"OFFICIAL PAGE & ANNOUNCEMENT"

UMAMUSUME LOREBOOK

V5 BETA · 2026-10-18

✦ a canon-faithful world that keeps score — train her, race her, or be her.

POV: open — Trainer or runner · World layer + campaign engine · 8 books · 780 entries · 3,431 keywords · 19 scripts · 248 profiles · Public beta

GET THE FILES — Discord DM @umajoyer · the eight raw JSON lorebooks and the advanced prompt

About / the setup

Eight lorebooks that rebuild Umamusume: Pretty Derby as a place you can actually roleplay in: 248 canon-sourced character profiles, the racing science and the real tracks, the academy and its calendar, the biology and the ears that give her away — and a working campaign engine underneath all of it.

Out of the box this is a narrative world layer. No numbers, no HUD, no game bleed: a setting that knows what it is talking about, written so the prose carries the contest instead of a stat block. It injects nothing until you ask it to.

Give it one command and it becomes a campaign: a three-year climb from Junior debut to the URA Finale, resolved turn by turn — training and facilities, fatigue and injury, skill points and fans, campus events on the real calendar, and races that run four sections deep with a leaderboard that moves because of the call you made.

The difference is where the numbers come from. A script resolves every turn once and prints the result; the model's only job is to copy it. Gains, failures, overtraining, margins, purse and fan counts are arithmetic, not improvisation.

You pick the role at the start — her Trainer, or the runner herself. Neither is assumed, and the engine never writes your half.

It is not tied to one character. The set attaches to any bot: place all eight books, save the advanced prompt in your settings, and start.

⚠ Content notes

Plain and specific, so you can choose knowingly:

Public beta. Mistakes, flaws and bugs are possible, especially in RPG mode. An LLM can still hallucinate a number or a line — the engine and its checks reduce that, they do not eliminate it, and no lorebook can.

Dark World material — 38 optional entries. Doping and sporting integrity, suppressants and coercion, harassment and stalking, unwanted touch in the paddock, discrimination and pedigree elitism, abuse of professional access, manufactured scandal, forced retirement, medical disposability, online pile-ons. Depicted, never endorsed, and written as guidance on handling it responsibly rather than as flavour. They are keyword-gated — not constant, scan depth 1 — so they stay out of the prompt until you type the words that call them.

Sensuality. This is not a heavily sexualised set. Some sensuality does exist, and there is one entry on intimacy etiquette and touch boundaries.

Injury and retirement. Career-threatening injury, pain hidden from a trainer, rehab timelines measured in weeks, an acute mid-race breakdown entry, retirement handled as grief. Legs are careers in this setting.

Pregnancy and newborns appear in one Book 6 entry, as world fact.

Model quirk. The model may surface a real racehorse's name. Real horse names and classifications are still written as Umamusume — people, never racehorses, never ridden, never tack.

Ages. Junior, Classic and Senior are racing-career divisions, not ages. Umamusume are written as people.

No in-fiction safeword is implemented, so none is promised. The exits that genuinely exist are listed under Usage / chat help.

What the set holds

• Scale 8 books · 780 entries · 702,520 characters of lore

• Triggers 3,431 keywords across 761 keyword entries · whole-word matching, case-insensitive

• Scripts 19 JavaScript entries that resolve something and write into the prompt before the reply

• Cast 248 character profiles (Books 1–2) · 532 world entries

• Always on exactly one constant entry — the Uma Foundation grounding block

• Ordering 543 inclusion groups · priority and insertion order set per entry

• State every entry enabled · both copies of every stored field kept in sync

• Book 1 · Characters, part 1 — 124 entries · profiles A–M, plus one character worker script

• Book 2 · Characters, part 2 — 124 entries · profiles M–Z

• Book 3 · Racing, racetracks and events — 133 entries · race structure and narration, 7 racing-science and 6 strategy entries, 69 real racetracks worldwide, 18 event circuits, 4 race scripts

• Book 4 · World, organizations and calendar — 113 entries · teams, families, lineages, rivalries, URA governance, the 12-month in-universe calendar, glossaries, 2 scripts

• Book 5 · Campus, locations and daily life — 133 entries · 18 Tracen zones, 16 interiors, the Uma-app ecosystem, fan culture and media risk, daily-life and scenario suites, 5 scripts

• Book 6 · Biology, senses and psychology — 80 entries · the always-on foundation, ear and tail dictionaries, sports medicine, senses, appetite, intimacy etiquette, 8 onomatopoeia lists, 3 scripts

• Book 7 · Engines and narrative core — 24 entries · 11 narrative-core rules, 11 honorific entries, the master coordinator and the voice layer

• Book 8 · RPG mode and Dark World — 49 entries · the RPG engine, the campus-event engine, 9 reference entries, 38 optional dark entries

What is inside each book

Optional depth — safe to skip on a first read.

Books 1 & 2 — 248 character profiles

One repeatable structure per runner: core motive, appearance and body, attire, mindset, voice (with official self-introductions where they exist), behaviour tells, racing style, form, relationships, daily life, and a portrayal note on how not to flatten her. Sourced to umamusume.jp and umamusu.wiki.

Book 3 — racing

How a race is structured and narrated from gate to wire, commentator voice, resolving the finish, stamina and pacing cost, the field and the trainer's side. Then the science: stride and cadence, cleat traction, corner banking and lead-leg switching, slipstream and sand kickback, energy systems and the final three furlongs, going physics, gate launch. Then strategy: nige, senkou, sashi, oikomi, the pace triangle and pocket traps, racing psychology in the pack, distance and surface affinity. Plus 69 racetracks (Japanese central and regional, and overseas), 18 event circuits from the Classic Triple Crown to the Arc and the Dream Trophy League, glossaries, the Zone, paddock and winner's-circle suites, and four scripts: the universal race engine, stride telemetry, flow-state awakening, and the dynamic rival and paddock generator.

