Sisyphus

01.1The joke that became the product

A friend said he was going to start a League of Legends ranked grind. I was bored and thought: "what if I build something to track him and potentially make fun of him :p".

That was basically the first product brief. The earliest version only needed to notice the climb and give the group something to react to.

Then more people in the squad were playing ranked and asked to be tracked too. The project stopped being about one friend and became a shared record of the group.

A shared record

STARTED AS A JOKEA friend said he was going to grind ranked.One friend → the squad

Sisyphus-BotAPPToday at 9:41 PM

Juno#0000

✅ Victory · Ahri (MID) · 30-25

Ranked Solo/Duo · 35m 42s
GOLD IV — 42 LP +18 LP

Core Line8/3/11 KDA · 6.33 ratio
212 CS · 7.1/min
Level 17

OutputDamage 24,812 · 31% share
Gold 13,420 · 24% share
Kill participation 63%

Map WorkVision score 27
Wards 12
Cleared 4

StoryClean win: Juno gained +18 LP on Ahri with 8/3/11.

Spotlight• Juno moved the boulder +18 LP.
• Juno joined 63% of team kills.
• Juno lit the map with 27 vision score.

ContextTeam kills 30 · Enemy kills 25
Damage share 31% · Gold share 24%

ItemsShadowflame · Blackfire Torch · Swiftmarch · Cosmic Drive · Oblivion Orb · Needlessly Large Rod

Team and full scoreboard views are available below

📊 Overview🔵 Blue Team🔴 Red Team🏅 Full LeaderboardRemember

01.2The real value was the shared story

Without Sisyphus, ranked progress is split across match history, Discord messages and memory. You can look up a game, but the group loses the surrounding story: who was grinding, how LP moved, what people predicted, what everyone argued about afterwards and which matches became memorable.

That is why I think of Sisyphus as a chronicle rather than a stat lookup. A game becomes a recap. The recap becomes a conversation. The result affects fictional markets and records. The history accumulates into something the squad can revisit.

01.3Who it is for

Sisyphus is deliberately shaped around one private Discord squad. Anyone in the group can opt in to tracking, and the main product surface is Discord because that is where the group is already talking while games happen.

The dashboard is secondary. It gives the same history a richer analytical view, but the product would feel wrong if the group had to leave Discord for every interaction.

02.1Discord first, dashboard second

Discord is where the social moment happens, so recaps, queue state, live-room updates, predictions and reactions belong there. The website is better for long-lived exploration: LP journeys, activity, markets, records and the Boulder Archive.

This division kept the product from becoming a dashboard that happens to have a bot. The bot is the product surface; the dashboard is the deeper history.

Discord first

Sisyphus-BotAPPToday at 8:22 PM

Kairo#0000

Today's Stats

Rank PLATINUM IV — 42 LP

Today✅ 2 ❌ 1 ➖ 0
+18 LP

Ranked Solo/Duo only

📊 Today🏆 All-time

THE BOULDER CHRONICLERANKED RECORD

THE BOULDER CHRONICLE / THE SQUAD LEDGER

The Squad Ledger

Tracked players13

Games recorded1,507as of 13 November 2026

02.2Why JSON stayed the source of truth

The bot still uses JSON as its primary state store. For one small private guild, that has been a deliberate fit-for-scale choice rather than something I am trying to hide.

A heavier database would add operational complexity without solving the main problems I have today. The migration trigger would be a real change in the product: more users, multiple guilds, significantly more history or enough concurrency that whole-file state updates become a limitation.

The rule has been to scale the architecture when the product needs it, not because a hypothetical future version might.

02.3Riot and OP.GG are complementary, not interchangeable

The hardest technical theme in Sisyphus is external truth. Riot is useful for live presence, account information and LP-related state. OP.GG provides ranked history and match-detail information that fills different gaps.

The problem is timing. A match can finish before every source reflects the new state. If the bot posts the first value it sees, LP can be wrong, an old match can appear as new, or a downtime recovery can flood the channel with stale recaps.

So the ingestion path is built around reconciliation: bounded first-run backfill, recap age caps, duplicate guards and delayed LP re-fetch after match completion.

