AI deck analysis

The AI analyzer (feature 06) gives you a structured second opinion on any valid 50-card deck: detected strategy, archetype, strengths, 3–5 mechanical suggestions, and a 0–100 coherence score broken down on four axes (consistency, power level, versatility, anti-meta).

How to use it

1. Open /deck-builder/$deckId and finalize the deck (exactly 50 cards, legality OK).
2. Click Analyser avec l'IA in the right-panel footer.
3. A slide-over sheet opens with the analysis. Click Appliquer on any suggestion to add / remove / replace cards in the deck.
4. Use the FR / JP chips to translate on-demand (cached after the first translation).
5. Voir l'historique at the bottom of the sheet opens /deck-builder/$deckId/analyses — every past analysis for this deck, freshest first.

Plan gating

If you're on Free, the analyze button opens an upgrade prompt instead of running the analysis (we want you to feel the Nakama value).

Models + cost cap

• Free tier: Claude Haiku 4.5 (claude-haiku-4-5-20251022).
• Nakama+: Claude Sonnet 4.6 (claude-sonnet-4-6-20260213).
• Translation: Haiku 4.5, lazy, cached.

The platform-wide monthly budget cap defaults to 15 000 cents (150 €). When usage reaches 80 % of the cap the service flips to degraded; at 100 % it flips to down and refuses new analyses until the 1st of the next month (or sooner if an admin bumps the cap from /admin/ai-monitor).

When the service is down, the sheet shows a friendly explanation and a "Notify me when it's back" opt-in (userProfiles.notifyOnAiBack). The cron that resets the state at the start of each month also emails opted-in users.

Disclaimer

These suggestions are statistical hints. Player skill and reading the opponent do not enter into this analysis.

This wording is locked into both the system prompt sent to Claude and the footer of the sheet. The analyzer never quotes Bandai's verbatim card-effect text — only our own effectSummary strings (encoded by admins under feature 04).

Anti-abuse (Free tier)

To keep the Free tier from being farmed:

1. Email verification (OTP) is enforced (feature 02).
2. Signups with a disposable-email domain (disposable-email-domains package) are flagged — Free analyses are refused for those accounts.
3. Each signup IP is checked against proxycheck.io; confirmed VPN / proxy / Tor IPs flag the user. Free analyses are refused while the flag is uncleared. The flag clears automatically on first paid Nakama subscription.
4. Browser-fingerprint hash + IP rate-limit (3 signups / 24 h) reduce multi-account abuse from the same machine.

None of these checks affect Nakama+ paying users, who keep unlimited access regardless.