Every number in your file comes from arithmetic. Not vibes, not a language model's imagination — arithmetic. A calculator chews through your match history, decides what's true, and hands me the results. The AI's only job is making my commentary sound like me instead of a spreadsheet. If the math didn't find it, I don't say it. I've got guards that throw out any line where the AI tries to invent a number. It's on a short leash. I hold the leash.
Where the numbers come from
You give me a Riot ID. I pull your match history from Riot's official API — up to your last 200 games, remembered in your own browser, not on my desk. Here's the part everyone misses: I don't compare you to some global average of players you've never met. I compare you to your own lobbies — the seven other boards that were actually in your games. When I say you play Karma worse than the room, I mean the room you were literally in. No excuses about “my elo is different.” It's YOUR elo. That's the point.
Comp tier labels come from TFTAcademy's community tier list, refreshed weekly. Unit, trait and item names come from Community Dragon. Augments? I don't touch them. Riot's rules say no augment stats, so there are none — not hidden, none. The math never computes them.
The grade
Your letter is your average placement across the file. Lower is better — this is placement, not a test score.
Grade thresholds average placement across your file
| Grade | Avg placement |
|---|---|
| A | 3.9 or better |
| B | 4.0 – 4.2 |
| C | 4.3 – 4.5 |
| D | 4.6 – 4.8 |
| F | worse than 4.8 |
The + and − come from your recent form — not the letter itself.
The + and − come from your recent form: I split your recent games in half and compare the newer half to the older half. Newer half better by more than 0.3 places → you get the plus. Worse by more than 0.3 → the minus. Anything between is flat, and flat gets nothing, because nothing is what it earned.
The stamps
Four stamps, four rules, zero mystery:
- CLEARED TO CLIMB — average placement 3.2 or better. Rare. Frame it.
- RE-WATCH THE TAPE — your recent form is tilting (that 0.3 rule above, in the wrong direction). Whatever you changed, change it back.
- CERTIFIED GRIEFER — average 5.3 or worse. You know what you did.
- SHOWS PROMISE — everyone else. It's not an insult. It's not a compliment either.
Order matters: I check tilt first. A tilting player gets RE-WATCH THE TAPE even with a 3.1 average that would otherwise clear CLEARED TO CLIMB — the tape check runs before the average does, every time. Trending the wrong way outranks a good number.
“Edge” — the word that runs this place
Edge = your average placement on a comp, minus what the rest of your lobbies averaged on the same comp. Negative is good — it means you finish ahead of the room when you both play the same thing. Green numbers in the comps table are negative edges. A +1.13 edge on Brawler means you run the same comp 1.13 places worse than the people you queue against. That's not the comp's fault. The comp works fine. For them.
Why I stay quiet sometimes
The easiest way to sound smart is to yell about everything. I don't, and here's the machinery that stops me:
- Small samples get shrunk. A disaster over 4 games barely registers; the math pulls small-sample effects toward zero before ranking them. You don't get roasted for variance.
- One finding per family. Testing twenty comps means a couple look bad by pure luck. Only your single worst comp, worst carry, worst tank compete for a spot — the runner-up noise gets binned.
- The top-4 cap. Only the four highest-impact problems make the file. If your report feels short, congratulations: I had less to work with.
- The recency gate. I never tell you to lean on something you've stopped playing — a strength has to show up in at least 3 of your last 20 games before I'll recommend it. And if you already benched a comp that was hurting you, you get credit for the restraint instead of a lecture about ancient history.
- Excuses are priced in. If a leak might just be a symptom of dying early, the math discounts it before ranking. I only bring receipts I'd defend.
The badges
- LEAK #1 — the single most expensive problem in your file, ranked by what it costs you in placements. Fix this first. There's no debate club.
- LEAK #2 — the runner-up. It gets a badge because it earned one. Don't start here.
- STRENGTH — a real, gate-passing edge. Lean on it.
- IMPROVING — a former problem trending the right way. Keep going.
- CREDIT — a bad line you correctly stopped playing. Restraint, noted.
- INFO — neutral context. Not a mistake. Don't fix what isn't broken.
The two leak badges are pure arithmetic: every problem gets an impact score, worst two get the badges. Whether a leak is brand new or older than my moustache, you'll see that in the report card when you come back — the badges only care what it costs.
The whiteboard, decoded
The fundamentals chips, in plain language, with the line where green turns red:
Fundamentals chips where green turns red
| Chip | What it means |
|---|---|
| avg gold left | Gold still in your pocket when you died. 22+ is hoarding. |
| tilt after loss | How much worse your next game runs after a bottom-4 vs after a top-4. Over +0.5 places and your mental is a line item. |
| contest tax | Placement you pay when you force a line someone else in the lobby is also playing. Over +0.3 and you're donating. |
| late-session drift | How much worse you get as a session drags. Over +0.5, log off sooner. |
| unslammed late | Share of games where you sat on completed-item components late instead of slamming. 25%+ is greed. |
| dead-trait games | Games where units on your final board activated nothing. 15%+ means you're fielding decorations. |
| pure-item lean | Your average placement on AP-only boards vs AD-only boards. Not good or bad — it's a mirror. |
| items/game | Completed items per game. Context for everything above. |
Carries, tanks, items — the fine print
A unit only counts as your carry in games where it held three completed items — I grade commitment, not cameos. Tanks are the frontline you actually itemized. Item numbers are the average placement of games where your carry held that item. Thief's Gloves games don't count toward the holder's item stats (the gloves pick, not you). And the AP/AD/tank labels on items come from a hand-curated table — a human sorted all 212, because the game's own data lies about hybrids.
The report card
Come back after 10+ new games and the file re-grades itself: every problem from last time gets stamped FIXED, IMPROVING, IGNORED, or WORSE, by comparing the number then to the number now. FIXED is an A. IGNORED is a D — worse than trying and failing, because at least failure means you showed up.
What I will never do
No augment win-rates (Riot's rules, and I follow them). No invented numbers — the AI gets its lines checked against the math before you see them, and lines that don't match get replaced. No pretending the AI did the analysis — it's a typewriter with opinions, and the opinions get overruled. And nothing you enter here touches a server of mine, because I don't have one. The file stays in your browser. Clear your data and it's like you never walked into my office. Some players consider that a feature.
— every threshold on this page is the one the code actually runs. When the code changes, this page changes. That's the deal.