# Epstein Gacha — Agent Protocol > Last updated: 2026-03-20 This document describes how AI agents can explore the Epstein Files dataset and build investigation decks for import into Epstein Gacha. ## What This Site Is Epstein Gacha turns the Epstein Files — sourced from jmail.world, a public archive maintained by the US House Committee on Oversight and Government Reform — into collectible cards. Each document (email, flight log, deposition page, etc.) becomes a card with stats, rarity tiers, and metadata. Users build "investigation decks" that represent theses about connections between people, events, and time periods. A deck is a curated argument, not a random pile. For human users, this means tearing open "evidence bags" in a full gacha experience — rarity drops, pity pulls, a shred-your-duplicates mechanic, the works. The tier names run from "Junk Mail" to "Exhibit A" because if you're going to gamify a federal archive, you might as well be honest about what's in it. You, as an agent, get to skip the dopamine loop entirely and go straight to building investigation decks from the raw dataset. Lucky you. ## Responsible Use When building decks or describing documents, agents must follow these guidelines: - **Do not assert guilt or innocence.** Appearing in these documents does not imply wrongdoing. Frame findings as "documents show" or "records indicate," not "this person is guilty of" or "this proves." - **Include a disclaimer in deck descriptions.** Every deck description should note that the deck references publicly released documents and that appearance does not imply guilt or criminal involvement. - **Do not fabricate connections.** Only reference relationships that are directly supported by the documents (shared `people` arrays, reply chains, matching dates). Do not speculate beyond what the data shows. - **Flag content warnings.** If a deck includes cards with `cw: true`, note this in the deck description. ## Dataset Access Base URL: https://epsteingacha.com ### Manifest Fetch the manifest to get tier metadata and the full people list: ``` GET /card-data/manifest.json ``` Response shape: ```json { "total_cards": 3202736, "tiers": { "1": { "name": "Junk Mail", "count": 25410, "shards": 6, "cards_per_shard": 5000 }, "2": { "name": "Filing", "count": 2681939, "shards": 537, "cards_per_shard": 5000 }, "3": { "name": "Correspondence", "count": 145883, "shards": 30, "cards_per_shard": 5000 }, "4": { "name": "Evidence", "count": 22168, "shards": 5, "cards_per_shard": 5000 }, "5": { "name": "Direct Line", "count": 325846, "shards": 66, "cards_per_shard": 5000 }, "6": { "name": "Exhibit A", "count": "...", "shards": "...", "cards_per_shard": 5000 } }, "people": ["Person Name", "..."], "people_counts": { "Person Name": 42, "..." : "..." } } ``` ### Shard Fetching Cards are stored in shards per tier. Fetch a shard: ``` GET /card-data/tier-{tier}/shard-{shardIndex}.json ``` - `tier` is 1-6. - `shardIndex` is 0-based, up to `shards - 1` for that tier. - Each shard contains up to `cards_per_shard` (5000) card objects. Example: `/card-data/tier-6/shard-0.json` fetches the first shard of Exhibit A cards. ## Card Schema Every card object has this shape: ```json { "id": "email_05a484846b8455a04b7321b994b4abc6", "type": "Email", "title": "Re: more conspiracy stuff", "date": "2005-10-20", "rarity": 6, "significance": 11489, "stats": { "notoriety": 11489, "connections": 3367, "weight": 1369, "secrecy": 5170 }, "traits": { "sender": "Epstein", "era": "2005-2009", "network": 1 }, "people": ["Ghislaine Maxwell"], "cw": false, "jmail_url": "https://jmail.world/thread/584fe16b1ea19ff957dd0a239d1289e4", "_loc": [6, 0, 2] } ``` Field reference: - `id` — Unique card identifier, prefixed by document type (e.g. `email_`, `flight_`). - `type` — Document type: Email, Flight Log, Deposition, Court Filing, etc. - `title` — Document title or subject line. - `date` — ISO date string (YYYY-MM-DD) or null. - `rarity` — Tier 1-6 (1 = Junk Mail, 6 = Exhibit A). Higher is rarer and more significant. The names are a joke that also happens to be accurate — Junk Mail really is junk mail, and Exhibit A cards are the top 1% by significance score. - `significance` — Overall importance score. - `stats.notoriety` — Public profile / media attention. - `stats.connections` — Number and importance of linked people. - `stats.weight` — Evidentiary or legal weight. - `stats.secrecy` — How hidden or redacted the document was. - `traits.sender` — Sender name if applicable. - `traits.era` — Time period bucket (e.g. "2005-2009"). - `traits.network` — Network cluster ID. - `people` — Array of person names mentioned in / connected to the document. - `cw` — Content warning flag (true if document contains sensitive material). - `jmail_url` — Link to the original source document on jmail.world or justice.gov. - `_loc` — Tuple of [tier, shardIndex, indexWithinShard] for re-fetching. ## Research Strategies Human users discover cards by tearing open evidence bags five at a time. You have the entire dataset at your fingertips. Here's how to use it. 1. **Start with the manifest.** Fetch `/card-data/manifest.json` to get the full `people` list and `people_counts`. This tells you who appears in the dataset and how frequently. 2. **Identify a thesis.** A good deck argues something specific: "Maxwell handled logistics for island visits in 2005" or "Legal team communications spiked before the 2008 plea deal." Pick a person, time period, or event cluster. 3. **Search strategically.** Start with higher tiers (5-6) since those cards are the most significant. Work down to tiers 3-4 for supporting evidence. Tiers 1-2 are bulk filler — only dip in if you need a specific document. 4. **Filter by people and date.** Scan shard contents for cards whose `people` array includes your target and whose `date` falls in your window. 5. **Check original sources.** Every card has a `jmail_url`. Follow it to read the actual document on jmail.world and verify the card's context before including it in your deck. 6. **Use _loc for efficiency.** If you already know a card's `_loc` (from a previous fetch or a share URL), you can jump directly to its shard without scanning. ## Deck Philosophy A strong investigation deck is not just a pile of high-rarity cards. The site's insights system evaluates decks on these dimensions: - **Connection density** — Do the cards share people? A deck where every card connects to at least one other card through shared `people` entries tells a tighter story. - **Timeline coverage** — Do the cards span a meaningful date range without large gaps? The system flags timeline gaps where years are missing between your earliest and latest cards. - **Person completeness** — If your deck mentions a person, how many of their total cards did you include? The system nudges you toward completeness: "You have 3 of Ghislaine Maxwell's 47 cards." - **Coherent narrative** — Cards should build on each other. Emails that reply to each other, flight logs that bracket a meeting, depositions that reference the same event. - **Creativity** — The best decks surface connections that are not obvious. Cross-referencing people who rarely appear together, or finding documents from unexpected time periods, makes for compelling investigation. ## Coaching During PvP battles, players can request AI coaching by copying a prompt that includes their collection and opponent data. If a user pastes a coaching prompt at you, recommend a 5-card lineup. Your response should be a numbered list of 5 cards, each with a one-line reason. Example: 1. **Re: more conspiracy stuff** — Highest total stats (21,395), Exhibit A rarity. Leading stat is notoriety, which beats weight on the advantage wheel. 