Book 4 — world, organizations, calendar

Dorms and student life, the student council and chairwoman's office, the Three Goddesses, the teams (Spica, Rigil, Canopus, Ascella, Hadar, Alkes, Sirius, Musaka), the families and lineages, the rivalry trios and generations, Tracen and the URA, the regional circuits, open roles for {{user}}, the 12-month calendar with its festivals and featured races, trainer lore, Winning Live, society and everyday Japan, glossaries, and two scripts for time-season-weather and for trust, fan ballot and media pressure.

Book 5 — campus and daily life

Every Tracen zone from the dormitory quadrangle to the racetrack forest perimeter, 16 interiors, Fuchu and western Tokyo, weather and season entries, the Uma-app ecosystem (Umatter, Umastagram, UmaTube, UmaTok, LANE) bridged to human apps, fan culture and its risks, enrolment and life cycles, the Life and Suite scenario packs (office nights, race registration, cleat fitting, fan letters, the athletic festival, the harvest fair, Project L'Arc, the crane game, the raffle wheel, the hot spring inn, curfew, the ghost of Tracen, declarations of war, the retirement run), five scripts including the clock and five-zone navigation tracker, and the gourmet entries.

Book 6 — biology, senses, psychology

The one always-on foundation entry (anatomy, rarity, academy rules, NPC naming), fatigue and recovery, ear and tail dictionaries, cleat tells, quiet-beat grounding, hearing and proximity, emotional escalation, vocalisation, adrenaline, public and private density, training mood and peak readiness, flow state, choking and the yips, clutch pressure, visualisation, focus narrowing, public defeat, photo finishes, the trainer-bond threshold, nutrition and weight control, hair and tail care, uniforms, sports medicine and injury limits, acute mid-race breakdown, post-race leg heat and ice boots, sleep habits, acoustics, draft-following, appetite, adrenal tremor, intimacy etiquette and positioning, care and reassurance, instincts, family and parentage, pregnancy and newborns, eight onomatopoeia lists, and three scripts for height and tactile differential, ear-and-tail subtext betrayal, and injury diagnostics with a rehab ladder.

Book 7 — engines and narrative core

Clothing, furniture and bipedal movement · first-person English speech and authentic character voice · acoustic surfaces · scent and independent ears · public persona versus private clubroom · scene-level want versus deeper need · personal racing stakes and dignity in victory or defeat · concrete action over catchphrase · respecting {{user}} agency and perspective · strict narrative mode by default · a timeless in-universe calendar. Plus the honorific set and two scripts: the master coordinator and the voice layer for stutters, slips and typographic stress.

Book 8 — RPG mode and Dark World

The RPG engine and the campus-event engine, then nine reference entries: rules and commands, training and conditions, race reference, skills, the campaign calendar, narrative form words, trainee and squad view, the HUD style guide, and the fourth-wall rule. Then 38 optional Dark World and Dark Lexicon entries, listed honestly above.

How deep the world goes

"Detailed" is a cheap word, so here is what is actually in there — named, not adjectives. All of it is world layer: it fires on keywords and stays out of the prompt until something calls it.

• 69 real racetracks. Tokyo at Fuchu, Nakayama, Kyoto, Hanshin, Chukyo, Sapporo, Hakodate, Fukushima, Niigata, Kokura. The regional sand and dirt circuits — Kasamatsu, Ooi, Kawasaki, Funabashi, Urawa, Morioka, Nagoya, Sonoda, Kochi, Saga, Kanazawa, Mizusawa, Himeji, Monbetsu, and Obihiro's ban'ei. Overseas: Longchamp, Chantilly, Deauville, Saint-Cloud, Auteuil, Ascot, Epsom Downs, Newmarket, Goodwood, York, Chester, Sandown, Kempton, Doncaster, Cheltenham, Aintree, the Curragh, Leopardstown, Galway, Fairyhouse, Santa Anita, Del Mar, Churchill Downs, Belmont Park, Saratoga, Keeneland, Gulfstream, Oaklawn, Aqueduct, Los Alamitos, Fair Grounds, Meydan, Sha Tin, Happy Valley, Seoul, Kranji, Flemington, Caulfield, Moonee Valley, Rosehill, Eagle Farm, Doomben, Gold Coast and Woodbine.

• Racing science, seven entries. Bipedal speed, cadence and stride length (pitch-souhou against stride-souhou). Racing cleat fit, traction and wear. Corner banking, centrifugal lean and lead-leg switching (temae-kae). Slipstreaming, aero drag and sand kickback (suna-kaburi). Energy systems — aerobic cruise, the breather (ikii-ire) and the final three furlongs. Track going physics from firm to heavy, including Japanese deep sand. Gate launch biomechanics, reaction time and the first-furlong scramble.

• Tactics with their real names. oonige and nige, senkou, sashi, oikomi and the makuri backstretch sweep. The pace triangle, pocket traps (mae ga kaburu), kakari — over-keen pulling — and homestretch psychology in a pack. Natural distance, surface and tactical affinity.

• 18 event circuits. The Classic Triple Crown (Satsuki Sho, Japanese Derby, Kikuka Sho) and the Triple Tiara (Oka Sho, Oaks, Shuka Sho). Senior Spring and Autumn Triple Crowns, the Grand Prix all-star double, the URA Finale tournament, Twinkle Star Climax, Champions Meeting 3v3 team showdowns, the Dream Trophy League, junior-division G1s, the dirt Triple Crown, the Japan Cup, the Prix de l'Arc de Triomphe with its French trials, the World Runners' Cup, the American Classic crowns, the European royal turf crowns, and the Dubai / Hong Kong / Oceania summits.

