Turn 3Deploy Gundam Barbatos 1st Form
Develops the strongest legal board / Preserves one active resource
- Deploy Gundam Barbatos 1st Form8.42 heuristic
- Play Improved Technique6.15 heuristic
- Deploy Graze Custom5.88 heuristic
Gundam Match Lab
Gundam Match Lab
Build Better Gundam Decks
Analyse consistency. Test matchups. Optimise your deck using thousands of simulations.
1 Load Deck
→
2 Analyse Consistency
→
3 Improve Deck
→
4 Simulate Matchups
00 / Command
What do you want to prove?
Your deck is ready for analysis.
01 / Build
Current Deck
Edit deck listPaste or revise the 50-card main deck 50-card text list
01A / Analyse
Draw Consistency
Ready
Unofficial fan-made analysis tool. Shared reports contain simulation estimates and may be limited by unsupported card behaviour.
Detailed AnalysisDeclared support, Burst odds, and turn-by-turn access
Load a deck to begin analysis.
04 / Library
Card Library
Loading
Card catalogue
Browse cards
Browse cards by name, set, colour, type, keyword, or effect support.
Filters 0 active
Search the local card catalogue by card, effect, trait, or your current deck.
Quick searches
02 / Improve
Does this change improve my deck?
Compare your Current Deck against a Modified Deck using stable paired experiments. Each pair has its own derived seed and alternating first-player seat; the two game histories remain separate.
Current Deck→Modified Deck→Measure Improvement→Understand Why
Current Deck
Baseline referenceCurrent deck detailsReview the baseline list and declared support
Load a legal 50-card deck in My Deck.
View cards
Modified Deck
Variant testModified deck detailsReview or edit the test list
View or edit cards
Live Comparison
No changes yet
Deck Changes
No changes yet
Clone your current deck, then replace at least one card to begin the comparison.
Experiment Design
Waiting
Test Settings
Optimises for the highest simulated win percentage.
Simulation Plan5,000 paired games
Hand tests10,000 per deck
Expected runtimeMedium
Larger samples improve sampling confidence, but cannot correct card effects the simulator does not execute.
Paired Difference
Waiting
Run Experiment
Make at least one valid card change to continue.
- Preparing decks
- Generating pair seeds
- Running simulations
- Comparing outcomes
- Building recommendation
Clone the current deck, make a swap, then compare.
Measured Outcome
Waiting
Results
Experiment Output
Your results will include
- RecommendationWhich deck performed better
- Estimated improvementThe measured effect size
- ConfidenceSample and reliability context
- Why the result changedTraceable measured contributors
- Detailed comparisonCurrent and modified metrics
No result data is shown until the experiment finishes.
Deck Optimizer
Choose Objective
Balance overall performance, target the matchup, or improve reliable card access.Choose Opponent or Meta Field
Every candidate faces the same selected opponent.Select Swap Limit
More candidates broaden the search and increase runtime.Run Candidate Search
Legal one-card swaps are screened, then retested with a second seed.
Advanced simulation settings
Larger samples improve confidence but do not correct unsupported card behaviour.
Measured Candidates
Review Recommendations
Ready to test candidate swaps.
Recommendations will appear here.
Each result will show the swap, objective signal, second-seed retest, and confidence.
03 / Test
Matchup Test
Run thousands of automated games to measure performance against one opponent or the weighted tournament field.
Choose a legal pilot deck and opponent to configure the test.
How the simulator decidesSee legal choices, heuristic rankings, and model limits
The simulator ranks legal actions with versioned heuristics. Scores compare available actions; they are not probabilities or confidence values.
Published benchmark
Validated: 4 of 4 deterministic checks passed
5,000 mirror games / GD05 constructed / rules-neutral legal mirror / 28 Aug 2026
Policy gml-heuristic-v2 / engine gml-duel-engine-v2
PassedLegal actions and game invariantsObserved 0 illegal actions / 0 invariant violations / target 0 violations
PassedSeed reproducibilityObserved exact seed 73 repeat / target exact repeat
PassedPaired-seat mirror balanceObserved 55.0% first-seat wins / target 52%-56%
PassedRelative heuristic responseObserved 91.8% stronger-curve wins / target at least 75%
Read How AI Plays
Release trendV44100.0% first-seat / gml-heuristic-v1 / limitedV4555.0% first-seat / gml-heuristic-v2 / passed
This benchmark measures deterministic legality, rules-neutral seat balance, and relative heuristic response. Declared effect support is reported separately; this does not measure human-level play or predict tournament results.
