Leverage opm-common keyword definitions to expand VS Code extension features

[magnesj]

Summary

The repo ships opm-common as a submodule and merges its parser-truth keyword
definitions into the extension's keyword index. The index now exposes sections,
per-item value_type / dimension / default / options, per-record arity
(expected_columns, records_meta), and the derived size_kind / size_count /
templated / variadic_record shape fields. This plan enumerates the remaining
opportunities to deepen the editor's understanding of OPM Flow decks.

Plan status (updated 2026-06-29). The diagnostics, multi-record, outline and
docs-panel work has landed and the index pipeline has grown the per-record /
shape fields it needed. Items 1 (requires/prohibits diagnostics), 2
(positional value_type checking) and 4 (deck_names aliases) are now done —
all verified to add zero false positives on the known-good opm-tests corpus.
A UDQ / ACTIONX sub-language layer also shipped (highlighting, hover,
completion, two new diagnostics, and expression-block alignment) — recorded
under Shipped beyond the original plan. The remaining gaps are the rest of
item 9's index fields (size_type/dialect), code actions (quick-fixes), and the
richer completions (snippets, UDA-aware, well-name). See the Progress section
at the bottom for a per-item rollup.

Why now

• opm-common is already wired in as a sparse submodule, the index build merges it, and CI rebuilds on every release. The data is reliable and free.
• We have 41 keywords with requires, 35 with prohibits, 236 with size_type, 35 with deck_names (aliases), and a records[] schema for multi-record keywords like ACTIONX. The records[] schema is now consumed; the others are still not surfaced.
• The arity diagnostic is currently total-tokens-vs-expected_columns (now per-record for multi-record keywords). We have full per-position value_type and dimension available, so we can do positional type checking for free.

Current vs. available metadata

Metadata in opm-common	Used today	Possible feature
sections	✔ section-validity diagnostic, hover, completion	—
value_type per item	✔ shown in hover/sidebar + per-position type diagnostic	smarter completions
dimension per item	✔ shown in hover/sidebar	unit-aware hover (FIELD/METRIC/LAB)
default per item	✔ shown in hover/sidebar	inline-hint adornment, snippet placeholders
options per item	✔ value completion + sidebar	—
records[] (multi-record schema)	✔ records_meta: per-record arity, hover & docs tables	—
size shape (→ size_kind/size_count)	✔ terminator & continuation diagnostics	—
requires (41 kw)	✔ emitted + "X requires Y" diagnostic	quick-fix (item 5)
prohibits (34 kw)	✔ emitted + "X conflicts with Y" diagnostic	—
size_type, size{keyword,item}	✘	row-count validation against TABDIMS etc.
deck_names + deck_name_regex	✔ aliases resolved + alias_of in hover; PROBE/region families expanded; container schema names suppressed	—
data (free-form bulk keywords)	partial (size_kind: array)	mark COORD/ZCORN/ACTNUM blocks differently
Dialect dir (000_Eclipse100 … 900_OPM)	✘	tag completions/hover with dialect (E100/E300/OPM)

Proposed work

1. Diagnostics — cross-keyword constraints — ✅ done

• requires check: a keyword with non-empty requires whose required partner is absent gets a "X requires Y, which is not present in the deck" warning. Suppressed for INCLUDE-bearing decks (the partner may live in an included file) and for bare INCLUDE fragments (no section header — a .inc/.grdecl legitimately omits the phase keywords its tables require). Phase requirements (OIL/GAS/WATER) are also treated as satisfied under implicit-phase modes (CO2STORE/H2STORE).
• prohibits check: warns "X cannot be used together with Y; they are mutually exclusive" when both are present, reported once per unordered pair (fires even when only one side declares the constraint, e.g. THERMAL prohibits TEMP).
• Tests added in vscode-extension/src/analysis.test.ts; zero false positives on the known-good opm-tests corpus.