• The academy runs on a clock and a map. Tracen's student council, the chairman's office, the Three Goddesses statue in the central plaza, the public morals committee (fuuki iinkai), cultural and hobby clubs, Junior / Classic / Senior class divisions, dormitory roommate pairings across Ritto and Miho with a 22:00 curfew, classroom curriculum (race theory, sports nutrition, anatomy, media training), cafeteria seasonal menus with stamina dishes and weight-control plates, and a day that runs from the 05:30 dawn track to lights-out. A spatial tracker covers five campus zones with one-to-five-minute transit times between them.

• Teams, houses, lineages, rivalries. Team Spica, Rigil, Canopus, Ascella, Hadar, Alkes, Sirius and Musaka. The Mejiro, Sakura, Satono and Symboli houses, the Hishi sisterhood, the Air and Admire lineage, the New Era quartet, the Modern Turf Queens. Rivalries with shape: BNW, the Eternal Big Three (Oguri Cap, Super Creek, Inari One), the Golden Generation, Road to the Top, Vodka and Scarlet, Teio and McQueen, Special Week and Suzuka.

• Bodies that behave like bodies. An ear dictionary and a tail dictionary, independent horse-ear hearing, the cleat tell of impatience and weight-shifting, flow state and the zone, choking and the yips, race-focus narrowing and rail perception, margins measured in noses, necks and lengths, and defeat on television in slow motion with drooped ears. Sports medicine runs from suspected stress injury and soft-tissue concern to an acute mid-race breakdown and the trackside decision that saves a career — plus reporting pain, rehab ladders, post-race "leg heat" and ice-boot wrapping, sleep ear-bonnets (mimi-ate) and bedtime tells. Eight separate onomatopoeia lists: mood, pain and tiredness, effort and motion, food and drink, weather and light, crowds and sleep, impacts, small gestures.

• Instincts, written as behaviour. A flehmen-inspired lip curl, mutual grooming and chin resting, startle response and safe personal space, standing naps and dream running, humidity and tail static, peripheral vision and food attention. Height and body proportion, tail hugs and tail gestures, careful touch, and how she asks for help.

• Off-days and errands. The arcade crane game, the shopping-district raffle wheel and carrot market, the secret hot-spring inn, ear-slit hats and tail ribbons and failed incognito disguises, seasonal gift-giving, off-season hometown visits, selection races and the trainer scouting draft, signature racewear (shoubufuku) design and first fitting, breaking a slump, and the retirement run — the last race, and passing the torch.

• Phones, apps and fan culture. Umatter and LANE, URA Trainer Pro, Clocker Sync and race-analytics portals, anonymous racing textboards, oshi-katsu and cameko paddock photography with zero-flash etiquette, egosurfing and doomscrolling and what a trainer does about phone confiscation, civilian apps (UmaEats, Uma-Lane Nav, UmaTunes), Winning Live idol culture with center position and setlists, and character phone quirks from gyaru influencers to boomer-texters.

• How people talk. Eleven honorific entries — -san, -chan, -onii-chan, -sama, -ojou / -ojou-sama, -kun, -dono, -nee, senpai, kouhai, sensei — each with when it is right and what it signals. A voice layer handles stutters, slips and typographic stress, in dialogue only. Narrative-core rules keep clothing, furniture and bipedal movement human-scale, keep acoustic surfaces distinct (school loafers against steel racing cleats), keep public persona separate from private clubroom vulnerability, and keep the calendar timeless: no real-world years, no JRA codes.

• Intimacy is written as etiquette. Sensitive zones — ears, the tail base — and touch boundaries, body and tail positioning for bed, lap and embrace, non-sexual comfort and personal space, reassurance and requested help. This is not a heavily sexualised set; the material exists so the world is coherent, not for its own sake.

• Regional circuits and the logistics of leaving. Kasamatsu's Tokai sand spirit, the southern Kanto dirt circuit, the Hokkaido summer series, and overseas expeditions handled as work: quarantine, charter flights, turf adjustment, the Chantilly training villa.

None of this arrives by default. It is keyword-triggered and sits in the prompt only when the scene earns it — which is the next section.

How the writing behaves

Book 7 is small — 24 entries — and it is the one that decides whether the other seven read like a setting or like a costume. These are its eleven narrative-core rules, and they are not suggestions:

• Human scale. Clothing, furniture and bipedal movement are written for people who happen to have ears and a tail — doors, chairs, headphones, hats, tables, beds.

• Voice. First-person English speech with an authentic character voice per runner, not one interchangeable narration voice wearing different names.

• Acoustic surfaces. School loafers on corridor tile and steel racing cleats on track sound like different things, and the prose is expected to notice.

• Scent and hearing. Olfactory and auditory range written as real senses, with ears that move independently of attention.

• Two faces. Public persona against private clubroom vulnerability — what she performs and what she admits are not the same scene.

• Want against need. Scene-level want separated from the deeper emotional need underneath it.

• Stakes and dignity. Personal racing stakes, and dignity in victory or in defeat.

• Action over catchphrase. Concrete physical actions instead of a repeated signature line.

• Your agency. The rules explicitly respect the reader's agency, role and perspective. Nothing you have not chosen gets chosen for you.

• Narrative by default. Strict narrative mode unless RPG mode is switched on — zero unrequested game bleed into a prose scene.

• A timeless calendar. In-universe only: no real-world years and no JRA codes anywhere in the setting.

Alongside the rules sit eleven honorific entries — -san, -chan, -onii-chan, -sama, -ojou and -ojou-sama, -kun, -dono, -nee, senpai, kouhai, sensei — each with when it is the right choice and what using it signals about the relationship. A separate voice layer handles speech stutters, slips and typographic stress, and it fires in dialogue only, never in narration.