03A / Test / Advanced
Scenario Lab
Configure a precise board state and manually test one side against the AI. This advanced tool is intended for rules interactions and edge cases.
Configure the opening Main Phase.
04A / Library / Help
Help & Rules
Getting Started
Load, understand, improve, prove
- LoadPaste a 50-card list or load a tournament result.
- UnderstandReview draw access and declared effect support.
- ImproveCompare a modified list or test measured swaps.
- ProveRun matchups against one deck or the current field.
Official text
Search the Comprehensive Rules and FAQ
Exact official wording with source links. Results load when you start typing.
Match Lab glossaryPlain-language and technical definitions
Release notesRecent product changes
06A / Releases
Changelog
V51
Added / Fixed / Improved / Improved / Data
AddedFixedImprovedImprovedData- Added: Added an experimental Card Atlas IR interpreter: it reads a bounded subset of the reviewed effect programs (choose, damage, rest, activate, destroy, return, recover, draw, and discard with unit selectors, literal and count quantities, and count or presence conditions), plans the eligible targets and amounts for each resolution, and reports whether the simulator agreed. Interpreter agreement appears in the matchup Evidence; clauses resolve through the interpreter only when explicitly allowlisted, and nothing is allowlisted by default.
- Added: Added measured execution agreement: every effect resolution in a matchup run is compared with its Card Atlas program, the agreement rate and disagreeing clauses appear in the matchup Evidence, the Card Inspector shows per-clause agreement from the last run, and the implementation backlog is prioritised by measured disagreement.
- Added: Added an official rules and rulings search to Help & Rules: every Comprehensive Rules 1.8.0 section and every official FAQ ruling from the Card Atlas snapshot, shown verbatim with source links and scope labels for excluded or mixed-scope items.
- Fixed: Corrected Gundam Dynames resting an enemy Unit. Its text contains both "rest this Unit" and "choose 1 enemy Unit", and the generic rest branch fired on any text holding both words, so an enemy Unit was rested alongside the intended damage. The branch now requires the rest to be aimed at the enemy Unit. The measured-agreement shadow had scored the stray rest as a match; only the interpreter, which plans its own targets, caught it.
- Fixed: Corrected Sugai's Gelgoog (GQ) damaging any enemy Unit: "Lv.5 or higher" had no target filter, so a Lv.1 Unit was hit. Deux Murasame's 【Destroyed】 effect never found a target because "undamaged enemy Unit" matched the "damaged enemy" filter first and the list emptied itself. Both filters are corrected.
- Fixed: Corrected three cards whose wording gives the choice to the opponent — "All enemy players each choose 1 of their active Units. Rest them." (Gundam Full Armor (Thunderbolt) (EX), Eternal Road, Sazabi) — resolving against nothing; the opponent now gives up the least valuable Unit the text allows.
- Fixed: Corrected Gundam Geminass 02 returning an enemy Unit without paying for it. "You may choose 1 active friendly Unit that is Lv.3. Rest it. If you do, …" had no handler for resting one of your own Units as a cost, so the return resolved for free while the loose rest branch rested an enemy Unit in its place. The cost is now paid with the least valuable Unit the text allows, and declined when no such Unit exists. Friendly-target filters also now read qualifiers written after "Unit" — "that is Lv.3", "that is Lv.4 or lower", "that is Lv.3 or higher" — which were being dropped entirely, so a friendly effect restricted by level had been choosing among every friendly Unit.
- Fixed: Corrected Commands being played without their leading target. Comprehensive Rules 10-1-8-1-1 and 10-1-8-1-2: a Command whose effect requires choosing a target cannot be played when that target cannot be chosen, judged on the portion preceding "Then" or "If you do". Extreme Hatred ("Choose 2 of your active Units. Rest them. If you do…") was being played with one active Unit and did nothing.
