Genesis AI — Complete File Trigger Map

User clicks Upload button → sets uploadOpen=true → UploadModal mounts

On open → POST /api/upload/reset (clears previous job)
User picks file → handleFile() runs:
① Validates extension (.xlsx/.xlsm/.xls)
② Builds FormData({ file })
③ POST /api/upload
④ Starts startPolling() — polls every 1.5 s

① Validates extension (server-side guard)
② Checks no job already running (HTTP 409 if busy)
③ Saves xlsx → _uploads/filename.xlsx
④ _find_pipeline_python() — finds Python with pandas
⑤ Spawns background thread → _run_pipeline(xlsx_path)
⑥ Returns HTTP 200 immediately

Streams subprocess stdout live to terminal
_annotate() prints step banners at key lines
After completion → copies output/*.json → react-dashboard/public/data/
Sets _job["status"] = "done"

Parses CLI args (--input, --output)
Orchestrates all pipeline steps in sequence
Calls every function below and writes all JSON outputs

Single source of truth for all thresholds.
Imported (not called) — all constants read at import time.
No magic numbers anywhere else in the pipeline.

load_source_data():
 → Opens xlsx, finds 1_Source_Data sheet header row
 → Reads Brand, Date, Channel, Region, Pack Size Group, Value, Volume, Distribution
 → Computes Price = Value / Volume
 → Parses dates, drops null rows

read_brand_config():
 → Reads 2_Brand_Config sheet
 → Returns focal brand(s) + competitors list

build_wide_pivot():
 → Called once per Channel × Region combination
 → Creates wide monthly table with focal brand + competitor columns
 → Adds: Vol_F, Price_F, Dist_F, Cat_Vol, Vol_Up, Vol_Down

run_elasticity_models():
 → Tests all OLS spec combinations per grain
 → Base + competitors + seasonal dummies
 → Selects best Adj-R² with own-price coef < 0
 → Clamps elasticity to [−6, 0]

ols(y, X):
 → numpy.linalg.lstsq solver
 → Returns betas, t-stats, p-values, R², Adj-R²

Produces 5-sheet diagnostics Excel:
 → descriptive: mean, std, CV, trend per grain
 → correlation: Pearson |r|, collinearity flags (>0.70)
 → rpi_dist: relative price index vs competitors
 → cannibalization: cross-pack volume correlations
 → summary: model-readiness pass/warn/fail per grain
Writes: output/{focal}_Step1_Diagnostics.xlsx

assign_proxies():
 → Grains with wrong sign get borrowed elasticity
 → Priority 1: interpolate adjacent packs
 → Priority 2: borrow from nearest pack same CH/Region
 → Priority 3: borrow from nearest pack any region
 → Adds Final_OwnE, IsProxy, ProxyMethod columns

compute_freq_anchors():
 → Finds dominant pack per Channel × Region by vol %
 → Returns {CH|Region: {anchor, vol_sal}}

build_grain_metrics(): vol sal, val share, MS yr25/yr24, price/ml per grain
build_model_export(): flat list of grain elasticities + metrics → models.json
build_stats_json(): brand KPIs (vol growth, avg ε, anchors, Cr value) → stats.json
print_summary(): prints grain/proxy/Adj-R² summary to terminal

→ trend.json: monthly time-series for all brands (volume, value, price, dist)
→ ms.json: focal brand market share yr25 vs yr24 per grain
→ vol_salience.json: pack vol % of focal brand total
→ val_share.json: pack value % of focal brand total
→ comp_ms.json: all brands market share yr25 vs yr24
→ vtm.json: category + per-brand volumes with MS and vol change

Called twice:
① channel="MT", sheet 3_PPA_MT → ppa_mt.json
② channel="TT", sheet 4_PPA_TT → ppa_tt.json
Per-brand PPA matrix: SKU count, MRP, price/ml, relative gradient, gross contribution %, RPI

→ interaction.json: cross-brand Pearson r of monthly volumes per CH×Region
→ growth_decomp.json: %-point contribution per grain to brand vol growth yr24→yr25
→ pgi.json: Price Gradient Index per channel yr24 vs yr25

→ recs_full.json: pricing rec cards per Channel × Pack
±5% scenarios: score, feasibility, vol/val/GC impact for each
Uses elasticity from final_df + GC% from PPA sheet

Copies all output/*.json → react-dashboard/public/data/
Also copies recs_full.json → recs.json (alias)
Sets job status to "done" so frontend poll loop stops
React now has two sources: live API + static fallback

Polls status every 1.5 s until status="done"
On done: handleReload() → calls onSuccess(filename)
Closes modal

Updates header badge to show filename
Persists filename in localStorage
Calls reload() from DataContext → triggers full data refresh

Fetches all 16 keys sequentially
Each key: tries FastAPI first → falls back to Vite static if server down
setProgress() updated per key (progress bar)
Once all done: setData(results) → all tabs re-render with fresh data

Loading screen clears once loading=false
Active tab component mounts with fresh data from context