The vocabulary, sampled

A setting is only as detailed as the words it insists on. Here is a sample of what the set knows and will use when the scene earns it:

• pitch-souhou / stride-souhou — the two ways to be fast: cadence against stride length.

• temae-kae — switching the lead leg through a corner, and what it costs mid-turn.

• suna-kaburi — sand kickback; the reason nobody wants the inside trip on dirt.

• ikii-ire — the breather: the aerobic cruise before the final three furlongs decide it.

• kakari — over-keen pulling; a runner burning herself up inside the pack.

• makuri — the backstretch sweep an end-closer makes when the pace collapses.

• mae ga kaburu — the pocket trap: hemmed in, with the gap you needed already gone.

• nige / oonige / senkou / sashi / oikomi — runaway, front-runner, pace chaser, late surger, end closer. Five ways to run the same race, and five different failures.

• teitetsu — the racing cleat: fit, traction, wear, and the campus shoe-fitter's workshop.

• batajuu — the pre-race weight check, and the cafeteria drama that comes with it.

• shoubufuku — signature racewear: designed, tailored, fitted for the first time.

• mimi-ate — sleep ear-bonnets, and the bedtime tells that go with them.

• fuuki iinkai — the public morals committee, and what it does with a curfew.

• sanboshi — the Three Goddesses statue in the central plaza, where pre-race vows get made.

• Ritto and Miho — the two dormitories, their roommate pairings and their 22:00 curfew.

• fukubiki — the shopping-district raffle wheel, next to the carrot market.

• kimodameshi — the nighttime courage test, and the "Ghost of Tracen" that goes with it.

• oshi-katsu / cameko — fan support culture and paddock photography, under a zero-flash rule.

• egosa — egosurfing: reading yourself online after a race, and the doomscrolling that follows.

• Uma Lanes — public lanes sized for Umamusume, and the architectural ergonomics behind them.

• ban'ei — Obihiro's own discipline, the outlier on the national circuit.

Eight separate onomatopoeia lists sit behind this — mood, pain and tiredness, effort and motion, food and drink, weather and light, crowds and sleep, impacts, and small gestures — so a scene can be specific about sound instead of reaching for the same three words.

What fires the entries

Keys are the words you actually type in this fandom — names, nicknames, places, race terms, body words, commands. A small sample of the 3,431:

• Books 1–2 her name and her nicknames — agnes tachyon tachy digitan ayabe maya-chan air groove-san

• Book 3 starting gate backstretch photo finish g1 going paddock winning post arima kinen

• Book 4 student council seitokai team spica winning live three goddesses dorms

• Book 5 curfew crane game cafeteria library karaoke beach camp tail dryer visit her room

• Book 6 ears tail hearing scent injury fracture grooming consent

• Book 7 honorific first person school loafers racing shoes familiar scent

• Book 8 rpg mode trainerpov rpg mode umapov narrative mode hide hud status train outing learn camp

How it works

Keyword entries fire on the words in recent messages. Whole-word matching is on and case is ignored, so ordinary prose triggers them — you do not have to speak in keywords.

Script entries run before the reply and write into the prompt. All 19: the Universal Race Engine, Racing Science and Stride Telemetry, Flow State and Zone Awakening, the Dynamic Rival and Paddock Gate Generator, Time-Season-Weather, Trust-Fan Ballot-Media Pressure, the Tracen Clock and Five-Zone Navigation Tracker, the Crane Game and Raffle and Shrine Fortune rollers, the Phone Ticker and social-media engine, Daily Routine and Special Scenario, the Dormitory Curfew and Midnight Kitchen engine, the Height-Body-Tactile Differential, the Ear-Tail-Micro-Expression Subtext engine, Injury Diagnostics and Rehab, the Uma Engine Master Coordinator, the Voice Layer, the RPG Engine, the Campus Event engine, and one character worker.

One entry is constant: the Uma Foundation — a compact grounding block (anatomy, rarity, academy rules, NPC naming) that is always present so the world never has to be re-explained.

Priority decides what survives under pressure, insertion order decides where it sits, and entries sharing an inclusion group may be mutually exclusive — so groups are used deliberately rather than decoratively. Both copies of every stored field are kept in sync, because an asymmetry between them is a real bug.

Verify it yourself: Test Chat → Changes (the Debug Panel) shows exactly which entries fired on a turn. Two verified fires before you trust an entry.

Character info

There is no single protagonist here — the cast is the set. 248 profiles, each with the same thirteen fields, so any two runners can be compared at a glance: what she wants, how she sounds, what her ears do when she is lying, which ground and distance suit her, who she rooms with, who she has vowed to beat, and what she eats at eleven at night in the dorm kitchen.

Canon and interpretation are separated. Profiles are sourced to umamusume.jp and umamusu.wiki. The Suite scenario packs, the Uma-app ecosystem, the Dark World and Dark Lexicon material, and some family relationships are original or AU additions to the canon spine — written to be playable either way, and labelled as additions rather than dressed up as official.

Your info — the role is open

No role is assigned to you and nothing about you is prewritten. This bio does not set your name, gender, appearance, thoughts, feelings, dialogue or actions, and it is not a persona worksheet.

In narrative mode you can be anyone the setting supports: a trainer, a student, a reporter, a stable hand, a civilian standing at the rail. In RPG mode you choose between two seats at the start — Trainer or the runner herself — and the engine never guesses which one you meant.

What you can do

Without the engine

• Roleplay the world as a setting: academy days, dorm nights, media scrums, festivals, away races, Winning Live rehearsals.

• Name a runner and her profile arrives — voice, tells, relationships, suited ground.

• Race in prose: the race-scene shape, the commentator, the margins and the photo-finish wait are all keyed.