02.4Private community data stays private

The raw bot state contains identifiers and operational information that should never become public just because the code is on GitHub. Historical state even included PUUIDs, Discord IDs, message/channel IDs and other private records.

The dashboard therefore consumes a sanitized, allowlisted export rather than raw bot state. Member identifiers are replaced by opaque keys, and dashboard access is scoped through Discord OAuth and guild membership.

03.1From a tracked player to a recap

The bot polls tracked players for active Ranked Solo/Duo games. Riot live state helps identify active games, while OP.GG supplies recent ranked history and detail. The system identifies unseen matches, rejects stale recap candidates and re-fetches LP after a match to reduce timing errors.

Remakes under two minutes are canonicalized as draws. Related fictional markets are voided and refunded rather than settled as a normal win or loss. Otherwise the outcome updates history, records and market state before the recap is posted.

The recap itself is intentionally dense but social: result, rank/LP transition, KDA, CS/min, damage, gold, vision, items, team context and drill-down views. Friends use it both as a summary and as evidence in the inevitable "that loss was not my fault" argument.

One outcome, one recap

01 · MATCH FINAL

Juno · Ahri · Win +18 LP

02 · RECAP POSTED

Sisyphus-BotAPPToday at 9:41 PM

Juno#0000

✅ Victory · Ahri (MID) · 30-25

Ranked Solo/Duo · 35m 42s
GOLD IV — 42 LP +18 LP

Core Line8/3/11 KDA · 6.33 ratio
212 CS · 7.1/min
Level 17

OutputDamage 24,812 · 31% share
Gold 13,420 · 24% share
Kill participation 63%

Map WorkVision score 27
Wards 12
Cleared 4

StoryClean win: Juno gained +18 LP on Ahri with 8/3/11.

Spotlight• Juno moved the boulder +18 LP.
• Juno joined 63% of team kills.
• Juno lit the map with 27 vision score.

ContextTeam kills 30 · Enemy kills 25
Damage share 31% · Gold share 24%

ItemsShadowflame · Blackfire Torch · Swiftmarch · Cosmic Drive · Oblivion Orb · Needlessly Large Rod

Team and full scoreboard views are available below

📊 Overview🔵 Blue Team🔴 Red Team🏅 Full LeaderboardRemember

03 · FULL SCOREBOARD

Sisyphus-BotAPPToday at 9:42 PM

Juno#0000

🏅 Full Scoreboard · 35m 42s

🔵 Blue Team ✅ VictoryAhri 8/3/11 CS:212

🔴 Red Team ❌ DefeatRenekton 3/7/10 CS:254

📊 Overview🔵 Blue Team🔴 Red Team🏅 Full LeaderboardRemember

03.2Queue Beacon to Live Game Room

One of the more product-like state transitions is the Queue Beacon. Instead of posting a new message for every phase, one Discord message can become the Live Game Room and update in place.

The room tracks the game, teams, watcher count, voice presence, streaming state and a final snapshot. Watchers are aggregated from the relevant voice channel and the tracked player is not counted as their own watcher.

When the game disappears, the room is finalized and the historical record can preserve things such as peak watchers and stream time.

One message changes in place

Sisyphus-BotAPPToday at 9:41 PM

Juno is pushing the boulder.

Queue Beacon

Juno is pushing the boulder on Renekton.

QueueRanked Solo/Duo

StreamNo stream yet

Watching0 watching

Watcher IntentNo watchers yet

No scouting. No spoilers. Just the room around the game.

Sisyphus-BotAPPToday at 9:41 PM

Juno is pushing the boulder.

Queue Beacon

Juno is pushing the boulder on Renekton.

QueueRanked Solo/Duo

StreamNo stream yet

Watching0 watching

Watcher IntentNo watchers yet

No scouting. No spoilers. Just the room around the game.

Sisyphus-BotAPPToday at 9:41 PM

Juno is pushing the boulder.

Queue Beacon

Juno is pushing the boulder on Renekton.

QueueRanked Solo/Duo

StreamStreaming

Watching2 watching

Watcher Intent🫡 1 · 💀 0 · 👀 1

No scouting. No spoilers. Just the room around the game.

