Iron & Chalk — Meal Plan Architect

Profile & Daily Macro Targets

Fill the blue cells; everything else recalculates. Phase macros come from Phase Reference, modifiers from Adjustments. Activity Multiplier is auto-computed below.

Client Information

Client scope (for hosted multi-client use)

Useful when hosting online. Each slug gets its own localStorage. URL:

Client name

Client #

Instagram

Notes

Athlete Profile

Total Body Weight (kg)

Body Fat (%)

Age (years)

Activity Multiplier (auto)

Weekly Weight Change Goal (kg)

Derived Stats

Lean Body Mass (kg)

BMR (Katch-McArdle, kcal)

TDEE (BMR × Activity)

Linear Caloric Adj (kcal/d)

TDEE-adjusted Target (kcal)

Phase Selection

Phase

Range Position within Phase

Phase Primary Goal

Individual Adjustment Modifiers

Carb Tolerance Tier

Age Bracket

Thyroid Status

Cardio, Climate & Sweat Rate — drives the Hydration tab

Do you perform cardio? Average minutes per cardio session Adds 500-1000 ml / hr of cardio Cardio days of the week (check the days you do cardio — can include rest days)

Hot/humid climate or non-A/C facility? Adds 500-1000 ml / day Performed the 60-min Sweat-Rate Test? Optional but elite. Calculates exact L/hr loss.

Baseline Phase Coefficients (g/kg of TOTAL bodyweight)

Macro	g/kg	Adjustment (g/kg)	Final g/kg	Grams/day	Kcal/day
Protein	—	—	—	—	—
Carbohydrates	—	—	—	—	—
Fats	—	—	—	—	—
Phase Cal Adj (%TDEE)	—	—	—	—

Adjustment = Carb-tolerance + Age/Thyroid + NEAT layered on top of the chosen phase. Range Position blends Low → Mid → High (Lowest..Low..Slightly Low..Mid..Slightly High..Higher..Highest).

Final Daily Targets — use these for meal building

Protein

—

Carbs

—

Fats

—

Total kcal

—

kcal

vs TDEE

—

kcal

Physiological Warning Flags

Activity Multiplier Calculator

Fill the blue inputs. The Effective Multiplier auto-feeds the Activity Multiplier (above).

1 · Inputs

Average daily steps (incl. work)

Job type

Training sessions per week

Avg minutes per session

Training intensity

2 · Energy Breakdown (kcal/day)

Resting metabolism (BMR)

Ambulatory NEAT — from steps

Occupational load — from job

Training (EAT) — daily avg

Thermic effect of food (TEF)

MAINTENANCE / TDEE

EFFECTIVE ACTIVITY MULTIPLIER →

3 · Engine Constants (advanced)

kcal per kg per 1,000 steps

TEF — % of maintenance

Metabolic damage adjustment (kcal)

Table 1 — Master Macronutrient Table (12 Phases)

All values in grams per kilogram of total bodyweight. Caloric adjustment is % of TDEE. Mid = avg(Low,High).

Phase	P Low	P High	P Mid	C Low	C High	C Mid	F Low	F High	F Mid	Cal Low	Cal High	Cal Mid	Primary Goal

Source: Sub-section 5.1, Table 1. Use Low end for conservative dosing, High for aggressive, Mid as default starting point.

Adjustment Modifier Tables

Tables 3, 4 & 5 — these layer on top of the baseline phase macros.

Table 3 — Carb Tolerance & Insulin Sensitivity

Tier (Marker Profile)	Carb Adj (g/kg)	Fat Adj (g/kg)	Protein Adj (g/kg)	Preferred Carb Sources

Table 4 — Age & Thyroid Status Modifiers

Demographic / Condition	Protein Adj	Carb Adj	Fat Adj	Notes

Note on N.E.A.T.

N.E.A.T. is now fully driven by the Activity Multiplier Calculator below your profile (steps + job type). The old Table 5 has been retired — the calculator already auto-translates your daily movement into a multiplier that TDEE is multiplied by, so a separate N.E.A.T. tier table is redundant.

