7 min read · Published
A field-by-field walkthrough of the JSON LoL2LLM hands to ChatGPT or Claude: how the export checkboxes map to JSON keys, a real output sample, and what to watch for when you paste it.

CHAMPION LENS
Orianna
Nocturne
Ezreal
Nautilus
FioraLoL2LLM pulls the fields worth analyzing out of the Riot API match data and writes them as JSON with shortened key names. One thing to keep straight: the checkbox labels on the export screen ("Basic Metadata", "Combat", and so on) are not the key names that appear in the JSON (game, me, and so on). This guide maps one to the other, explains what each key holds, and flags what to watch for when you hand it to an AI.
Here is how each checkbox maps to the JSON. Every box is checked by default.
| Checkbox | JSON key | What's inside |
|---|---|---|
| Match Context › Basic Metadata | game | Mode, duration, patch, result, queue ID |
| Match Context › Team Macro | teams | Both teams' bans, objectives taken, total kills, total gold |
| My Performance › Combat | me.combat | Kills, deaths, assists, KDA, damage to champions, damage taken, and more |
| My Performance › Economy | me.economy | CS, gold earned, level |
| My Performance › Vision | me.vision | Vision score, wards placed and killed, control wards bought |
| My Performance › Loadout | me.loadout | Items, summoner spells, runes |
| Participants › Allied Team | allies | A summary of your four teammates |
| Participants › Enemy Team | enemies | A summary of all five enemies |
| Participants › Lane Opponent | lane_opponent | Detailed stats for the enemy in your role |
The real output looks like this. It's a live match (KR_8270926232) exported with every box checked and anonymization on, taking the jungler as "me". Some fields are folded onto one line for readability, and "…" marks omitted parts.
{
"game": {
"mode": "CLASSIC",
"dur_s": 1530,
"patch": "16.12",
"result": "Loss",
"queue": 420
},
"teams": {
"ally": {
"bans": [235, 555, 34, 126, 143],
"obj": { "dragon": 1, "baron": 0, "tower": 3, "herald": 0 },
"total_k": 26,
"total_g": 52862
},
"enemy": { … }
},
"me": {
"name": "Player (Me)",
"champ": "Vi",
"role": "JUNGLE",
"combat": {
"k": 11, "d": 4, "a": 7, "kda": "4.50",
"dmg_champs": 22141, "dmg_taken": 32119,
"largest_spree": 8, "multi_kill": 2, "cc_s": 34
},
"economy": { "cs_total": 168, "cs_min": 17, "cs_jg": 151, "gold": 12988, "lvl": 14 },
"vision": { "score": 55, "wards_placed": 10, "wards_killed": 13, "ctrl_bought": 10 },
"loadout": {
"items": [3155, 6610, 2055, 3026, 3111, 3071, 3364],
"spells": ["Flash", "Smite"],
"runes": { "primary_style": "Precision", "primary_picks": [8010, 9111, 9104, 8017], … }
}
},
"allies": [
{ "name": "Ally (TOP)", "champ": "Ambessa", "role": "TOP", "k": 2, "d": 6, "a": 6,
"kda": "1.33", "dmg_champs": 22773, "gold": 8668 },
…
],
"enemies": [ … ],
"lane_opponent": {
"name": "Enemy (JUNGLE)", "champ": "Naafiri", "role": "JUNGLE",
"k": 6, "d": 5, "a": 7, "kda": "2.60",
"dmg_champs": 20016, "dmg_taken": 23454, "cs": 195,
"gold": 11617, "lvl": 15, "vision": 41,
"items": [6699, 3071, 6695, 2019, 2422, 3364]
}
}game holds the mode (mode; Summoner's Rift games are CLASSIC), game length in seconds (dur_s), patch (patch), result (result), and queue ID (queue; 420 is Ranked Solo/Duo). The patch helps keep the AI from judging your game against a different patch's meta. Game length is the denominator for CS/min and every other per-minute figure: a 6/0/2 in a 20-minute stomp is a different game from a 6/0/2 in 40 minutes of teamfighting.
