NAME SELECT *` so it doesn't need to enumerate every sensor column by hand. Regenerated committed artifacts: - `parquet/petrobras_3w/schema.md`, `schema.json`, `schema.sql`, `README.md` regenerated against the upgraded fixture — `schema.sql` now declares all 27 sensor columns with hyphen-quoted identifiers, `schema.json` carries the full column / FK / PK metadata for all four tables, `schema.md` and `README.md` carry every per-table description and every canonical query example. - Root `README.md` and `parquet/index.html` regenerated via the website integrator — the tab count was stuck at "1 file" since #19 and the README blurb still claimed the catalog and Observations "ship in follow-up issues". Both now reflect the realised four-table state. Key decisions: - Committed schema docs are produced from the FIXTURE pipeline, not from a full upstream publish (3+ GB unstageable). The docs describe shape, not content — bumping fixture column inventory to match upstream's 27-column inventory makes the fixture-derived docs production-correct. The actual data parquets (`wells`, `instances`, `observations/`) still come from the upstream publish at deploy time; only `event_types.parquet` is byte-stable across the two paths because it is dataset.ini-driven. - Per-Well CV split uses `well_id %% n_folds` as the fold key. Stable across refreshes and zero-state (no separate fold-assignment file needed); consumers picking another assignment can replace the `%% 5` clause without touching the rest of the query. - `dataset.ini` parsing is reused in the test conftest via `parse_dataset_ini` rather than hard-coding the 27 column names — any future upstream column rename surfaces at fixture build time as a parse-time mismatch, the same fail-loud property the production pipeline has. Files modified: - `scripts/export/petrobras_3w/schema_doc_generator.py` — sensor- description inline lookup + glossary-intro prose update + new CV-split README section. - `tests/petrobras_3w/conftest.py` — 27-sensor fixture columns from upstream `dataset.ini`. - `tests/petrobras_3w/test_smoke.py` — assert no stale "#22 lands" prose; assert sensor descriptions are inlined in observations table; assert the four canonical queries (a/b/c/d) are present; switch the per-instance-drift parity test's INSERT to `BY NAME SELECT *`. - `parquet/petrobras_3w/{schema.md,schema.json,schema.sql,README.md,LICENSE-3W-DATA.md}` — regenerated. - `README.md`, `parquet/index.html` — refreshed via the website integrator. Notes for next iteration: - Pre-existing test failure in `scripts/export/test_integration.py` (Volve-era, missing `parquet/wells.parquet`) is unchanged, same as every #19..#23 noted. - The deploy step that copies the full ~2,228-file Observations tree to `dev-petrodb.ocortez.com` is out of scope for this PR — the pipeline running in production is what materialises those bytes. The committed schema docs accurately describe what the deployed tree contains. Closes #24 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Replaces the GitHub Pages workflow (which would fail on the 1 GB cap with
the 3W tree) with a single Cloudflare Pages workflow that handles both
environments via `wrangler --branch ${{ github.ref_name }}`: push to
`main` lands in production, push to `stage` in the preview deployment.
The 3W generated tree is gitignored — the workflow rebuilds it from the
pinned upstream tag (ADR-0002) and uploads the whole `parquet/` directory
so committed datasets ride along. Pipeline re-runs are cached on the 3W
sources + per-branch `BASE_URL`, so non-3W changes deploy without
rebuilding. `PUBLIC_BASE_URL` is now host-only via env, with the
`/petrobras_3w` segment derived from the dataset's directory.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Sync 3W dataset + Cloudflare Pages migration into dev
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.