Your daily loop across the seven areas — grid to signal to consensus to forecast — and the refusals you will meet.
You live in the Demand Planning Suite (/app/demand-suite), and your day runs left to right across it. Start at the editable key-figure grid (/app/grid) — the runtime frame (SIQ-MOCK-15). It stacks five figures per SKU × market × period: the statistical baseline, sales input, consensus, sell-out consumed, and net. Exactly two of those five are yours to edit — sales input and consensus. The rest are the engine and the facts, and they refuse your pen on purpose.
The grid stores overrides only. The statistical row is engine evidence and cannot be typed over; sell-out consumed is a fact ("facts are not editable"); and net is DERIVED — net = max(0, consensus − sell-out consumed) — so it refuses every edit with its own reason. When you do edit a judgment cell, a basis is mandatory: no basis, nothing saved. A blank basis is refused, a negative quantity is refused ("must be ≥ 0"), and a valid override lands as rung C — "graded on actuals" — so your judgment is measured, not just recorded.
Next, read the signal (/app/signal). Multi-echelon sensing shows you the inflation and divergence between POS, distributor and primary demand that a statistical baseline cannot see. The doctrine here is strict and load-bearing: flags only — a flag never adjusts history. If the channel is stuffed or a signal diverges, SangamIQ raises a flag for a human to work; it will not silently rewrite your demand_history. Correcting the record stays a human task.
Then reconcile at consensus (/app/consensus). Your baseline meets the commercial number, and the two are never merged behind your back — the gap is shown in units AND value, never a bare percentage. Agreement is maker-checker: you cannot close your own round (maker ≠ checker), and when a different user agrees, the one audited number lands as a single consensus demand stream. Re-negotiating the same family and period updates that same stream in place — it never duplicates.
Finally, override the forecast where you genuinely know better (/app/forecast). Every override carries a reason, is audit-logged, and is graded against naive as forecast-value-add — if a simple method would have beaten you, the product tells you. And the hard floor: below nine months of history SangamIQ refuses to fit a model at all. That refusal is deliberate — a confident forecast on three months of data is a lie with extra steps.