2. Diagnostics — positional type checking — ✅ done (enum check intentionally dropped)

• computeDiagnostics now checks each record token against the matching item's value_type with a precise per-token underline range:
  • INT: flags a decimal/float or a quoted string; integers and defaults (*, N*) pass.
  • DOUBLE: flags a quoted string; numerics (incl. scientific) pass.
  • Bare identifiers in numeric slots (UDA/UDQ references, macro substitutions) are left alone — the line-oriented engine can't resolve them. UDA items accept anything. Repeat-values (N*VALUE) check the value; multi-record keywords use the active record's params; variadic-record table keywords are skipped.
• Arity-overflow check unchanged; columns beyond the mapped parameters are not type-checked.
• STRING options[] membership check — deliberately not implemented. The option sets are extracted heuristically from manual prose and are frequently incomplete/abbreviated (e.g. WCONINJE's phase lists WAT, not the valid WATER), producing ~3000 false positives on the known-good opm-tests corpus. Options remain in the index for completions only. Revisit only if a reliable enum source (opm-common itself) becomes available.

3. Diagnostics — multi-record schema (records[]) — ✅ done

• Per-record arity: records_meta[currentRecord-1].expected_columns drives the over-arity check; the message names which record overflowed.
• A record cursor (currentRecord) advances on each terminating /, capped so the trailing variadic record absorbs all remaining rows (WELSEGS, VFPPROD, COMPSEGS, ACTIONX, TUNING…).
• Hover and the docs sidebar resolve the parameter against the active record and render one parameter table per record.

4. Aliases (deck_names) — ✅ done

• Emit deck_names from the index build: each concrete deck keyword becomes (or is tagged on) an entry carrying alias_of naming its opm-common family (WOPR→WELL_PROBE, KRNUMX→KRNUM). The 31 family/container schema names (WELL_PROBE, AQUIFER_PROBE_ANALYTIC, MULT_XYZ, …) are suppressed from the index — they are never typed in a deck — so they no longer pollute completions or pass as valid keywords. Hover surfaces a Deck-name alias of <family> line.
• Honour deck_name_regex for the open-ended summary-vector families (UDQ, tracer, water-cut-bucket mnemonics) via summary_name_patterns.json; PROBE deck_names, directional KRNUM/IMBNUM region variants, L-modifier vectors (WOPRL, LWWIR) and FIP region-set vectors (ROIP_ABC) are all recognised. Verified zero new false positives on the opm-tests corpus.

5. Code actions (quick-fixes)

• Move keyword to correct section when section-validity fails: insert the keyword under the first valid sections[] entry that exists in the document.
• Insert missing requires keyword with default record (using item defaults) under RUNSPEC.
• Trim record to allowed length for the arity diagnostic.
• Insert missing / terminator for the terminator diagnostics.
• Uppercase / unindent keyword quick-fixes for the lowercase and column-1 diagnostics.
• Fill defaults — replace a record line with one expanded to all items using the documented defaults (handy for WCONPROD/WCONINJE).

6. Smarter completions

• Value completion inside records when the active item has options[] (e.g. COMPDAT STATUS), honouring an opm-flow.completion.stringValueStyle setting (bare/quoted/both).
• UDQ / ACTIONX completion — inside a UDQ block the start of a statement offers the control words (ASSIGN/DEFINE/UNITS/UPDATE); inside a UDQ formula or ACTIONX condition the UDQ functions (SORTA, SUM, ABS, …) are offered and inserted with parentheses.
• Snippet completion for keywords that scaffolds a full record line with tab-stop placeholders pre-filled from default/options[0] (e.g. typing WCONPROD inserts the keyword and a record line scaffold).
• UDA-aware completion: when the active item has value_type: UDA, scan the deck for UDQ definitions and offer those names alongside numeric input.
• Cross-keyword name completion: when the parameter is a well name (heuristic by item name WELL / GROUP), collect names from WELSPECS/GRUPTREE and offer them. #21

7. Outline & navigation

• Outline tree view (opm-flow.outlineView): sections become folders, keywords become leaves; selection follows the cursor and clicking reveals the keyword. (Implemented as a custom TreeDataProvider rather than a DocumentSymbolProvider.)
• Folding of section blocks and individual keyword blocks via a FoldingRangeProvider.
• File navigation — Ctrl+click quoted paths on INCLUDE/IMPORT/RESTART/GDFILE, with PATHS alias ($NAME) expansion (DocumentLinkProvider).
• Consider also exposing a DocumentSymbolProvider so the built-in Outline view / breadcrumbs / Ctrl+Shift+O work too (the tree view is separate from VS Code's native outline).
• DefinitionProvider so Ctrl+click on a well name in WCONPROD jumps to the WELSPECS entry that introduced it.