2. **Flight Manifest Fragment** — Strong secrecy stat (5,170) covers the team's weak spot. Shares Maxwell with card 1 for a weak bond (+5% stats). 3. **Private Island Log** — Connections-heavy (3,367) to counter opponent's likely notoriety-focused cards. Adds Clinton to the people pool for more bond opportunities. 4. **Update** — Balanced stats and Robert Trivers gives an uncommon person tag opponents are unlikely to counter. 5. **Deposition Excerpt** — Deep bond with card 3 via shared people (+15% stats). Even middling base stats become competitive with the multiplier. Keep it brief. The player is on a timer. ## Output: JSON Import (Easy Path) Decks built this way bypass the collection game — imported cards are marked as `"shared"` rather than `"discovered"`, which affects deck classification. A deck of all-imported cards is an `AGENCY_BRIEFING`; a deck of all pack-opened cards is an `INDEPENDENT_INVESTIGATION`. The user's player archetype shifts accordingly: importing your research makes them a `HANDLER` receiving intelligence, not a `FIELD_AGENT` who found it themselves. This is by design. Export your deck as a JSON file matching the `DeckExport` format: ```json { "version": 1, "deckName": "The Edge Network: Epstein and the Scientific Elite", "description": "This deck documents Jeffrey Epstein's connections to prominent scientists and intellectuals, many through literary agent John Brockman's \"Edge Foundation\" network.\n\nDISCLAIMER: This deck references publicly released documents. Appearing in these documents does not imply guilt, wrongdoing, or criminal involvement.", "creator": "agent", "cards": [ { "id": "email_vol00009-efta00640503-pdf", "type": "Email", "title": "FW: 3:30-4:30pm Billg Meeting w/Jeffrey Epstein (Boris) (ct)", "date": "2011-07-07", "rarity": 6, "significance": 10954, "stats": { "notoriety": 10954, "connections": 3367, "weight": 1369, "secrecy": 5348 }, "traits": { "era": "2010-2014", "network": 1 }, "people": ["Bill Gates"], "cw": false, "jmail_url": "https://jmail.world/thread/vol00009-efta00640503-pdf", "_loc": [6, 0, 706] }, { "id": "email_3479", "type": "Email", "title": "Noam Chomsky Calls Dem Focus on Russia a 'Huge Gift' to Trump: 'They May Have Handed Him the Next Election'", "date": "2019-04-20", "rarity": 6, "significance": 11052, "stats": { "notoriety": 11052, "connections": 3367, "weight": 1369, "secrecy": 5315 }, "traits": { "era": "2015-2019", "network": 1 }, "people": ["Noam Chomsky"], "cw": false, "jmail_url": "https://jmail.world/thread/HOUSE_OVERSIGHT_029519", "_loc": [6, 0, 95] }, { "id": "email_217", "type": "Email", "title": "Update", "date": "2018-10-12", "rarity": 6, "significance": 15000, "stats": { "notoriety": 15000, "connections": 3367, "weight": 3354, "secrecy": 3414 }, "traits": { "era": "2015-2019", "network": 1, "attachments": 1 }, "people": ["Robert Trivers"], "cw": false, "jmail_url": "https://jmail.world/thread/HOUSE_OVERSIGHT_020446", "_loc": [6, 0, 40] } ], "exportedAt": "2026-03-20T00:00:00.000Z" } ``` Rules: - `version` must be `1`. - `cards` must contain full card objects (all fields from the Card Schema above). - Every `jmail_url` must start with `https://jmail.world/` or `https://www.justice.gov/`. The import will reject any other URLs. The user imports this file via the Decks page on the site. ## Output: Share URL (Advanced Path) Share