- Fixed: Corrected the Card Atlas interpreter treating a cost as paid when fewer Units were available than the text requires, letting an inactive step count as succeeded for the steps after it, comparing a boolean presence condition as a number (every "if you have another … Unit in play" read as false), and judging plans it could not know — a
<Blocker>requirement no IR selector carries, or a step that depends on a program bound to another clause. Those last two now report as unsupported rather than as disagreements. - Fixed: Corrected a draw inside a battle-destroy trigger being withheld at the trigger's own timing. The earlier fix that stops future-event draws paying out early also stripped the sentence when it was the effect being resolved, so Trowa Barton discarded 1 without drawing 1; Gundam Maxter's and Amuro Ray's triggers were checked and were unaffected.
- Fixed: Corrected a named discard cost going unrecorded. Gwadan, Gundam Mk-II (Titans) and Gundam AGE-1 Normal paid their "discard 1 red card"-style costs through card-specific handlers that moved the card without logging it or reporting it, so the Card Inspector and measured agreement showed the payoff arriving for free. The generic resolver also only recognised the bare "discard 1. If you do" wording. Both paths now pay with a card matching what the cost names, record and log the discard, and when no such card is in hand the effect it buys does not happen.
- Fixed: Corrected measurement gaps that made correctly resolved effects read as disagreements: a top-deck look, reveal and add were seen by the board diff only as the hand growing and reported as a draw; a rest or discard paid as an ability's cost was not reported at all; and a step the board made impossible — no damaged Unit to recover, no rested Unit to activate, no card in hand or trash to deploy or add, a "Then, if…" condition that was false — was scored as a failure to follow the card. Each is now reported by the resolver and excused from the comparison. A top-deck reveal also now honours the card description it names instead of keeping any card on top.
- Fixed: Corrected Willgem entering play without the 3 damage its own 【Deploy】 deals to it; damage aimed at "this Base" had no handler.
- Fixed: Corrected five cards whose targets are written for multiplayer ("1 Unit with 2 or less HP belonging to each enemy player") resolving against nothing in a two-player duel. Strike Freedom Gundam (EX), Hi-Nu Gundam (EX), Warship Cruise, Tiffa Adill & Freeden, and Zeta Gundam Ⅲ P2 Type now find the single opponent's Unit.
- Fixed: Corrected declared effect support overstating three cards that redirect an attack already under way — Master League Begins, Overcoming Hardships, and Armed Intervention. The simulator implements \<Blocker\> redirection but not these, and they now report as partially modelled rather than executed.
- Fixed: Corrected keyword amounts replacing each other instead of adding. Comprehensive Rules 13-1-1-2 and 13-1-2-5 state that a new copy of \<Repair\> or \<Breach\> adds its amount to the amount already there, so a Unit given \<Breach 1\> and then \<Breach 2\> now has \<Breach 3\> rather than \<Breach 2\>; 24 cards grant \<Breach\> and 11 grant \<Repair\>. The keywords printed without an amount still cannot be given twice, per 13-1-4-2, 13-1-5-3, 13-1-6-2 and 13-1-7-2.
- Fixed: Corrected \<Suppression\> resolving one Shield's 【Burst】 before destroying the second Shield. Comprehensive Rules 13-1-7-1 and 13-1-7-4 state that both Shields are damaged and revealed simultaneously, so both are now destroyed before either 【Burst】 resolves.
- Fixed: Corrected optional costs being paid for an effect that could not happen. Comprehensive Rules 10-2-2 and the official Wings of Light rulings state that an effect requiring a target does not activate when no target can be chosen; fourteen cards are written as "You may \<cost\>. If you do, choose 1 enemy Unit…" and now decline the cost rather than spending a Unit, a Resource, or cards from hand or trash for nothing. Per 10-1-8-1-2 a mandatory first portion still resolves either way, so the eight cards written without "You may" are unchanged.
- Fixed: Corrected Gundam Dynames never paying its 【Deploy】 cost: the pattern that rests the Unit could not span the sentence break in the printed text, so it dealt 2 damage for free.
- Fixed: Corrected rest effects wasting themselves on an already-rested Unit. Comprehensive Rules 1-3-2-1 states that putting an entity into a state it is already in performs no action, so Amuro Ray and cards like it now choose an active enemy Unit ahead of a rested one, and a rest that could not be performed is no longer written into the game log as though it happened.
- Fixed: Corrected Strike Freedom Gundam paying an optional cost for nothing: its paired 【Attack】 discarded 2 cards before checking whether an enemy Unit existed to return, so against an empty board it spent two cards and did nothing. The discard is now paid only when there is a Unit to return.