Weekly Carb Cycling Planner

Modeled on Table 7 — Joe Bennett / Neil Hill split. Edit the blue carb % cells; protein & fats compute. The weekly average auto-compares to your daily target.

Daily target (kcal)

—

Daily protein (g)

—

Daily carbs (g)

—

Daily fats (g)

—

Day	Training Split	Carb %	Protein (g)	Carbs (g)	Fats (g)	Kcal	Phys. Goal / Notes
WEEKLY AVG			—	—	—	—	—
vs TARGET (deviation)			—	—	—	—	—

Carb % > 100% = above-baseline training day · < 100% = rest day. Fats are auto-calculated to land near total daily target kcal.

Carb % Rhythm — see the week at a glance

Daily Meal Builder

Pick a weekday from the dropdown — the target row pulls that day's macros from the Weekly Cycle Plan. Daily Target (flat) uses the single daily target from Profile. Choose foods from the dropdowns and enter raw/dry gram weights; macros auto-fill and validate against the target.

Switch weekday

▾

★ Favorites

No favorites yet — click ★ on any food in the search list below to add it here.

Target P

—

Target C

—

Target F

—

Target kcal

—

kcal

TDEE-Adj kcal

—

kcal

—

#	Protein (g)	Carbs (g)	Fats (g)	Kcal	Notes
DAILY TOTAL	—	—	—	—	—
DEVIATION vs TARGET	—	—	—	—	—
DEVIATION VS TDEE-adjusted Target (kcal)	—	—	—	—	—
VALIDATION	—	—	—	—	—
Macro split (% kcal)	—	—	—	—	—

If a macro shows ✗ ADJUST, change portion weights in 10–25 g increments on the right macro source. Carb sources move carbs, oils/nuts move fat, protein sources move protein. Target: all 4 columns green (within ±10 g P/F, ±20 g C, ±50 kcal).

Daily Nutrient Summary — % of target covered by today's foods

Show only low (< 50%) Target

Show in grams / mcg

%DV is calculated against standard adult reference intakes (US FDA label values). This is a quick coverage check — it doesn't track absorption, interactions, or your individual needs.

Saved Meal Plan

Stripped, read-only view of every day you marked as Saved from the Meal Builder. Empty / unused food slots are hidden — only what you actually eat shows up.

View

Shopping List

Aggregated raw/dry gram weights for every food on the days you select, with a 10% buffer to absorb scale error, spillage, and last-minute cooking loss. The list is sorted by total quantity (largest first).

Scope

Food	Total (raw g)	+10% Buffer	Buy (g)	Buy (kg / pack)	kcal	Tier / Category
TOTAL	—	—	—	—	—	—

The "Buy" column is the total raw/dry weight with the 10% buffer applied — that's what you should put on the scale at the store. "Buy (kg / pack)" rounds up to the nearest practical unit. Whole eggs are auto-converted from grams to count (50 g raw ≈ 1 large egg). Day pills with ● have saved data.

Food Database — Macros per 100 g (RAW / DRY / UNCOOKED)

~90 common foods synthesized from Efferding (Vertical Diet) digestibility-first, Aceto carb hierarchy, Palumbo fat sources. Type exact names in the Meal Builder dropdowns.

0 foods

Food	Protein	Carbs	Fats	Fiber	Kcal	Tier / Category	Allergens

Macros per 100 g, raw / dry / uncooked weights (USDA FoodData Central + bodybuilding standards). Cooking only changes water weight, not the macro ledger here.

Table 2 — S/A/B/C/D Tier Rankings by Phase

Sanity-check overlay on your phase choice. S/A-tier is your defaults; B is situational; C/D should be avoided. Filter by phase or edit any cell to customize.

Phase	S-Tier (Optimal)	A-Tier (Highly Effective)	B-Tier (Situational)	C / D-Tier (Avoid)

Click any cell to edit. Changes persist to localStorage and replace the default for your client.

Warning Flags & Health-First Checkpoints

Hard physiological thresholds and the bi-annual bloodwork panel. Bloodwork overrides any macro plan.