teams has an ally and an enemy block, each with bans (bans), dragons, barons, towers, and heralds taken (obj), total kills (total_k), and total gold (total_g). Bans are champion IDs, not names, and -1 means no ban. This is the team-level frame your individual stats sit inside; kill participation, for example, is your kills plus assists divided by total_k.
me holds your name, champion, and role (name, champ, role) plus whichever sub-categories you checked: combat, economy, vision, loadout.
combat: kills, deaths, assists (k, d, a), KDA (Perfect with zero deaths), damage to champions (dmg_champs), damage taken (dmg_taken), largest killing spree (largest_spree), largest multikill (multi_kill), and a crowd-control figure (cc_s, Riot's timeCCingOthers). Damage taken is the clue to how much a tank or support soaked.
economy: total CS (cs_total) split into minions (cs_min) and jungle monsters (cs_jg), gold earned (gold), and level (lvl). Note that cs_min is the minion count, not CS per minute — make sure the AI doesn't mix them up. CS/min is cs_total divided by dur_s / 60.
vision: vision score (score), wards placed (wards_placed) and killed (wards_killed), and control wards bought (ctrl_bought). This is what you lean on when asking the AI to judge a support's vision work.
loadout: items (items), summoner spells (spells), and runes (runes). Spells and rune trees come as names, but items and individual runes come as numeric IDs. Items cover every filled inventory slot, so components and the trinket show up alongside finished items.
The AI won't always translate item IDs into the right names. When you ask for a build review, add the names of your main items to the prompt so nothing gets misread.
allies holds your four teammates and enemies all five opponents. Each entry is lighter than me: name, champion, role, kills, deaths, assists, KDA, damage to champions (dmg_champs), and gold. Without it, though, an AI cannot tell whether your 6/2/8 is good or bad — it has no peer comparison. Damage share, for example, is your dmg_champs divided by the sum of dmg_champs for you and your four teammates.
lane_opponent holds the enemy who played your role: kills, deaths, assists, and KDA, plus damage to champions, damage taken, total CS (cs), gold, level, vision score (vision), and items. A prompt like "compute CS, damage, and death deltas vs lane_opponent and judge who came out ahead" gives you a metric-grounded verdict instead of a vibes-based one. Keep in mind these are end-of-game totals, so they can't show how the lane phase itself went.
The opponent is matched by role, so supports get the enemy support. In games without assigned roles (ARAM, for instance), lane_opponent is not included. Between two supports, comparing vision score and damage taken shows the gap in vision work and in how much each soaked on the front line.
"Hide usernames" is on by default. While it's on, names become role labels: you are Player (Me), teammates look like Ally (TOP), and enemies like Enemy (JUNGLE). Turn it off and each player's Riot ID game name appears instead. Only the names change; every number the analysis uses stays the same. Leave it on when you want to discuss the game publicly (Reddit, Discord, X) without dragging other players' names along.
Even with every box checked, one game is about 3,900 characters (measured on the match above), so size isn't the problem. The problem is focus: numbers unrelated to your question tend to pull the answer toward them. Curated sets by question:
You vs your lane opponent: Basic Metadata + Combat + Economy + Lane Opponent
Teamfight / macro analysis: Team Macro + Combat + Allied Team + Enemy Team
Build review: Basic Metadata + Loadout + Combat + Lane Opponent (which includes the opponent's items)
Once you understand what each field carries, you can match the export to the question. Specific input, specific question, specific answer — that's how you turn an LLM into a useful coach instead of a generic feedback machine.
Found this article useful? Leave a like, and share your thoughts or questions in the comments.
AhriRead the post-game scoreboard the way a coach would. Role-aware benchmarks, what each number actually measures, and how to combine them before handing the JSON to an AI.
Lee SinWhat APM (actions per minute) means in League of Legends and why a high APM doesn't make you a better player. LoL has no official APM, so we measured ability casts per minute (Q/W/E/R) from Riot match data by rank, role and champion.
JhinWhen to use which of LoL2LLM's five personas, plus the one-line additions that turn a generic answer into a sharp one.