MCOC Prestige Tools

A free, roster-aware prestige optimisation tool for Marvel Contest of Champions. Deployed at mcoc.help. Architecture overview in architecture-v5.md.

Repo structure

prestige-tools/
├── apps/
│   └── web/                Next.js 15 app (static export → Cloudflare Pages)
│       ├── app/                Route handlers
│       ├── components/         Client components (picker, table, recs view)
│       └── lib/                Format helpers, roster storage, data loader
├── packages/
│   └── engine/             Pure-TS @prestige-tools/engine
│       ├── src/
│       │   ├── types.ts        Champion / Roster / Move / CostGate
│       │   ├── bhr.ts          BHR computation with per-rank sig curves
│       │   ├── prestige.ts     Top-30 averaging
│       │   ├── optimise.ts     Atomic moves enumeration & ranking
│       │   ├── ceiling.ts      Ceiling view (long-term planning lens)
│       │   └── costs.ts        Four-axis cost gating
│       └── __tests__/      Vitest suite — reproduces §16 verified data
├── data/
│   ├── formulas/
│   │   └── multipliers.json    Rank × ascension × sig-curve constants
│   ├── champions/
│   │   └── seed.json           254 champions (full 7-star roster)
│   └── _verified/
│       └── known-rosters.json  Regression target (Dave's verified roster)
└── scripts/
    ├── seed-from-mcochub.ts        Ingestion stub (full impl Phase 2)
    └── scrape-fandom-portraits.ts  Populate portraitUrl from Fandom CDN

Prerequisites

• Node.js ≥ 20
• pnpm ≥ 9 (npm install -g pnpm if you don't have it)

Get running

pnpm install
pnpm -F @prestige-tools/engine build  # required: builds the engine to dist/
pnpm test                              # engine Vitest suite — reproduces §16
pnpm dev                               # Next.js on localhost:3000

The engine builds to packages/engine/dist/. The Next.js app consumes the built output rather than source — re-run the engine build after any engine code change. (Phase 1.x will improve this with a watch script.)

Visit http://localhost:3000/roster/ to add champions, then / for the recommendations.

Populating portraits

The seed ships with portraitUrl: null on every champion — the champions page renders class-icon placeholders until you populate the portrait URLs:

pnpm scrape-portraits

This walks every champion, hits the Fandom wiki API, extracts the headshot URL, writes it back to data/champions/seed.json. ~5 minutes for the full 254 (rate-limited to 1 request/sec).

Idempotent — re-running only fetches champions still missing a URL. Use --force to refresh all. Use --only "Name1,Name2" to target specific champions. Failures are logged to scripts/scrape-failures.json for manual review.

See architecture-v5.md §17 for the portrait sourcing decision.

Roster sharing

New in iter6. Users can generate a short link (mcoc.help/r/?id=abc12345) showing their roster as view-only — useful for alliance-war / AQ planning. Backed by Cloudflare Pages Functions and KV storage; entries auto-expire after 6 months. See architecture-v5.md §22 for the design rationale and DEPLOY.md for the KV namespace setup steps.

The share feature requires the ROSTERS KV binding to work. Without it (local dev with pnpm dev, or missing dashboard binding in production), the rest of the app works fine — only share creation and the /r/ page fail.

What works in this iteration

• Roster picker: typeahead search across all 254 7-star champions, inline rank/sig/ascension entry. Ascension dropdown auto-disables for non-ascendable champions. Adding the same champion twice overwrites the existing state.
• Bulk paste import: paste your whole roster as text, the parser understands many formats (Lizard R5 sig 200 A2, maestro r4 200 a2, Pavitr 4/200/1, etc.). Shows matched / ambiguous / unmatched counts before committing.
• Roster table: every owned champion with current BHR, ceiling, headroom Δ, prestige-impact-if-maxed, in-top-30 indicator, plus a summary bar (champion count, top-30 prestige, cutoff BHR, highest BHR).
• Recommendations view (atomic): top-10 ranked moves with cost-gate badges (rank-cats / sig-stones / ascension), partitioned into "proceed" and "deferred — ascend first" per the v5 deferral rule.
• Recommendations view (ceiling): top-10 long-term plays sorted by prestige delta if fully developed. The Phase 0 "Blue Marvel +449" insight shows up automatically when you've got him in your roster.
• Persistence: roster lives in localStorage. Nothing leaves your device.
• Champion list + detail pages (statically generated for all 254 champs).
• Variant D Hybrid styling: editorial typography, Marvel red accents, burst rendering reserved for the #1 recommendation card.
• Production-ready deploy config: DEPLOY.md walks through the Cloudflare Pages setup; .nvmrc pins Node 20.

What's NOT here yet

• Sortable / filterable roster table: columns are defined but click-to- sort and filter chips (class, ascendable, in-top-30, cost-gate) are Phase 1.x work.
• URL-hash sharing: encodeRosterToHash / decodeRosterFromHash are implemented in lib/roster-storage.ts but not wired to the UI.
• Champion detail page synergies / immunities: deferred until the metadata pipeline lands.
• Drift detection / community correction flow: Phase 2.
• Class assignments may have minor errors: 254 champions classified by rule; the rare unusual variant (Sam Wilson Cap, scope of Mystic vs Skill for some characters) might need manual correction post-launch.

Data sources

• MCOCHUB (InsaneSkull) — primary BHR reference: https://mcochub.insaneskull.com
• mcoc.gg (BrutalDX) — secondary cross-check: https://mcoc.gg
• Fandom wiki — champion metadata (CC-BY-SA): https://marvel-contestofchampions.fandom.com

See architecture-v5.md §17 for the full sourcing chain.

Engine accuracy

End-to-end check against Dave's verified §16 roster:

Champions:        30 / 30 within ±30 BHR
Exact matches:    17 / 30
Aggregate Δ:      −140 BHR across 30 champions
Predicted prestige: 36,115
Observed prestige:  36,120 (off by 5)

The accuracy is bounded by in-game BHR rounding (every value is rounded to the nearest 10), so per-champion ±10 is the noise floor. The R3 multiplier (0.6906) was independently confirmed against Onslaught, Silver Surfer, and QuickSilver — verified during Phase 0 from auntm.ai data.