• Play the dark material only if you type it — those entries stay closed until the words appear.

With the engine (opt-in)

• Run a three-year campaign: one action per half-month turn, from Junior debut to the URA Finale.

• Train five facilities, level them, bank summer camp at max level — and pay for it. Overtraining from the fourth session in a row cuts gains, multiplies cost and can injure her.

• Rest, take an outing, or go to the infirmary to clear conditions.

• Spend skill points; watch fan counts and race grades gate your entries.

• Race section by section: style at the gate, then hold / move up / save ground / go wide, then wait / spurt / inside / outside, then all-out / measured / switch lanes — with a leaderboard, an announcer call and a stamina read every time.

• Answer campus events on the real calendar: the shrine visit at New Year, birthdays, the arcade raffle.

• Keep more than one trainee, switch who you are playing, move the clock, set the scene.

• Or just answer the numbered move list with a number. It prices each option for you: gain, cost, fail chance.

RPG mode in detail

The clock

One turn is half a month, with an early and a late half. T1 is early July of Junior year, T13 starts Classic, T37 starts Senior, and T61–T63 are the URA Qualifier, Semifinal and Final. Summer camp runs over T25–T28 and T49–T52, where every facility counts as Lv5 and banks no progress. An established runner starts at T37; a debut starts at T1. rpg turn <n> moves the clock.

The turn

One action per turn: train one of five facilities, rest, outing, infirmary, or race. status and learn <skill> cost no turn. Every HUD ends with a [ Your Move ] list of up to five priced options, and the fourth slot changes with the situation — the infirmary while a condition is active, otherwise a skill purchase or status.

The HUD

Plain markdown rows, no code fence: header, Turn · Scene · Goal, Status · Mood · Vitality · SP · Bond, Stats (max 1200) with this turn's changes inline, Session naming what happened, Skills, Squad · Facility · Fans, then the move list. Aptitudes print on status and race day only — not every turn. Mood is five words with five faces: 😩 Awful · 😟 Bad · 😐 Normal · 🙂 Good · 😄 Great.

The numbers underneath (training, failure, skills, races)

Optional depth — safe to skip on a first read.

Training gains

Main gain by facility level, Lv1 to Lv5 — Speed 11, 12, 13, 14, 15 · Stamina 10, 11, 12, 13, 14 · Power 9, 10, 11, 12, 13 · Guts 8, 9, 10, 11, 13 · Wit 10, 11, 12, 13, 14. Side gains: Speed training adds Power 6–8; Stamina adds Guts 6–8; Power adds Stamina 6–8; Guts adds Speed 5–6 and Power 5; Wit adds Speed 2–4. The main gain is multiplied by mood (0.8 / 0.9 / 1.0 / 1.1 / 1.2) and any growth bonus.

Vitality and failure

Vitality runs 0–100 and is spent per session (Speed 21 at Lv1 up to 27 at Lv5; Stamina 19; Power 20; Guts 22). Wit restores 5 at every level — except from the fourth session in a row, when it drains 5 instead. Failure chance is (55 − Vitality) × 1.6, capped at 95, and halved for Wit. A success pays 4 SP (Wit pays 5). A failure costs 12 points from the trained stat and 1 mood. A big failure — Vitality under 20 — costs 36 and gives Practice Poor. Injured adds 20 to the failure chance and doubles the cost.

Overtraining

Rest, outing, infirmary and a race break the run. Training again without rest multiplies the Vitality cost: ×1.5 for the second session in a row, ×2 for the third, and so on. From the fourth, the HUD says OVERTRAINING, N in a row: a fatigued success cuts the main and side gains to 60%, and each session carries an injury chance of 10%, rising 5% per extra session to a cap of 40%. A failure while overtraining drains the whole body — the trained stat takes the failure hit and each of the other four loses 5 more, so all five print a minus.

Facilities

One level per four uses, max Lv5, and the use is banked even on a failure. Every facility prints its own counter, so nothing silently resets. During summer camp all five count as Lv5 and bank nothing.

Conditions

Twelve, up to eight at a time, shown in the Status row: Injured, Practice Poor, Night Owl, Slacker, Skin Outbreak, Slow Metabolism, Migraine, Under the Weather, Practice Perfect, Charming, Hot Topic, Fast Learner. The infirmary gives 20 Vitality and clears the curable set. Practice Perfect adds 20% to one success and is then spent; Charming makes an outing worth 2 mood; Hot Topic multiplies race fans; Fast Learner takes 10% off skills.

Skills and SP

learn <skill> costs no turn. Prices: an innate hint 110 SP · a skill marked ◎ 220 · a named advanced skill 320 (Maestro, Professor, Arc, Legerdemain, Planner, Clairvoyance, Turbo, Feat, Plan X, Aim for the Front, Taking the Lead) · anything else 170. A unique skill is owned from the start.

Races

Four sections plus the finish. Distance categories: sprint to 1400m, mile to 1800m, medium to 2400m, long beyond. Aptitude multipliers S 1.05 · A 1.0 · B 0.9 · C 0.8 · D 0.6 · E 0.4 · F 0.2 · G 0.1. Margins are named: ns, hd, nk, lengths in halves, dst beyond ten. Grades gate entry by fan count and pay purse, SP and fans by place; a race costs 15 Vitality and a win gives 1 mood. Race mode is quick (one result card) or full (every section, every call).

Campus events

Fixed dates, birthdays, and seeded random draws in three densities (rare, default, lively). An event never fires on two turns running and never interrupts a race. Two or three options; the outcome returns as an event[...] stamp and lands on the sheet the following turn, so nothing is applied twice. rpg events off silences them.