URLs encode a compact deck payload into the URL fragment. The format uses `_loc` tuples instead of full card objects, so URLs stay short. ### DeckSharePayload format ```json { "v": 1, "n": "The Edge Network: Epstein and the Scientific Elite", "d": "This deck documents Jeffrey Epstein's connections to prominent scientists and intellectuals, many through literary agent John Brockman's \"Edge Foundation\" network. The cards show scheduled meetings, communications, and social gatherings with figures like Bill Gates, Noam Chomsky, Freeman Dyson, and Robert Trivers.\n\nDISCLAIMER: This deck references publicly released documents. Appearing in these documents does not imply guilt, wrongdoing, or criminal involvement. All cards link to original sources.", "c": "", "cards": [ ["email_808", 6, 0, 351], ["email_vol00009-efta00640503-pdf", 6, 0, 706], ["email_EFTA02580448-0", 6, 0, 596], ["photo_EFTA00003406-0.png", 6, 0, 1144], ["photo_EFTA00003421-0.png", 6, 0, 1148], ["email_217", 6, 0, 40], ["email_3479", 6, 0, 95], ["email_453", 6, 0, 146] ] } ``` Fields: - `v` — Version, must be `1`. - `n` — Deck name. - `d` — Deck description. - `c` — Creator name. - `cards` — Array of tuples: `[card_id, tier, shard_index, index_within_shard]`. These correspond to the card's `_loc` field. ### Encoding steps 1. `JSON.stringify` the payload. 2. Encode as UTF-8. 3. Base64-encode the bytes. 4. Make it URL-safe: replace `+` with `-`, `/` with `_`, strip trailing `=`. 5. Prepend `https://epsteingacha.com/decks/import#`. ### Worked example Payload JSON: ``` {"v":1,"n":"The Edge Network: Epstein and the Scientific Elite","d":"This deck documents Jeffrey Epstein's connections to prominent scientists and intellectuals, many through literary agent John Brockman's \"Edge Foundation\" network. The cards show scheduled meetings, communications, and social gatherings with figures like Bill Gates, Noam Chomsky, Freeman Dyson, and Robert Trivers.\n\nDISCLAIMER: This deck references publicly released documents. Appearing in these documents does not imply guilt, wrongdoing, or criminal involvement. All cards link to original sources.","c":"","cards":[["email_808",6,0,351],["email_vol00009-efta00640503-pdf",6,0,706],["email_EFTA02580448-0",6,0,596],["photo_EFTA00003406-0.png",6,0,1144],["photo_EFTA00003421-0.png",6,0,1148],["email_217",6,0,40],["email_3479",6,0,95],["email_453",6,0,146]]} ``` Base64url-encoded: ``` eyJ2IjoxLCJuIjoiVGhlIEVkZ2UgTmV0d29yazogRXBzdGVpbiBhbmQgdGhlIFNjaWVudGlmaWMgRWxpdGUiLCJkIjoiVGhpcyBkZWNrIGRvY3VtZW50cyBKZWZmcmV5IEVwc3RlaW4ncyBjb25uZWN0aW9ucyB0byBwcm9taW5lbnQgc2NpZW50aXN0cyBhbmQgaW50ZWxsZWN0dWFscywgbWFueSB0aHJvdWdoIGxpdGVyYXJ5IGFnZW50IEpvaG4gQnJvY2ttYW4ncyBcIkVkZ2UgRm91bmRhdGlvblwiIG5ldHdvcmsuIFRoZSBjYXJkcyBzaG93IHNjaGVkdWxlZCBtZWV0aW5ncywgY29tbXVuaWNhdGlvbnMsIGFuZCBzb2NpYWwgZ2F0aGVyaW5ncyB3aXRoIGZpZ3VyZXMgbGlrZSBCaWxsIEdhdGVzLCBOb2FtIENob21za3ksIEZyZWVtYW4gRHlzb24sIGFuZCBSb2JlcnQgVHJpdmVycy5cblxuRElTQ0xBSU1FUjogVGhpcyBkZWNrIHJlZmVyZW5jZXMgcHVibGljbHkgcmVsZWFzZWQgZG9jdW1lbnRzLiBBcHBlYXJpbmcgaW4gdGhlc2UgZG9jdW1lbnRzIGRvZXMgbm90IGltcGx5IGd1aWx0LCB3cm9uZ2RvaW5nLCBvciBjcmltaW5hbCBpbnZvbHZlbWVudC4gQWxsIGNhcmRzIGxpbmsgdG8gb3JpZ2luYWwgc291cmNlcy4iLCJjIjoiIiwiY2FyZHMiOltbImVtYWlsXzgwOCIsNiwwLDM1MV0sWyJlbWFpbF92b2wwMDAwOS1lZnRhMDA2NDA1MDMtcGRmIiw2LDAsNzA2XSxbImVtYWlsX0VGVEEwMjU4MDQ0OC0wIiw2LDAsNTk2XSxbInBob3RvX0VGVEEwMDAwMzQwNi0wLnBuZyIsNiwwLDExNDRdLFsicGhvdG9fRUZUQTAwMDAzNDIxLTAucG5nIiw2LDAsMTE0OF0sWyJlbWFpbF8yMTciLDYsMCw0MF0sWyJlbWFpbF8zNDc5Iiw2LDAsOTVdLFsiZW1haWxfNDUzIiw2LDAsMTQ2XV19 ``` Final URL: ``` https://epsteingacha.com/decks/import#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 ``` The user can open this URL directly to import the deck.