- Fixed: Corrected measured execution agreement counting impossible steps as disagreements. A step the board left no legal target for — Rewloola's Shield when the Shield Area is empty, or a damage step with no enemy Unit in range — is now reported by the resolver and excused from the comparison instead of being recorded as a failure to follow the card. Excused steps are counted and shown in the matchup Evidence so they stay visible. Measured agreement over 400 games reads 100.0% for both decks, with 251 steps excused.
- Fixed: Corrected a startup fault that could leave the workspace open but unresponsive: the rules reference and the legality ruleset were requested without checking the response, so a single failed or incomplete request stopped startup before the run controls were connected, and the page then ignored every click. All seven startup steps now load independently, each reporting its own failure and opening with its documented default, and the Help & Rules panel reads correctly when the rules reference is the part that is missing.
- Fixed: Corrected paired Pilot effects resolving as though the Unit carrying them had produced them: an 【Attack】 effect printed on a Pilot (such as Kira Yamato's AP reduction) was credited to the Unit, so per-card behaviour and effect attribution used the wrong card. Measured execution agreement against the reviewed card text rose from 90.5% to 99.5% for the affected deck once each text resolved as its own card.
- Fixed: Corrected a card-advantage error affecting every card that draws on a future event: an effect written "when they destroy an enemy card with battle damage, draw 1" paid the card out the moment the effect resolved. Strike Freedom Gundam's Deploy also never armed its real trigger, because the pattern that arms it could not span the sentence break in the printed text. Against a Strike Freedom field this moved the measured opponent win rate from 38.5% to 30.2% over 400 games.
- Fixed: Corrected keyword reminder clauses (Blocker, Repair, Breach, Support, High-Maneuver) that the effect manifest reported as unimplemented although the simulator executes those keywords; 96 such clauses now count as executed, none remain "review required", and Blocker decks no longer lower matchup confidence or appear in the implementation backlog for that reason.
- Fixed: Stopped the Matches and Games count field from writing a raw event object into the matchup status line.
- Fixed: Generated the Changelog "Updated" date from the newest release instead of a fixed string.
- Fixed: Fixed the Card Atlas index failing to load when the app opened directly on a Deck, Improve, Test, or Library link, and stopped a failed load from being remembered for the rest of the visit.
- Improved: Renamed "card effect coverage" to "declared effect support" throughout Deck, Improve, Test, and the glossary, and defined it as a classification of printed clauses rather than a measurement of how games resolved.
- Improved: Generated the Season 2 legality ruleset (banned card, copy limit, banned pairs, and the effectless Unit profile) from the Card Atlas legality constraints at build time, with the April 2026 ruleset kept as hand-maintained history.
- Improved: Derived Atlas capability labels for the optimizer, countermeasures, interaction map, and coverage map from typed operation, target, modifier, keyword, and timing facets instead of regular expressions over card text.
- Improved: Projected reviewed Card Atlas clause dispositions into the effect manifest so reminder, structural, and no-effect text is never counted as engine work, and pinned the Atlas semantic artifact hash in the frozen snapshot so semantic repairs are visible as drift.
- Improved: Split each card's text into one clause per timing block so clauses match the Card Atlas program boundaries (1,134 clauses); a Deploy resolution is now compared only with the Deploy effect's program rather than every effect on the card.
- Improved: The simulation worker loads the Atlas IR bundle and interpreter only for duel jobs; the consistency run every page load performs no longer downloads or parses them.
- Improved: Matchup, comparison, and optimizer runs show "Preparing card data" while the Card Atlas index loads instead of appearing to ignore the click.
- Improved: The engine's board-changing primitives now report structured events (damage, rest, activate, destroy, return, recover, draw, discard, exile, Resource placement, AP change) that drive both measurement layers; log wording is no longer parsed for measurement.
- Improved: Script and data cache keys are derived from file content at build time (
npm run assets:keys, run byprebuild) instead of being edited by hand. - Improved: Effect telemetry is merged by one engine helper for both measurement channels; the app no longer re-aggregates it.
- Improved: Widened the Atlas shadow-trace vocabulary so resolver steps map onto Atlas operation codes (Shield-to-hand and trash-to-hand adds, destroy, return, deploy, rest, activate, pair, Resource placement, AP changes, look and reveal); token creation and cost payment are treated as not observable.