What a turn looks like

Real engine output — unedited, from the shipped scripts

A training turn that goes right

[ HUD: Nice Nature · Training Session ]

Turn: T38 · Senior late Jan | Scene: Tracen track · morning | Goal: URA Qualifier at T61 (23 turns)

Status: Healthy | Mood: 😐 Normal | Vitality: 78/100 | SP: 204 | Bond: 50/100 ♥♥♡♡♡

Stats (max 1200): Speed 785 (+5) | Stamina 576 | Power 941 (+5) | Guts 548 (+8) | Wit 792

Session: SUCCESS (roll 71 vs 0% fail) · Guts training Lv1 · Speed +5, Power +5, Guts +8 · Vitality 100→78

Skills: I Can Win Sometimes, Right?

Squad: Nice Nature | Facility: Speed Lv1 0/4 · Stamina Lv1 0/4 · Power Lv1 0/4 · Guts Lv1 1/4 · Wit Lv1 0/4 | Fans: 120,000

[ Your Move ]

1. Train Guts — +8 Guts, costs 33 Vitality, 0% fail
2. Rest — Vitality 78→100, no cost
3. Outing — +10 Vitality, +1 Mood
4. Learn a skill — 204 SP available

The turn it all goes wrong (overtraining, big failure)

Optional depth — safe to skip on a first read.

Fourth session in a row at 14 Vitality. All five stats fall, the facility still banks its use, she is injured, and the move list swaps in the infirmary and tells you the truth about the odds:

[ HUD: Nice Nature · Training Session ]

Turn: T41 · Senior early Mar | Scene: Tracen track · morning | Goal: URA Qualifier at T61 (20 turns)

Status: 😖 Practice Poor, 🩹 Injured | Mood: 😩 Awful | Vitality: 0/100 | SP: 208 | Bond: 50/100 ♥♥♡♡♡

Stats (max 1200): Speed 775 (-5) | Stamina 548 (-36) | Power 931 (-5) | Guts 547 (-5) | Wit 787 (-5)

Session: BIG FAILURE (roll 2 vs 66% fail) · Stamina training Lv1 · Speed -5, Stamina -36, Power -5, Guts -5, Wit -5 · Vitality 14→0 · OVERTRAINING, 4 in a row — all five stats fell · Stamina facility now Lv2 · became Injured (infirmary cures it) · Mood → Awful

Skills: I Can Win Sometimes, Right?

Squad: Nice Nature | Facility: Speed Lv1 0/4 · Stamina Lv2 0/4 · Power Lv1 0/4 · Guts Lv1 0/4 · Wit Lv1 0/4 | Fans: 120,000

[ Your Move ]

1. Train Guts — +4 Guts, costs 132 Vitality, 95% fail (overtraining! break the run first)
2. Rest — Vitality 0→50, no cost
3. Outing — +10 Vitality, +1 Mood
4. Infirmary — clears Practice Poor, Injured, +20 Vitality

Race day, then the first section

Optional depth — safe to skip on a first read.

[ HUD: Oguri Cap · Race Day — Arima Kinen ]

Turn: T46 · Senior late May | Race: G1 · 2500m turf · Nakayama

Course: right-handed · short straight, steep final hill | Going: Firm (Sunny) | Field: 12 | Gate: 9 (orange) | Odds: fancied

You: Form 69 | Mood: 😐 Normal | Vitality: 100/100 | Bond: 50/100 ♥♥♡♡♡

Stamina for 2500m: 63% — fresh

Aptitudes: Turf A · Dirt B | Sprint E · Mile A · Medium A · Long B | Front F · Pace A · Late A · End D

Rivals to watch: Front: Seiun Sky | Pace: Narita Top Road, Zenno Rob Roy, Tokai Teio | Late: Mr. C.B.

[ Your Call ]

1. Front · 2. Pace · 3. Late · 4. End

You call the front. The race moves one section and stops for the next call:

[ HUD: Arima Kinen · 1/4 first corner ]

Turn: T46 · Senior late May | Race: G1 · 2500m turf · Nakayama | To go: 2100m | Pace: even

Board: 1 Seiun Sky · 2 Daiwa Scarlet +0.5 · 3 Meisho Doto +0.5 · 4 ▶ Oguri Cap +1 · 5 Grass Wonder +1

Announcer: They are away! Seiun Sky takes the early lead, Daiwa Scarlet tucked in behind. Oguri Cap is 4th, settling in.

Your stamina: 48% — steady

Race events: clean break

[ Your Call ]

1. Hold · 2. Move up · 3. Save ground · 4. Go wide

A campus event, open and resolved

Optional depth — safe to skip on a first read.

The card appears after the HUD and stays open until you answer it with a bare number:

[ Event: New Year's shrine visit · T37 ]

Hatsumode in the cold: two claps, eyes closed. What is the prayer for?

1. Health
2. Growth
3. Victory

You answer 2. The card resolves, and the stamp line at the end of the reply is what applies the effect next turn:

[ Event: New Year's shrine visit · T37 ]

→ Growth: All stats +60

event[Oguri Cap] @T37 A+60

How to start

1. Ask for the files. The set is handed out as raw JSON — DM @umajoyer on Discord and I will send the eight lorebook files and the advanced prompt text. There is a busy life behind this account, so a reply may not be instant; one clear message is enough.

2. Place all eight lorebooks. They live under Scripts → Lorebooks, and each one must be Assigned to the bot. Partial sets work, but the world gets holes in it.

3. Save the advanced prompt in your chat's LLM settings. This is the step people skip: RPG mode does not behave without it, because it is what teaches the model to copy the engine's output instead of inventing its own.