Warning Flags — Hard Physiological Limits

Health-First Checkpoints — Bi-Annual Biomarker Panel

Panel / Marker	What it tells you / threshold to act on

Evidence Ratings

Prescription	Evidence Level & Source

Hydration Protocol

Strict application of Sub-Section 16.1: Daily Hydration Protocols. LBM-based baseline + training + cardio + climate + protein clearance, distributed hour-by-hour.

Rest Day Target

—

L/day

Training Day Target

—

L/day

Hourly Sweat Rate

—

L/hr (if tested)

Inputs

Drives the calculations below

Do you perform cardio? Avg minutes per cardio session Adds 500-1000 ml / hr of cardio Cardio days of the week (check the days you do cardio)

Hot / humid climate or non-A/C facility? Adds 500-1000 ml / day Did you perform the 60-min Sweat-Rate Test? Optional but elite. Calculates exact L/hr loss.

Calculation Breakdown

Show your work

Enter inputs to see the breakdown.

Day-by-Day Breakdown — per-week targets, dynamic from your cycle + cardio days

Day	Activity	Carb %	Cardio?	Target (L)

Daily Schedule

Front-loaded morning, peri-workout peak, nighttime taper.

Time	Volume	Additives	Rationale

Quality & Longevity

Non-negotiable elite protocol

Progress Tracker

Weight, body fat, blood pressure, measurements, sleep, steps, and blood work — all time-stamped, all charted, all yours.

Body Weight (kg)

Date Weight (kg)

Body Fat (%)

Date BF %

Blood Pressure

Date Sys Dia

Sleep (hours / night)

Date Hours

Daily Steps

Date Steps

Measurements (cm) — custom fields

Blood Work — custom fields

Progress Photos — front / side / back by date, AI visual comparison

Session date

Label (optional)

Front

no photo

Side

no photo

Back

no photo

Photo Sessions — saved photos by date

Visual Comparison — pick two sessions, the model judges direction

Baseline (earlier)

Current (later)

Bodybuilding Meal Plan Architect

Operationalizing Sub-section 5.1 — Macronutrient Prescriptions (12-phase periodization model).

📸 Progress Photos — local AI vision setup

Yes, this is possible. Here's how the picture tracker is wired and how to plug in a local vision model for real "is the client going in the right direction?" judgment.

What ships by default (zero setup, fully offline)

Upload front / side / back photos by date — auto-resized to 400×600 to keep localStorage lean.
Each photo gets a 64-bit dHash perceptual fingerprint computed in pure JS at upload time.
Comparison: Hamming distance between the two sessions' hashes, per pose. Returns 0–64 (0 = identical, 64 = completely different) and a similarity %.
Verdict: "significant visible change" / "modest" / "minimal" based on the average distance. Detects change, not direction.

Optional upgrade: real LLM judgment via local Ollama

This gives you actual coaching-grade text — e.g. "Front: more shoulder separation and visible upper-ab outline. Side: lower-ab fat appears reduced. Overall trending toward the goal."

🍥 Easiest path: the bundled `iron-chalk` launcher (no OLLAMA_ORIGINS setup needed)

The iron-chalk binary included in the zip is a single self-contained executable that:

Serves the app on http://localhost:8765/ with no setup
Proxies /api/* calls to Ollama on :11434 and injects the CORS headers automatically — so you never need to set OLLAMA_ORIGINS
Probes Ollama on launch and shows whether your models are loaded
Auto-opens your browser

Just double-click the binary. No terminal, no script, no flags.