- Data: Refreshed the frozen Card Atlas snapshot to the current 0.3 semantic artifact, which preserves exact official punctuation and the repaired quantity, selector, and target facets, and, after the owner's 2026-09-02 rulings, carries
excludesSubjecton "other Units" selectors and binds count conditions to the cards being counted (Atlas repairs F32 and F33).
V50
Published / Added / Fixed / Improved
PublishedAddedFixedImproved- Published: Published the V47, V48, and V49 internal builds together with this release; the live site moved directly from V46 to V50.
- Added: Added constructibility, printing identities, reviewed relation evidence, governing rule links, and exact official FAQ questions and answers to Card Atlas study records.
- Added: Added visible Atlas trust metadata while keeping source verification, draft-schema, pending-paraphrase, and publication gates intact.
- Fixed: Migrated the simulator adapter from the retired Atlas 0.1 payload to the independently verified
0.2.0-draftproduction loader and made older payloads fail closed. - Fixed: Corrected semantic study data such as GD05-102 HP recovery, battle-area-to-hand movement, and its Action rule reference.
- Improved: Separated Atlas facts and evidence from simulator effect execution in both the runtime contract and Card Inspector copy.
- Improved: Kept pending Atlas paraphrases out of the player interface; official ruling text is displayed verbatim from the verified source snapshot.
V49
Added / Improved / Fixed
AddedImprovedFixed- Added: Added an Atlas-powered Card Inspector with official clause text, linked concepts, rulings references, effect-support status, and one-click editable Arena scenarios.
- Added: Added quantity-weighted clause coverage maps, a prioritized rules implementation backlog, deck interaction mapping, and clause-level replay telemetry.
- Added: Added Atlas-grounded optimizer candidate screening and colour-legal countermeasure discovery for individual matchups and the loaded meta field.
- Improved: Expanded card search with Atlas timings, zones, traits, rulings, and inferred capability terms while keeping deterministic search behavior.
- Improved: Kept optimizer recommendations simulation-led: Atlas semantics select candidates, while paired games and second-seed retests determine measured results.
- Improved: Extended Scenario Lab with an editable Trash zone and made Decision Inspector traces distinguish resolver activity from rulings and game impact.
- Fixed: Labelled Action-only Command setups as scenario drafts and explained that the current Arena cannot resolve them without a battle action window.
- Fixed: Restored the visible Card Inspector close symbol and contained completed optimizer tables at desktop and mobile widths.
V48
Added / Fixed / Improved
AddedFixedImproved- Added: Added a reviewed Card Atlas adapter and clause-level effect-support manifest for all 919 simulator card records.
- Added: Added canonical Gundnode and Pluma Unit tokens, including Gundnode's printed Breach 1 behavior.
- Fixed: Restored the missing recovery choice on GD05-102 Wings of Light and the missing Pilot-recovery choice on GD05-106 Mutual Attraction.
- Fixed: Updated the Season 2 legality attribution to the current official banned and restricted card notice.
- Improved: Made modal Command choices deterministic, target-aware, and visible in the Decision Inspector.
- Improved: Based card-effect coverage on executed clauses while keeping statistical confidence and benchmark validation separate.
V47
Fixed / Improved / Moved
FixedImprovedMoved- Fixed: Separated card-catalogue scope from executable effect-engine scope so EB01 cards no longer conflict with the GD01-GD05 simulation claim.
- Fixed: Hid pilot-loading and Current Meta reference controls when they do not apply to the selected matchup.
- Improved: Added source, date, freshness, and live/cache/fallback provenance beside the Home meta examples.
- Improved: Shortened completed Matchup results with a compact reliability strip and collapsed Replay and Evidence sections.
- Improved: Hardened mobile Card filters and Tuning deck summaries without closing controls that contain focus.
- Improved: Clarified the Card Library and deterministic Search Assistant hierarchy.
- Moved: Moved interface standards into a subordinate Developer Documentation disclosure.
V46
Consolidated / Moved / Preserved
ConsolidatedMovedPreserved- Consolidated: Replaced the separate first-visit dialog, workflow explainer, and command dashboard with one adaptive Home experience.
- Consolidated: Reduced the application shell to Home, Deck, Improve, Test, and Library while preserving every simulator capability.
- Moved: Moved Scenario Lab into Test as an advanced mode and combined Cards, Help, rules, glossary, methodology, and release notes under Library.