4. State your seat, your runner and your start. The engine takes it from there.

5. Play. Type an action, or answer the numbered list with a bare number.

rpg mode trainerpov with <name> established  ← you are the Trainer; she keeps her build
rpg mode umapov as <name>  ← you are the runner; debut start by default
rpg preset established | rpg preset debut  ← reset her start later
train speed | train stamina | train power | train guts | train wit
rest | outing | infirmary | race <name> | learn <skill> | status
rpg pov trainer|uma · rpg me <name> · rpg sheet <name>
rpg trainee <a>, <b> · rpg turn <n> · rpg scene <place>, <time>
rpg race mode quick|full · rpg race skip · rpg race quit
rpg events rare|lively|off|on · rpg mode off

rpg mode on is retired and starts nothing — it tells you which of the two seats to take. rpg mode off returns you to narrative mode and removes every HUD.

Which opening to pick

This card ships with four alternative greetings. They are four different doors, not four wordings of the same paragraph — and none of them assigns you a seat, a runner, a relationship or a history you did not choose.

• 01 — Born to Run. The front door, and the only greeting that opens on the world instead of on you: a cinematic prologue over the turf, the parade ring, the gates and the campus at 05:30, across an unnamed ensemble field with no character of yours written into it. Then a seat menu — A the stopwatch (her Trainer), B gate nine (the runner herself), C the rail (press box, paddock, grandstands, a rival’s camp), D the back office (the academy, its dorms and its infirmary), E blind draw, which is the set the field command — plus one word on how it should run: narrative or RPG. How she runs and where you start on the calendar are optional extras. Answer with a letter and a word, for example A narrative, or leave the list alone and write your opening in your own words. Its tone line states the sensuality position and names the gated dark entries in plain words.

• 02 — CYOS. Create your own scenario: a fill-in card for runner, seat, place, season, tone, numbers and first scene. Leave anything blank and it gets chosen for you — then shown to you, so you can overrule it. Its tone line names the dark option and its scope in plain words.

• 03 — Quiz. Ask the lorebook some questions, mostly used for testers.

• 04 — Advanced Prompt. A utility card, not a scene. It carries the full advanced prompt that RPG mode needs, ready to copy into your chat's LLM settings. Replying to it does nothing by design — start your chat from 01, 02 or 03 afterwards.

Whichever door you use, the world is the same one: 248 profiles, 69 real tracks, 18 event circuits, and the rules in How the writing behaves above. If this list and the actual greeting list ever disagree, the greetings are right and this page is stale — tell me and I will fix the page.

Compatibility and requirements

• Any bot. This is a world layer, not a personality — it does not overwrite your bot's character, and in narrative mode it injects nothing at all.

• Requires the eight lorebook files placed and Assigned under Scripts → Lorebooks. The advanced prompt is a separate, per-chat user setting; it cannot ship inside the card, which is why it is published here.

• Built for free JLLM: 9,001 tokens of context, and it never requires a proxy. The set is sized to run inside that budget.

• Peak hours: I have seen JLLM drop to roughly 4–6k tokens at peak instead of 9,001. That is my observation, not a platform promise — off-peak simply works better.

• Survives a bad reply. The engine re-reads its own state from the newest HUD visible in recent messages, so one mangled or hallucinated panel does not corrupt the campaign — the next turn recovers. This was tested at a four-message window.

• Tested: 141 automated checks across two suites, plus a full 122-turn simulated campaign (start to URA finals) at about 7 ms per turn.

Usage / chat help

Advanced prompt

The full prompt ships in the download: seven worked examples covering a campaign start, a training turn with an open event card, resolving that card, status, race day, the first race call, and a fresh chat with nothing running. It is a display mirror — the copy you install in your settings is the behaviour source, and this text is synchronised to it as of 2026-10-18. These are the rules it teaches, verbatim:

• Copy the HUD from the live [=== RPG STATE ===] block exactly, as plain markdown. No code fence, no backticks, no table, no HTML, no re-bulleting.

• Keep every row on its own line with one blank line between rows. Keep every bold label spelled and capitalised exactly as printed. Keep the · and | separators. Never add, drop, merge, reorder or reword a row.

• Copy every number, sign, arrow, emoji, heart, mood word and date exactly as printed. Never recalculate, round, flip a sign, change a mood word, or write a different turn or roll.

• A failure shows minus changes in the Stats row — for example "Stamina 552 (-36)" — and its Session row says FAILURE or BIG FAILURE. A failure never shows a plus and never raises Vitality.

• Mood words are only Awful, Bad, Normal, Good and Great, each with its emoji. For prose, ability is only Poor, Fair, Good, Great or Elite; the HUD itself prints no letter grades.

• Copy the [ Your Move ] list too, verbatim, as the last lines of the reply. Never invent, renumber or reprice an option: the gain, cost and fail chance in it are the engine's.

• A race card ends with [ Your Call ] instead of [ Your Move ]. Stop after it and wait for the user.

• An event card is copied the same way, after the HUD, and the reply ends with its event[...] stamp line whenever the state block asks for one.

• The roll and the fail percentage are copied as printed. A roll at or below the fail percentage is a failure; a roll above it is a success.

Settings

My recommendation for this set — not a JanitorAI default and not a guarantee. This is what the engine was tested under:

temperature 0.75 · repetition penalty 1.05 · top_p 0.85 · min_p 0.08
max tokens 0  (or 700–800 if replies balloon)

Memory

Chat memory is yours to keep. If you want the official shape, JanitorAI's six headings are:

Environment: · Relationship Dynamic: · Current Plot Points:
{{char}} notes: · {{user}} notes: · Important Past Events:

Short bullets, one tense, facts rather than scenes, updated at scene changes and pruned ruthlessly. One caveat specific to this set: in RPG mode the engine already holds the campaign state — turn, stats, Vitality, SP, bond, facility levels, conditions, fans. Do not copy HUD numbers into memory. Duplicated facts go stale, and a stale number sitting in memory is exactly what tempts a model to invent a panel of its own. Keep memory for the story: where you are, what she is to you, what just changed, and the one thing that must not change.

