# Eight by Eight

> A split planner and race-day strategy tool for eight-by-eight fitness races - the format popularised by HYROX: eight 1 km runs, each followed by one station (SkiErg 1000 m, sled push, sled pull, burpee broad jumps 80 m, row 1000 m, farmers carry 200 m, sandbag lunges 100 m, wall balls 100 reps), with a transition zone crossed 15 times. Independent tool; not affiliated with or endorsed by HYROX. HYROX is a trademark of its owner; SkiErg is a trademark of Concept2, Inc.

URL: https://eight-by-eight.skillsafe.ai/
API: https://eight-by-eight.skillsafe.ai/api.html

## What runs free in the browser (racekit.js, ribbon.js)

- Two pacing plans, A and B, for Open or Pro, men or women.
- Goal mode: a goal finish is spread over the 16 segments using a typical-finisher shape for the division, stations tilted by self-rated strength (strong x0.92, weak x1.10), renormalised to land on the goal.
- Numbers mode: race runs = fresh 1 km pace x (1 + compromised-run slowdown %), shaped so runs after heavy stations are slower and drifting by a fade % from run 1 to run 8; stations are your typed times; transitions = 15 x seconds per transition.
- What-if nudges: add or remove seconds on any segment.
- Race ribbon (SVG): every segment of both plans on one clock; a pasted result as a third row.
- B-minus-A chart per segment, split table with Plan A's running clock.
- Pacing flags: Run 1 more than 8% quicker than the average run, a run faster than the fresh 1 km pace or faster than any realistic 1 km (under 2:30), transitions under 15 s, station times far from typical, finishes near world-class.
- Paste a past result (lines like "Run 1 4:52", "SkiErg 4:31", "Roxzone 7:12", or a .txt/.csv file): the page reads which of the 17 parts it found, compares them with Plan A and ranks the biggest leaks. "Make this result Plan B" turns the result into Plan B so the next race can be planned against it.
- Exports: printable pace band (the plan on the open tab, with the strategy's call and cue per segment when one is on screen), CSV, Markdown, JSON. History of plan pairs saved to your account (collection `races`) with a browser mirror.

Reference splits are a rough typical shape per division for planning, not official results data.

## What the model does (metered, model gpt-terra)

- task "strategy" (built on @onewave-ai/game-strategy-simulator): reads both plans, the comparison, the pacing flags, any past-result leaks and optional athlete notes; returns a verdict (race Plan A, Plan B or a blend, expected finish, confidence, why), both plans weighed (upside, downside, risk, where each breaks), exactly 16 segment calls (hold / steady / push / recover, target split, cue, reason), a transitions target with tips, 3-5 contingencies, 2-4 what-ifs with their effect in seconds and whether they are worth it, and warnings.
- "Make these targets Plan B" turns the calls into Plan B so the called race sits on the ribbon next to Plan A.

## What the page checks in every reply

Missing segment calls, targets more than 15% from the picked plan (for a blend, from the nearer plan), targets plus transitions that do not add up to the stated finish (3%), more than six push calls, call words outside the four, pacing flags not addressed, and picking the slower plan on paper (explained, not blocked).

## Sources

- Race-day strategy lane: https://skillsafe.ai/skill/@onewave-ai/game-strategy-simulator