Steps to get Qwen-VL judging your photos

Step 1. Install Ollama (single binary, no Docker) and start it however you like (Launchpad app on Mac, just running on Linux, system service on Windows).
Step 2. Pull your vision model. With 16 GB VRAM, the q8-quantized 7-8B tier is sweet:

# Recommended for 16 GB VRAM
ollama pull qwen3-vl:8b            # ~8.6 GB, newest Qwen3-VL — DEFAULT in the app
ollama pull qwen2.5vl:7b-q8_0      # ~8.5 GB, previous-gen Qwen 2.5 VL 7B
ollama pull llama3.2-vision:11b-q8_0  # ~11 GB, Meta's 11B vision
ollama pull minicpm-v:8b          # ~8 GB, very strong for the size

# Lighter (faster, less VRAM)
ollama pull llava:7b              # ~4.7 GB, fast
ollama pull qwen2.5vl:3b           # ~3 GB, tiny but capable

Step 3. Double-click the iron-chalk binary. It will:

Open your browser to http://localhost:8765/
Probe Ollama and print which models are loaded
Stay running in the background — close the terminal to stop

Step 4. In the app, go to Progress → Progress Photos. The Vision provider is set to Ollama by default. The URL field shows /api (relative — this routes through the launcher's CORS proxy). The model is qwen3-vl:8b. Click Compare on any two saved sessions and the AI verdict appears below the perceptual-hash verdict.

If you'd rather run the app from the raw HTML/JS/CSS (no launcher)

You'd need to either set the CORS env var on Ollama, or serve the app and Ollama from the same origin. The launcher's CORS-proxy approach is the cleanest "no flags" option. If you do want to use OLLAMA_ORIGINS=*:

# Mac / Linux (run Ollama manually in a terminal)
OLLAMA_ORIGINS="*" ollama serve

# Windows (PowerShell)
$env:OLLAMA_ORIGINS="*"; ollama serve

# Windows (CMD)
set OLLAMA_ORIGINS=* & ollama serve

# macOS launchd service: edit ~/Library/LaunchAgents/ollama.plist
# Add <key>OLLAMA_ORIGINS</key><string>*</string> to EnvironmentVariables

# Linux systemd: edit /etc/systemd/system/ollama.service
# [Service]
# Environment="OLLAMA_ORIGINS=*"
# sudo systemctl daemon-reload && sudo systemctl restart ollama

Then change the Ollama URL field in the app from /api to http://localhost:11434/api for a direct call.

Building the launcher for your platform

Linux x64 binary is pre-built and bundled. For Mac / Windows, run this once (requires Bun):

# Mac (Apple Silicon)
bun build --compile --target=bun-darwin-arm64 --outfile=iron-chalk launcher.ts

# Mac (Intel)
bun build --compile --target=bun-darwin-x64  --outfile=iron-chalk launcher.ts

# Windows
bun build --compile --target=bun-windows-x64 --outfile=iron-chalk.exe launcher.ts

# Linux (other archs)
bun build --compile --target=bun-linux-arm64 --outfile=iron-chalk launcher.ts

Step 4. On the Progress tab, in the Progress Photos card header, set the dropdown to "Ollama (LLaVA / Llama-3.2-Vision)". Then on two saved sessions, click Compare. The AI verdict is appended below the perceptual-hash verdict.

How the AI call works

When you click Compare with the Ollama provider enabled, the app sends the two front and two side photos to POST /api/generate as base64, along with this prompt:

You are a physique coach comparing two progress photos. Image 1 = earlier, Image 2 = later (front, side, front, side).
For each image pair note: any visible changes in muscle definition, body fat, posture, or proportions.
Then give a 2-3 sentence verdict: is the client trending in the right direction (e.g., leaner, fuller, more defined) or not?
Be honest, specific, and avoid generic praise. Perceptual-hash similarity scores: front=…% side=…%

The response is rendered as a separate verdict card so you can see both the local signal (perceptual hash) and the AI's qualitative read side by side.

Tuning the prompt for Qwen 2.5 VL

Qwen 2.5 VL is a chat model — it responds well when the prompt is framed as a system + user role. If you want crisper output, you can edit queryOllamaVision() in app.js to send the Ollama /api/chat endpoint with a messages: [{role:"system", content:"…"}, {role:"user", content:"…", images:[…]}] payload instead of /api/generate. Qwen2.5-VL's chat format gives notably more grounded, less generic answers on physique comparisons.

Privacy

The perceptual-hash path is 100% browser — no image leaves your device. The Ollama path is also fully local (sent to localhost), so no image is ever sent to a third-party server.