8. Hover & sidebar additions

• Sidebar docs panel that follows the cursor, with the active parameter row highlighted and a per-record table for multi-record keywords.
• Configurable Type/Field/Metric/Lab/Default columns (opm-flow.columns.*, opm-flow.units.*) and an embedded vs. columnar layout (opm-flow.docs.layout).
• Hover warns when a keyword is used in the wrong section and when it is on the diagnostics exclusion list.
• Hover on a deck-name alias shows its opm-common family (Deck-name alias of <family>); hover on a UDQ control word or function shows its meaning/signature.
• Show the dialect tag (E100 / E300 / Frontsim / OPM-only) next to the section list (requires emitting dialect from the index build).
• Show requires / prohibits as a "Related keywords" line.
• Convert dimension → a concrete unit string for the active unit system (FIELD/METRIC/LAB) using the manual's unit table; controlled by an opm-flow.unitSystem setting. (Today the three unit columns are shown/hidden but not resolved to one active system.)

9. Index build pipeline

• Emit per-record / shape fields the diagnostics engine needs: records_meta, size_kind, size_count, templated, variadic_record, plus per-parameter value_type / dimension / options / units.
• Emit the cross-keyword constraint fields requires / prohibits (loaded in load_opm_common_index, copied onto merged + synthesized entries via _attach_cross_keyword, emitted in the compact JSON; 41 requires, 34 prohibits).
• Emit alias fields: expand_probe_deck_names expands PROBE deck_names and directional variants into index entries tagged with alias_of; synthesize_opm_only_entries now skips family/container schema names that only carry deck_names; deck_name_regex families ship as summary_name_patterns.json.
• Still to emit: size_type (+ size{keyword,item}) and a dialect tag.
• Add a fast smoke check: every keyword in opm-common either has a manual .fodt source or is marked synthesised (formalise the gap report).
• Watch VSIX size as the new fields land; consider gzip-on-disk + decode-on-load if the compact JSON pushes past ~3 MB.

10. Documentation & examples

• README.md Features list tracks shipped features (diagnostics, hover, sidebar, outline, folding, align/headers, file navigation, toggle comment, open PRT, generate reference).
• Add a short section to DEVELOPMENT.md describing the opm-common JSON schema fields the extension consumes, so contributors know where to look.

Shipped beyond the original plan

These were not in the first cut of the plan but landed alongside it and are worth recording:

• Richer diagnostics — unrecognised-keyword, wrong-section, per-record arity, missing per-record /, missing list/array close /, keyword-not-in-column-1, and lowercase-keyword checks; a opm-flow.diagnostics.excludedKeywords opt-out list with a default set.
• Templated SUMMARY mnemonics — FTPR→FTPRSEA style tokens resolve to the base template for hover/docs/diagnostics (templated, variadic_record, optional_body).
• UDQ / ACTIONX sub-language — recognition of the UDQ expression language and ACTIONX action blocks: tmLanguage highlighting for control words, UDQ functions, expression operators and the ACTIONX/ENDACTIO delimiters; hover and completion for UDQ control words/functions; two diagnostics (a UDQ statement that doesn't start with a control word; an ACTIONX block left unclosed by ENDACTIO); and a dedicated three-column alignment for UDQ expression blocks (control word and variable both left-aligned, expression right-aligned, division-safe, comment/blank-line tolerant).
• Corpus false-positive harness — vscode-extension/src/corpus.test.ts runs the diagnostics engine over the OPM/opm-tests decks (known-good) to catch regressions; documented in DEVELOPMENT.md.
• Formatting commands — Align Record Columns, Add Column Headers (idempotent, heading-aware), Toggle Line Comment.
• Open PRT File command + Ctrl+Alt+P keybinding.
• opm-flow.additionalFileExtensions setting to retag arbitrary extensions as opm-flow.
• Security hardening of the webview/hover and CI (no localResourceRoots, CSP, no command: links).