Sisyphus-BotAPPToday at 9:41 PM

Juno is pushing the boulder.

Queue Beacon

Juno is pushing the boulder on Renekton.

QueueRanked Solo/Duo

StreamStreaming

Watching2 watching

Watcher Intent🫡 1 · 💀 0 · 👀 1

No scouting. No spoilers. Just the room around the game.

Finalized

03.3Fictional markets and community state

The betting layer uses fictional points only. It exists because lightweight stakes make the squad pay attention to games they are already watching, not because I wanted to build a finance product.

Markets have explicit open, locked, settled and void/refund states. That state has to agree with the same canonical game outcome used by recaps and history, otherwise the social layer becomes inconsistent.

Profiles, records, goals, rivalries, memories and recurring recaps use the same accumulated history to turn repeated ranked games into a longer-running community record.

Explicit market states

Sisyphus-BotAPPToday at 10:17 PM

🔔 Market Resolution Notification!

✅ Market Settled — WIN

MarketNext Ranked Match — Wren

Tracked PlayerWren#0000

Total Staked450 pts

Winners1

ReasonMATCH_RESOLVED

THE BOULDER CHRONICLERANKED RECORD

THE BOULDER CHRONICLE / POINTS MARKETS

Points Markets

WalletsJuno · 650 pts; Kairo · 420 pts

MarketsOpen · locked · settled · void/refund

Market historyJuno · Ranked Solo/Duo SettledKairo · Ranked Solo/Duo Void / refunded

03.4Dashboard and sanitized export

The Boulder Chronicle dashboard is a separate FastAPI and React/Vite application. It does not read the raw bot state directly. A sanitized export defines the fields the dashboard is allowed to see and rejects forbidden keys.

The dashboard then turns that safe export into overview, player journeys, markets and community/archive views. Discord OAuth plus guild-membership checks keep the real dashboard private to the group.

03.5Operations for a passion project

The bot runs on my Mac. The dashboard and uptime tooling run on Oracle infrastructure that I also administer. That split is pragmatic, but it also means I own the operational edges on both sides.

The project includes process locking, health snapshots, backup validation and rotation, supervisor restart/backoff, deployment coordination, status publishing and log monitoring. None of that makes Sisyphus "enterprise software"; it is simply what was needed to keep a small service I care about running reliably.

03.6Refactoring and hardening as the project grew

The original bot became harder to maintain as features accumulated, so I split the monolith into modules. The motivation was straightforward: make new features easier to add without turning every change into a risk to unrelated behaviour.

A separate hardening pass replaced an earlier dynamic evaluation path with a strict fail-closed AST allowlist for OP.GG-derived expressions. I treat that as security hardening, not as a story about a known exploit.

Mistral also exists in a very bounded place: summarizing Git commits for release notes with a local fallback and approval gate. It does not drive match tracking, markets or game decisions.

04.1Live since May

Sisyphus went live on May 19, 2026. It has 13 tracked players and 1,507 games recorded as of 13 November 2026.

I use "games in view" deliberately rather than calling that a lifetime count. The live number is useful context, but the stronger outcome is that the system is still running and the squad still uses it.

04.2What the squad actually values

The features that matter are not always the most technically complicated ones. Friends use the win/loss and LP view, fictional betting, recaps and the dashboard. The recap is especially social because it gives everyone a shared artifact immediately after a game.

That is the product test I get with a friend-group project: if a feature does not become part of the routine, there is no abstract user story to hide behind.

04.3What I learned

Sisyphus taught me that small real products still need state correctness. External APIs can disagree or update at different times, and a wrong recap or LP change is immediately visible to the people using the bot.

It also reinforced a pragmatic architecture lesson. JSON is fine until it is not. A Mac plus an Oracle host is fine until the product needs something else. I would rather change those decisions in response to real scale than design a multi-guild platform nobody asked for.

For a small product, correctness and usefulness matter long before theoretical scale.

04.4If it grows

If Sisyphus eventually supports more guilds, much larger history or significantly more concurrent state, the storage and deployment model should change with it. A database and a more centralized service topology would make sense then.

For now, I want to preserve what made the project fun in the first place: it is software built around one group of people and the way they actually play together.