- Moved: Moved tournament field and freshness context into Test and kept card-effect coverage beside Deck and simulation results.
- Preserved: Preserved existing deck formats, deterministic simulation engines, result ownership, saved preferences, and legacy deep links.
- Preserved: Preserved Simple and Technical views, light and dark themes, the Card Assistant, optimizer validation, and Scenario Lab behavior.
V45
Fixed / Added / Improved
FixedAddedImproved- Fixed: Corrected the AI target policy that made the second seat repeatedly abandon base pressure for low-value unit trades; the 5,000-game legal mirror now records 55.0% first-seat wins.
- Fixed: Replaced residual first-paint placeholders and made the Meta Pulse tournament indicator switch between live, cached, and fallback provenance correctly.
- Added: Added a hard 52%-56% mirror-seat CI gate, engine and policy version stamps, per-check measurements, invariant monitoring, and V44-to-V45 benchmark history.
- Added: Added automatic tournament-result staleness warnings when the newest tracked event is more than three weeks old.
- Improved: Centralized engine-level objective limitations so one configuration controls Tuning warnings and confidence caps.
- Improved: Generated the in-app release history from this changelog during builds and documented earlier release-number amendments.
V44
Fixed / Added / Improved
FixedAddedImproved- Fixed: Redirected
/and/webappto the canonical/webapp/shell so the root can no longer drift into a separate experience. - Fixed: Added build-version stamps to HTML, response headers, and
/health, with route regression tests for redirects, nested SPA paths, HEAD responses, and user-agent independence. - Added: Baked a recorded AI decision into first paint and published a deterministic 400-game benchmark covering seed reproducibility, legal candidate ordering, paired-seat balance, and relative heuristic response. The current benchmark explicitly reports its observed first-seat bias as a limitation.
- Added: Added separate weighted-field and tournament provenance with explicit fresh, aging, stale, live, cached, and fallback states.
- Improved: Made simulation confidence depend on card-effect coverage as well as sample size.
- Improved: Split Tune deep links into
#tuning-manualand#tuning-optimizerwhile retaining old links as compatible aliases. - Improved: Normalized known GD05 archetype aliases and reconciled release numbering from V36 through V43.
V43
Added / Preserved
AddedPreserved- Added: Added a focused first-visit overview that explains the Build, Understand, Improve, and Test journey before entering the full workspace.
- Added: Added explicit Continue and Don't show again actions with keyboard focus containment, Escape handling, reduced-motion compatibility, and a mobile-safe layout.
- Preserved: Kept the existing Home workflow available after dismissal and left all simulator, deck, rules, and meta behavior unchanged.
V42
Improved / Fixed / Amendment - 2026-08-28
ImprovedFixedAmendment - 2026-08-28- Improved: Restored the target-card identity as crisp light and reverse SVG lockups, reducing the active logo payload from roughly 1 MB to under 5 KB.
- Improved: Moved the complete interface standards out of the player SPA and retained the standalone design-system page as the single reference.
- Fixed: Contained long Build status badges within the deck panel so legality text wraps cleanly instead of overlapping Draw Consistency.
- Fixed: Removed first-paint loading placeholders from Meta Pulse, recent results, and card browsing.
- Amendment - 2026-08-28: V42's public title was clarified during the V44 numbering reconciliation; the release scope and implementation are unchanged.
V41
Added / Updated
AddedUpdated- Added: Added a reproducible Egman GD05 primary-type meta updater and a direct Test vs. Current Meta navigation path.
- Added: Added a compact, validated meta snapshot to initial HTML so Meta Pulse and recent representatives are useful before JavaScript or live data responds.
- Updated: Replaced the previous GD05 top four with Strike Freedom Gundam (32.0%), Shining Gundam (22.4%), Nu Gundam (14.4%), and Gundam Barbatos 1st Form (8.0%) across 125 tracked Egman lists.
V40
Added
Added- Added: Added deterministic deck naming from colour identity, signature Mobile Suit or faction, and inferred playstyle, with a manual override in Deck Tools.
V39
Improved / Amendment - 2026-08-28 / Added
ImprovedAmendment - 2026-08-28Added- Improved: Separated Home from Build so the landing page presents only the product workflow and decision dashboard.
- Improved: Replaced the named-deck homepage status with a neutral legal-deck readiness message.