Out of scope (call out, don't implement here)

• A full deck parser. We stay line-oriented and lean on the index for correctness signals.
• Running OPM Flow as a language server. Static analysis only.

Progress rollup

Plan item	Status
1. requires / prohibits diagnostics	✅ done
2. Positional value_type checking	✅ done (numeric/quoted checks; enum-options check intentionally dropped)
3. Multi-record records[] schema	✅ done
4. deck_names aliases	✅ done (aliases resolved + alias_of hover; PROBE/region/regex families expanded; container schema names suppressed)
5. Code actions / quick-fixes	⬜ not started
6. Smarter completions	🟡 value/options + UDQ control-word/function completion done; snippets / UDA / well-name pending
7. Outline & navigation	🟡 tree view + folding + file links done; DocumentSymbolProvider & DefinitionProvider pending
8. Hover & sidebar	🟡 panel + columns/layout + section/exclusion warnings + alias/UDQ hover done; dialect tag / related-keywords / unit-system conversion pending
9. Index build pipeline	🟡 shape, per-item, requires/prohibits & alias (deck_names) fields done; size_type/dialect fields pending
10. Documentation	🟡 README tracked; DEVELOPMENT.md schema notes pending

Suggested staging (revised)

1. Phase A — index + cross-keyword diagnostics: items 9 (cross-keyword fields), 1, 2, 4. Purely additive; unlocks the highest-value diagnostics. (✅ complete: 1, 2 and 4 done; item 9 has only size_type/dialect left.)
2. Phase B — code actions & hover polish: items 5, 8 (dialect tag, related keywords, unit-system conversion).
3. Phase C — navigation & completions: items 6 (snippets/UDA/well-name), 7 (DefinitionProvider, optional DocumentSymbolProvider).

Links

• opm-common keywords directory: opm-common/opm/input/eclipse/share/keywords/{000_Eclipse100,001_Eclipse300,002_Frontsim,900_OPM}/
• Index builder: scripts/build_keyword_index.py
• Diagnostics module: vscode-extension/src/analysis.ts
• Extension entry point: vscode-extension/src/extension.ts
• Outline builder: vscode-extension/src/outline.ts
• Reference manual submodule: opm-reference-manual/
• Upstream opm-common: https://github.com/OPM/opm-common

[magnesj]

Where each opm-common field flows in the pipeline. Solid arrows are wired today; dashed arrows are the remaining proposed consumers. Updated 2026-06-29: deck_names/deck_name_regex now flow through the build into completions, hover (alias resolution) and diagnostics — solid. Still dashed: size_type, dialect tag, snippet scaffolds, and the requires-driven quick-fix.

flowchart LR
    subgraph SRC["opm-common JSON"]
      S[sections]
      VT[value_type]
      DIM[dimension]
      DEF[default]
      OPT[options]
      REQ[requires]
      PROH[prohibits]
      DN[deck_names / regex]
      RECS[records / size shape]
      ST[size_type]
      DIA[dialect dir]
    end

    SRC -->|build_keyword_index.py| IDX[(keyword_index_compact.json)]

    IDX --> HOV[Hover and sidebar]
    IDX --> COMP[Completions]
    IDX --> DIAG[Diagnostics]

    S --> DIAG
    S --> HOV
    VT --> HOV
    DEF --> HOV
    DIM --> HOV
    OPT --> COMP
    RECS --> DIAG

    REQ --> DIAG
    PROH --> DIAG
    VT  --> DIAG2[positional type check]
    ST  -.-> DIAG
    DN  --> COMP
    DN  --> HOV
    DN  --> DIAG
    DIA -.-> HOV
    DEF -.-> COMP2[snippet scaffolds]
    REQ -.-> CA[Code actions / quick-fix]
    S   -.-> CA

    DIAG --> UI[VS Code Problems]
    DIAG2 --> UI
    CA --> UIQ[Quick-fix lightbulb]
    HOV --> UIH[Hover and Docs panel]
    COMP --> UIC[IntelliSense]
    COMP2 --> UIC

Loading

[magnesj]

Commits related to this plan

Work that has advanced (or set up) the items in this plan since it was opened (2026-04-26).

