# Macro Desk

> Your diet phase, by the numbers: enter a bodyweight, a goal and a training week, and the browser computes the Renaissance Diet 2.0 numbers - calories for each kind of day, protein, carbs and fat, a meal template around your training, the phase length and magnitude, the projected bodyweight and the maintenance that follows - for free; then a coach writes the plan from those numbers, and three weeks later reads your weekly average weights and adjusts it.

Live at https://macro-desk.skillsafe.ai/ · API tutorial at https://macro-desk.skillsafe.ai/api.html · Tokens at https://macro-desk.skillsafe.ai/tokens.html

## What it does

One work object: one person - sex, bodyweight (lb or kg), optional body fat, age and height, goal (fat loss, muscle gain, maintenance), training experience, a seven-day week of rest / light / moderate / hard days, meals a day, wake, training and sleep times, a fat preference, plant-based or not - plus a free-text description of how they train and eat.

The free engine (no account, no model), the book's rules as written: maintenance at 12, 13.5, 14.5 and 16 kcal/lb for rest, light, moderate and hard days and its weekly average; the daily deficit or surplus implied by the chosen rate of bodyweight change at 3,500 kcal per lb, checked against the book's 300-500 (loss) and 200-400 (gain) kcal heuristics; protein at 1 g/lb (1.2 plant-based) on every day; carbs at the day type's starting point (0.5 / 1.0 / 1.5 / 2.0 g/lb) with floors (0.3 / 0.5 / 1.0 / 1.5); fat filling the remainder at or above 0.3 g/lb, or held at a preferred level with carbs filling; days where the floors bind are flagged. A meal template: first meal after waking, pre-workout 90 minutes before and post-workout 60 minutes after training, bedtime meal, the rest evenly spaced; protein between 1/8 and 1/4 of the day per meal, carbs 25%/35% (loss) or 20%/30% (gain) in the pre/post window, fat biased away from training. The phase: rate x weeks = magnitude, week-by-week projected bodyweight, judged against 0.5-1%/week, 6-12 weeks and 3-10% (loss; 2%/week and 16 weeks are hard limits) and 0.25-0.5%/week, 6-16 weeks, 3-10% (gain; 5% advanced, 24 weeks beginner); body-fat classification and floors (5% men, 15% women); the maintenance phase after (2/3-1x the cut; 4-8 weeks after a gain) and the transition calories. The check-in: two to five weekly averages, the least-squares trend in lb/week and %/week, the gap to target, the calorie change that closes it (3,500 kcal/lb over 7 days), the grams as fat or as carbs, the book's macro order (fat first when cutting, carbs first when adding, protein never), the +/-1.25% maintenance band, and phase progress against the planned weeks and the magnitude ceiling.

Two metered lanes (gpt-terra, one system prompt with a task router):

- `plan` - the coach writes the plan (calories by day type, macros, meals, the weigh-in rule, what follows the phase), judges the goal against body fat and experience (good / questionable / wrong_phase), places the five levels of the diet-priority pyramid for this person, gives food guidance grounded in what they said, judges ten named pitfalls, and checks the user's own current plan claim by claim. Verdict: ready / ready_with_changes / not_advisable.
- `adjust` - the coach reads the trend and the log, decides whether the scale is signal or noise, writes the adjustment (the calorie change, which macro, the new calories, how long to run before the next check), names the macro change, places the phase (continue / complete / stop_early) and flags adherence. Verdict: hold / cut / add / end_phase / wait.

Handoffs: "Check in on this plan" carries the plan's rate, calories and phase length into the adjust lane; "Rewrite the plan at this weight" carries the check-in's current weight back. Every number the model writes is read back against the engine and disagreements are shown. The desk refuses what the book refuses, adds its own safety limits (no cut when BMI is under 18.5 or would end under it, none past a 20% loss in one phase, a warning on any cut day under 1,200 kcal for women or 1,500 kcal for men), and defers to a clinician for pregnancy, eating disorders, illness, medication or anyone under 18.

## The contract

Run body: `{ "task": "plan" | "adjust", "about": string, "draft": string, "log": string (adjust), "facts": string }` where `facts` is the JSON the browser engine computed (`Diet.plan(spec)` for the plan lane; `{ checkin: Diet.adjust(spec), plan: summary }` for the adjust lane). The reply is one JSON object with `lane`, `title`, `headline`, `verdict`, `summary`, `notes_on_input`, `risks`, `next_steps`, `draft_check` and the lane body (`plan_text`, `goal_fit`, `priorities`, `food_guidance`, `pitfalls` for plan; `adjustment_text`, `trend_reading`, `macro_change`, `phase_status`, `adherence_flags` for adjust).

## Sources

Derived from the agent skill @borisghidaglia/rp-diet (https://skillsafe.ai/skill/@borisghidaglia/rp-diet/, repository borisghidaglia/science-based-lifter), which indexes The Renaissance Diet 2.0 by Dr. Mike Israetel, Melissa Davis, Jen Case and James Hoffmann (Renaissance Periodization). The engine was verified against an independent R implementation written from the chapter text on a grid of 2,030 plans and 60 check-ins, with three controls that must and do fail: a Mifflin-St Jeor TDEE against the per-pound maintenance estimate, a fat-fills split without the 0.3 g/lb floor against one with it, and a 20%-of-maintenance deficit against the rate-based 3,500 kcal/lb deficit. It also reproduces the book's own worked example (155 lb, fat loss, fat near 0.39 g/lb) within the book's rounding. Not medical advice; not a substitute for a registered dietitian; not affiliated with Renaissance Periodization or the skill's author.