Going further (when you want it)

Switch the app to Ollama's /api/chat endpoint and send Qwen-style system + user messages for noticeably better coaching prose.
Add the client's current Phase (cutting / lean bulk / maintenance) from the Profile as a system message, so the model knows which direction "good" means.
For pose-invariant similarity, layer in a small CLIP-ViT-B/32 in the browser via @xenova/transformers (~300 MB, fully local). Returns a 512-d embedding per image; cosine distance is more robust to small pose / lighting drift than dHash.
Long-term: a small fine-tuned regression head on the embeddings can estimate % body-fat change or lean-mass delta from a pair of images, going beyond "change / no change".

Sheets & Purpose

Section	Purpose & How to Use

Color Legend

User input — change these to drive everything else
Formula output — do not overwrite; will break dependents
Lookup value or assumption — verify it matches your phase
Pass indicator — within target window
Warning — outside safe physiological range

Quick-start checklist

Open Profile & Targets. Fill in the blue cells under Athlete Profile.
Pick your phase and range position. Then fill the Activity Calculator (steps, job type, sessions/week, minutes/session, intensity).
Pick your carb tolerance tier, age bracket, and NEAT level. The Final Daily Targets are your numbers.
Review the warning flags. If anything turns red, revisit your phase choice or your stats.
Open Weekly Cycle to design your carb cycle. Open Meal Builder to build a sample day and validate.
Rebuild the Meal Builder each time you swap a meal — the validation indicator catches drift.

Profile & Daily Macro Targets

Client Information

Athlete Profile

Derived Stats

Phase Selection

Individual Adjustment Modifiers

Cardio, Climate & Sweat Rate — drives the Hydration tab

Baseline Phase Coefficients (g/kg of TOTAL bodyweight)

Final Daily Targets — use these for meal building

Physiological Warning Flags

Activity Multiplier Calculator

1 · Inputs

2 · Energy Breakdown (kcal/day)

3 · Engine Constants (advanced)

Table 1 — Master Macronutrient Table (12 Phases)

Adjustment Modifier Tables

Table 3 — Carb Tolerance & Insulin Sensitivity

Table 4 — Age & Thyroid Status Modifiers

Note on N.E.A.T.

Weekly Carb Cycling Planner

Carb % Rhythm — see the week at a glance

Daily Meal Builder

Daily Nutrient Summary — % of target covered by today's foods

Saved Meal Plan

Shopping List

Food Database — Macros per 100 g (RAW / DRY / UNCOOKED)

Table 2 — S/A/B/C/D Tier Rankings by Phase

Warning Flags & Health-First Checkpoints

Warning Flags — Hard Physiological Limits

Health-First Checkpoints — Bi-Annual Biomarker Panel

Evidence Ratings

Hydration Protocol

Inputs

Calculation Breakdown

Day-by-Day Breakdown — per-week targets, dynamic from your cycle + cardio days

Daily Schedule

Quality & Longevity

Progress Tracker

Body Weight (kg)

Body Fat (%)

Blood Pressure

Sleep (hours / night)

Daily Steps

Measurements (cm) — custom fields

Blood Work — custom fields

Progress Photos — front / side / back by date, AI visual comparison

Photo Sessions — saved photos by date

Visual Comparison — pick two sessions, the model judges direction

Bodybuilding Meal Plan Architect

📸 Progress Photos — local AI vision setup

What ships by default (zero setup, fully offline)

Optional upgrade: real LLM judgment via local Ollama

🍥 Easiest path: the bundled iron-chalk launcher (no OLLAMA_ORIGINS setup needed)

Steps to get Qwen-VL judging your photos

If you'd rather run the app from the raw HTML/JS/CSS (no launcher)

Building the launcher for your platform

How the AI call works

Tuning the prompt for Qwen 2.5 VL

Privacy

Going further (when you want it)

Sheets & Purpose

Color Legend

Quick-start checklist

🍥 Easiest path: the bundled `iron-chalk` launcher (no OLLAMA_ORIGINS setup needed)