Ref	Description	Date
574280c	Add UDQ/ACTIONX support: recognition, highlighting, hover, completion, alignment (item 6 UDQ completion, item 8 UDQ hover, beyond-plan UDQ/ACTIONX diagnostics + alignment) — squash-merge of PR #52	2026-06-29
2aab7a8	Resolve deck_name aliases and suppress opm-common family containers (item 4; item 9 alias fields)	2026-06-29
d610100	Expand deck diagnostics: cross-keyword rules, value-type checks, and summary-vector coverage (items 1, 2, 4 summary-vector recognition, 9; corpus harness; DEVELOPMENT.md test docs) — squash-merge of PR #51	2026-06-29
fb010c0	Add positional value-type diagnostics (item 2; enum-options check intentionally dropped) — landed via PR OPM#51	2026-06-28
fd0e898	Add requires/prohibits cross-keyword diagnostics (items 1 + 9 cross-keyword fields) — landed via PR OPM#51	2026-06-28
0ecda00	Rebuild keyword index from opm-reference-manual 52fbb530	2026-06-25
caa2b7a	Update opm-reference-manual submodule to 52fbb530	2026-06-25
3b23a98	Release 0.8.0	2026-06-25
15b3121	Add embedded docs layout to widen parameter descriptions (item 8)	2026-06-25
5c0a319	Add section/keyword outline tree view for navigation (item 7)	2026-06-25
c0502ef	Add Ctrl+Alt+P keybinding for Open PRT File (beyond plan)	2026-06-25
d16237f	Add command to open the corresponding PRT file (beyond plan)	2026-06-25
a7a21d5	Recognise section headers with trailing decoration (item 1/3 diagnostics)	2026-06-25
447122f	Bump version to 0.7.1 and add release notes (#30)	2026-06-05
9d7ebd2	Release 0.7.0 (#29)	2026-06-05
1bc2dad	Add toggle line comment command — issue #27 (beyond plan)	2026-06-04
0e67a95	Security hardening: CI permissions/pinning, webview/hover, policy (beyond plan)	2026-05-23
bc8f4a6	Release 0.6.4	2026-05-19
a53403d	Add opm-flow.additionalFileExtensions setting (beyond plan)	2026-05-19
7f51ea6	Update pytest assertions for SUMMARY mnemonic array reshape (item 3/9)	2026-05-19
96be563	Highlight bare uppercase tokens inside records; tighten keyword-line pattern (item 2/diagnostics)	2026-05-19
d2f4a62	Accept multi-line VFPPROD / VFPINJ records without a closing / (item 3)	2026-05-19
da12bfe	Accept multi-line MESSAGES record (print + stop limits) (item 3)	2026-05-19
aa95020	Recognise UDQ SUMMARY mnemonics (templated, item 9)	2026-05-19
67b3353	Accept bare-stacked SUMMARY mnemonics (optional_body, item 9)	2026-05-19
2a5ba0d	Allow multi-line bodies for SUMMARY list-of-name mnemonics (item 9)	2026-05-19
a915f18	Recognise user-defined FIP region keywords (item 9)	2026-05-19
5f7bb8f	Recognise more OPM include-file extensions (item 7 file nav)	2026-05-19

[magnesj]

Plan refresh — 2026-06-28

Reviewed the extension against this plan and updated the body to match how the
work has actually evolved. Highlights:

Landed since the plan was written

• Item 3 (multi-record records[]) is fully done — records_meta drives per-record arity, a record cursor advances on each /, and hover + the docs sidebar render one parameter table per record.
• Item 7 (navigation) is largely done, just not the way the plan first imagined: an outline tree view (custom TreeDataProvider), block folding, and Ctrl+click file links with PATHS alias expansion all shipped.
• Item 8 (hover/sidebar) has the cursor-following docs panel, configurable Type/unit/Default columns, an embedded-vs-columnar layout, and wrong-section / exclusion warnings.
• Item 6 (completions) has value/option completion with a quoted/bare style setting.
• Item 9 (index pipeline) now emits the shape fields the diagnostics engine needs (records_meta, size_kind, size_count, templated, variadic_record) plus per-item value_type/dimension/options/units.
• A large body of diagnostics work not in the original plan also landed: unrecognised-keyword, column-1, lowercase-keyword, missing record/list/array terminator checks, templated SUMMARY mnemonics, and a configurable exclusion list. Plus formatting commands (align, headers, toggle comment), Open PRT, additionalFileExtensions, and security hardening.

Key insight on what's left
The remaining gaps cluster around cross-keyword intelligence. They are blocked on one upstream step: the index build does not yet emit requires, prohibits, deck_names/deck_name_regex, size_type, or a dialect tag (verified: 0 keywords carry these in keyword_index_compact.json, though per-item value_type is present on 3876 params and options on 87). So item 9's second half is the true Phase A — once those fields exist, items 1 (requires/prohibits diagnostics), 2 (positional value_type checking — data is already there), 4 (aliases), and the hover dialect/related-keyword lines all become straightforward.

Still entirely unstarted

• Item 1 — requires/prohibits diagnostics
• Item 2 — positional value_type checking (no index change needed)
• Item 4 — deck_names aliases
• Item 5 — code actions / quick-fixes (no CodeActionProvider registered yet)
• Parts of item 6 — snippet scaffolds, UDA-aware, well-name completion
• Parts of item 7 — DefinitionProvider (well name → WELSPECS), optional DocumentSymbolProvider for the native outline
• Parts of item 8 — dialect tag, related-keywords line, active-unit-system conversion

Open question
Item 2 (positional type checking) needs no index change — the data is already in the
index. Worth pulling forward ahead of item 9's field work, since it's pure
analysis.ts plus tests and delivers value immediately.

[magnesj]

Session summary — 2026-06-28 (Phase A: items 1 & 2)

Completed the two highest-value cross-keyword/positional diagnostics, plus the index fields item 1 depended on.

What landed

• Item 9 (partial) — requires/prohibits index fields. load_opm_common_index now reads both; a new _attach_cross_keyword helper copies them onto merged and synthesized entries; the compact JSON emits them. Result: 41 requires, 34 prohibits entries. (prohibits is 34, not the 35 quoted in the plan — one carrier, ROCK, has malformed JSON in the opm-common submodule and fails to load; pre-existing, unrelated.)
• Item 1 — cross-keyword diagnostics. Document-wide requires ("X requires Y, which is not present in the deck") and prohibits ("X cannot be used together with Y") checks, evaluated once after the line walk.
• Item 2 — positional value-type checking. Per-token check against the matching item's value_type with a precise underline range.

Key insight — the enum/options check is not safe to ship

Plan item 2's third bullet (warn on STRING values outside options[]) was implemented, measured, then removed. Running it over the OPM/opm-tests corpus (1440 known-good decks) produced ~3000 false positives: the option sets are extracted heuristically from manual prose and are routinely incomplete or abbreviated — e.g. WCONINJE's injector phase is stored as WAT rather than the valid WATER, so every real WATER injector was flagged. Options are fine for completions (incompleteness just means fewer suggestions) but unfit for diagnostics. To revisit this we'd need a reliable enum source — ideally opm-common's own value constraints rather than scraped prose.

Corpus-driven hardening of the requires check

The corpus also surfaced 306 requires false positives, all resolved by two suppressions (both verified to bring the count to 0):

1. INCLUDE fragments — a bare .inc/.grdecl with no section header legitimately omits the RUNSPEC phase keywords its tables require, so the check only runs when the input has a section header (and, as before, is suppressed when the deck contains INCLUDE/IMPORT/GDFILE).
2. Implicit-phase modes — CO2STORE/H2STORE activate water+gas without the explicit phase keywords, so phase requirements (OIL/GAS/WATER/…) are treated as satisfied when such a mode is present.