- Improved: Refreshed the GD05 Meta Pulse and Current Meta representatives from the August field and recent tournament lists.
- Amendment - 2026-08-28: This release was previously displayed under V38 before the V44 numbering reconciliation. The amendment preserves that published history without changing the current V39 scope.
- Added: Added a reproducible
npm run meta:updatemaintenance command for the current GD05 field snapshot.
V38
Improved / Amendment - 2026-08-28
ImprovedAmendment - 2026-08-28- Improved: Adopted the supplied target-card Gundam Match Lab lockup, with a matching mobile mark, favicon, dark-mode contrast plate, and updated design-system specimens.
- Amendment - 2026-08-28: V38 was previously displayed as "A focused Home" before the V44 numbering reconciliation. The target-card identity scope shown above is retained, and this note records the earlier public label.
V37
Fixed
Fixed- Fixed: Added a fail-open path for environments that deny the Worker Cache API while keeping security headers and uncached application responses intact.
V36
Fixed / Improved
FixedImproved- Fixed: Added safe shared caching for the application shell and extensionless SPA routes.
- Fixed: Renamed and protected Reset Deck so modified lists cannot be replaced without confirmation.
- Fixed: Removed redundant workspace chrome and suppressed empty matchup telemetry rows.
- Improved: Consolidated header utilities into Settings and retained one contextual action per workspace.
- Improved: Collapsed valid deck inputs and first-use tuning guidance to shorten repeated desktop and mobile workflows.
- Improved: Separated Card Browser and Card Assistant modes, with Browse Cards as the fresh-user default and compact mobile filters.
- Improved: Led matchup telemetry with the three largest observed differences and moved remaining evidence into a disclosure.
- Improved: Raised meaningful labels, help text, statuses, form values, and telemetry to readable minimum sizes.
- Improved: Replaced the mobile Meta Pulse carousel with a responsive grid and a focused Current Field signal on small screens.
Earlier releases V35 to V1
V35Isolated consistency, tuning, matchup, and arena result state so one workspace can no longer replace another workspace's conclusions.
V34Bumped cache-busting release tokens for shell styles, scripts, and worker assets to V34.
V33Internal packaging release with no player-facing changes. It was not published as a separate public application version.Internal
V32Rebuilt the responsive Match Lab identity around a sharper deck-comparison instrument mark with mobile-suit-inspired geometry.
V31Added a live Meta Pulse status rail for the leading tracked archetype, current-deck effect coverage, and tournament data source.
V30Reduced public navigation to the player-facing Build, Tune, Test, Cards, Advanced Scenario Lab, and Help destinations.
V29Simplified the primary navigation to Build, Tune, Test, and More.
V28Evolved the interface into a calm Industrial Instrument design language.
V27Added semantic design tokens and unified light and dark themes.
V26Added an in-app Changelog section under Help.
V25Routed exact static HTML paths through the Worker so cached app-shell responses receive CSP and common security headers.
V24Added CSP, HSTS, framing, MIME-sniffing, referrer, permissions, and opener protections.
V23Separated Manual Comparison and Deck Optimizer into distinct workflows.
V22Refreshed featured example decks from the current meta snapshot.
V21Added the guided product dashboard and grouped navigation.
V20Corrected production app-shell routing for browsers and automated clients.
V19Corrected replay card taps during playback.
V18Added the searchable Card Assistant with natural-language queries and starter actions.
V17Added Replay Card Inspector with active effects, modifiers, and interaction history.
V16Added the Automatic Deck Optimizer with measured candidate swaps and retesting.
V15Reorganized the simulator into tabbed workspaces.
V14Added Interactive Test Arena for manual play against the AI.
V13Added Effect Decision Inspector with requirements, targets, costs, and resolution steps.
V12Corrected the Gundam Throne Eins deploy effect and board presentation.
V11Added both players' hands to battle replays.
V10Ended games when direct Unit battle damage reaches a player with no Shields.
V9Expanded executable card-effect modelling across GD01 through GD05.
V8Added Burst shield probability references.
V7Audited GD01 through GD05 card rules and interactions.
V6Expanded the rules engine for GD05 mechanics.
V5Added paired deck comparison in the Deck Tuning Lab.
V4Added transparent card and effect modelling coverage.
V3Refined simulator visual design and responsive presentation.
V2Added tournament matchup simulation.
V1Published the initial Gundam TCG simulator.