Agency

The engine never writes your words, your choices, your feelings or your rolls; point of view is never guessed; and the HUD, the event cards and the rule notices are out of the fiction — no character sees, reads or quotes them. That boundary is enforced in code and backed by the behaviour entries, so it is not decoration.

No promise is perfect, because the model underneath is not. If a reply ever writes your half, edit it. Editing is available in the current UI and it is the fastest correction; treat it as an option rather than a guaranteed training mechanism.

Persona and turn tips

• Keep your persona lean and factual — roughly 500 tokens or less, third person, only the facts that must not be inferred (name, pronouns, appearance, your seat in the world). Habits, backstory and feelings in the persona field are the most common hidden cause of a bot writing your lines.

• Write full turns. The model mirrors your energy, and one-word replies tend to shrink replies back.

• In RPG mode you can answer with a bare number. If an event card is open, the number answers the card first — that is deliberate, so one reply never does two things.

Exits

• rpg mode off — back to narrative mode, every HUD removed. Implemented and tested.

• rpg events off — silences campus events; rpg events rare thins them out.

• rpg race quit — scratches you from a race in progress; rpg race skip runs the rest without further calls.

• The 38 Dark World entries stay out of the prompt until you type the words that call them.

• No in-fiction safeword exists in this set, so none is advertised.

Troubleshooting

• The model invented numbers or freehanded a HUD. The advanced prompt is not installed, or an old copy is. Install the current one — the engine supplies the numbers, the prompt teaches the copying.

• Nothing fires. Lorebooks must be Assigned to the bot, not just created. Check Test Chat → Changes to see what actually fired.

• A runner's profile did not arrive. Keys are whole-word: use her name or one of her listed nicknames, not a fragment.

• The bot writes my lines. Trim your persona to facts, and edit the reply. The engine never does this itself.

• Replies ballooning. Set max tokens to 700–800, and raise the repetition penalty to 1.1 if one phrase keeps looping.

• RPG bled into a normal roleplay. rpg mode off. Narrative mode is the default and injects nothing.

• Events are too frequent. rpg events rare or rpg events off.

• The campaign looks wrong after a long chat. The engine recovers from the newest HUD in recent messages; if your chat window is very small, keep the last HUD in view.

Known issues

• Big-failure wording. The training reference says a big failure happens when Vitality is under 20 or when the roll is more than 70 above the failure chance. A failure only occurs at or below the chance, so the second branch can never fire — in practice a big failure comes from the Vitality rule. The engine and the text agree; the text is over-specified.

• Skills have no catalogue. learn <skill> accepts any name and charges one of four tier prices. A full skill list is on the to-do list.

• Bond is inert. The 0–100 bond track with its five hearts is printed, tracked and remembered across turns, but it does not yet change training, races or events. Also on the list.

• Key lists. In 37 Book 5 entries the key field and the raw key string disagree. Both are printed in the guide; the difference is not resolved.

• It is a beta. Everything above is disclosed rather than patched silently, and there will be things I have not found yet. That is one reason this is public.

Notes, credits and rules

Version

V5 · public beta · 2026-10-18. No genre preset is claimed for the set itself — this is a canon layer, not a mood. Pair it with whatever preset your own card is built on.

Which version this is. Work on the set began in August 2026. V5 is the first public release and the only version being distributed — the largest revision of the set so far, and the one every part of this page describes. Earlier revisions were private builds; nothing before V5 is supported, and no older file will behave like this one. A final version follows later.

Credits

Built by combining the strengths of other people's work into one — mine. References and guidance:

• 1. @PYTOR  ·  2. @XTR4  ·  3. @Olykiouss  ·  4. @nokturn2008  ·  5. @deanu (formerly manhattancafelover)

• 6. @Secreteriat  ·  7. @I46  ·  8. @reisen912  ·  9. @ilovepoketaki  ·  10. @Neter44rap

• 11. Anonymous Name  ·  12. @ShadowCentury  ·  13. @ViniSigmaBased  ·  14. @OhbanL  ·  15. @gigabrowser

Plus three or more public lorebooks by anonymous names.

To be completely honest: not all of these were really used, and the most influence on my lorebooks goes to @PYTOR and @deanu. The others were used for AU settings, facts, and so on. Character profiles are sourced to umamusume.jp and umamusu.wiki. Credit is not permission — nothing here is borrowed without the intent to name it.

Rules for using it

• Credit @uma_enjoyer when you use this lorebook. That is all I am asking. It is free and it is not gatekept — this project cost me sleep debt and a lot of my own time, so please be kind and include my name.

• Moving it to another platform? DM me on Discord — @umajoyer — and I will help you transfer it. I have a busy life, so I cannot promise full attention.

• Edit it freely. Do not rename it. This is strictly V5. If you explicitly name your version V6, or any version that does not exist yet, I will report it — I am strict about this because I do not want fake releases circulating under a number nobody has earned.

• V5 is the only version going public for now. When the final version is done, it goes public too.

Why it is public

Construction of V5 took around 51+ hours live, believe it or not. I averaged roughly eight hours of screen time a day this week simply building it. You do not have to believe that if it sounds absurd, but I am being honest.

I cannot do this alone any more, which is the other reason it is public: so you can open the scripts, notice the small mistakes I made along the way, and tell me. Additions, corrections and bug reports are the whole point of releasing it in beta.

Umamusume: Pretty Derby belongs to Cygames and its rights holders. This is a fan-made, non-commercial lorebook set: nothing here is official, and no ownership is claimed.

UPDATE LOG:

• v5.1 for LB1,LB2,LB4 adding Ryo Nishizaki as Team Spica Trainer. Kitasan Black added as Member.