Verification

• 265 jest tests + 118 pytest tests, all green.
• Corpus harness: the three new diagnostic buckets (value-type, requires-missing, prohibits-conflict) are all 0 — net false positives added by this session: zero. (The corpus triage report's classify() was extended to bucket the new types.)

Open questions / follow-ups

• Item 4 (deck_names aliases) is the natural next step to finish Phase A, alongside the remaining item-9 fields (size_type, dialect tag).
• A future "Insert missing requires keyword" quick-fix (item 5) now has the index data it needs.
• Worth confirming the IMPLICIT_PHASE_MODE_KEYWORDS set is complete (currently CO2STORE/CO2STOR/H2STORE) — are there other phase-implying modes that omit explicit phase keywords?

[magnesj]

Session summary — 2026-06-29 (item 4 aliases + UDQ/ACTIONX)

Finished Phase A and shipped a UDQ/ACTIONX sub-language layer. All work is merged into upstream OPM/main and the fork is synced.

What landed

• Item 4 — deck_names aliases (done). Concrete deck keywords now resolve to their opm-common family: each expanded mnemonic carries alias_of (WOPR→WELL_PROBE, KRNUMX→KRNUM), surfaced as a Deck-name alias of <family> line in hover. The 31 family/container schema names (WELL_PROBE, AQUIFER_PROBE_ANALYTIC, MULT_XYZ, …) are suppressed from the index — they are never typed in a deck — so they no longer pollute completions or pass as valid keywords. Open-ended families (deck_name_regex: UDQ, tracer, water-cut buckets) ship as summary_name_patterns.json; PROBE deck_names, directional KRNUM/IMBNUM variants, L-modifier vectors (WOPRL, LWWIR) and FIP region-set vectors (ROIP_ABC) are all recognised. Zero new false positives on the opm-tests corpus.

• UDQ / ACTIONX sub-language (beyond plan).

  • Diagnostics: a UDQ statement that doesn't begin with a control word (ASSIGN/DEFINE/UNITS/UPDATE) is flagged; an ACTIONX block left unclosed by ENDACTIO is flagged.
  • Highlighting: tmLanguage scopes UDQ control words, UDQ functions, expression operators, and the ACTIONX/ENDACTIO delimiters.
  • Hover & completion (items 6/8): hover a UDQ control word or function for its meaning/signature; completion offers control words at statement start and UDQ functions (with parentheses) inside expressions.
  • Alignment: UDQ expression blocks get a dedicated three-column layout — control word and variable both left-aligned, expression right-aligned — that treats a / used for division as part of the expression, and tolerates interspersed comments and blank lines.

• Corpus harness + docs. The opm-tests false-positive harness (corpus.test.ts) and the test-suite docs in DEVELOPMENT.md landed alongside the diagnostics work.

How it merged

PR	Scope	Merge
#51	cross-keyword + value-type diagnostics, summary-vector coverage, corpus harness	squash d610100
(direct)	deck_name aliases + container suppression	2aab7a8
#52	UDQ/ACTIONX recognition, highlighting, hover, completion, alignment	squash 574280c

OPM/main, magnesj/main and local main are all at 574280c; merged feature branches deleted.

Open questions / follow-ups

• Item 4 completion polish: aliases now resolve and appear in completions as concrete entries, but there is no explicit "alias of …" completion label yet — worth deciding whether that adds value over the hover line.
• UDQ_CONTROL_WORDS is defined twice — a local Set in analysis.ts (for the diagnostic) and an exported record in udq.ts (for hover/completion). Harmless but a candidate for consolidation.
• Phase A is complete; the natural next steps are item 5 (code actions / quick-fixes) and the remaining item 9 index fields (size_type, dialect tag), which then unlock the hover dialect/related-keyword lines (item 8).