TemporalMind — Time Series Intelligence

Upload a CSV or Excel file. TemporalMind auto-detects timestamp, value, and hierarchy columns.

📂

Drop file here or browse

CSV · Excel (.xlsx, .xls) · Max 50 MB

CSV XLSX XLS

ℹ How it works

📁 Upload CSV/Excel with any structure

🔍 Auto-detect timestamp, value & hierarchy columns

✏️ Confirm or adjust column roles below

📊 Proceed to analyze or prepare data

Single Series

date, value

Hierarchy

date, region, store, sku, qty

Full decomposition, stationarity, outlier and intermittency analysis for any series or hierarchy node.

⬆ Upload and confirm a dataset in Tab 1 first.

Last run: — nodes · — succeeded · —

0 leaves

Value Column

# Observations

Missing %

Outlier %

Mean

Std Dev

% Zeros

Demand Class

Series Length

Noise Level

🌲 Hierarchy Explorer

Demand Class

Series Length

Noise Level

Outliers %

Missing %

Observations

🌲

No results yet

Run Hierarchy Analysis to populate this table

✦ Forecast Advisor has recommendations. Apply suggested interval and seasonal period for analysis.

Match these to Forecast Prep → Interval Selection to align outlier counts.

Apply imputation

Treat zeros as missing

IQR Z-Score IsolationForest LOF

Series Analyzer applies one treatment to every detected outlier. Per-outlier review is available in Forecast Prep.

📈

Select a series or hierarchy node, then run analysis

Results will appear here - charts, decomposition, stationarity tests

Understand relationships between your dependent, independent, and event variables. Detect lead/lag effects, event lifts, and Granger-causal links.

⬆ Upload and confirm a multi-variable dataset in Tab 1 first.

Build a model-ready series with train / validation / test / holdout / future splits.

⬆ Upload and confirm a dataset in Tab 1 first.

Last run: — nodes · avg MAPE — · —

Normal Mode Default model selection

Value Column

# Observations

Missing %

Outlier %

Mean

Std Dev

% Zeros

Demand Class

Series Length

Noise Level

🗂

Configure settings, then run preparation

Timeline, model comparison and feature matrix will appear here

TimelineFORECAST VIEW 🔭 Forecast Only

Run Forecast Preparation to populate the model comparison. You can still configure input forecasts and manage events before the first run.

Run Forecast Preparation to view per-model horizon forecasts. Input Forecasts and Events stay available before the first run.

Run Forecast Preparation to view independent-variable forecasts and event effects. These inputs remain available even when the node selection changes.

Select Variables

Fallback Strategy

Forecast Methods

Specify Forecasts

1Select input variables

Choose which variables to include in the template and import

Loading columns…

2Fallback strategy

Applied when a column is missing from the uploaded file, or has gaps in the forecast horizon

When a column is missing from the uploaded file (or has gaps in the forecast horizon), apply this fallback:

Default fallback for all columns:

Per-variable overrides Leave as "— global default —" to inherit from above

3Forecast method per variable

Choose how each independent variable will be forecasted — used when generating model-based inputs

Each independent variable can use its own forecast method. A global default is applied if no per-variable override is set. Event variables skip this step — they default to zero.

Global default method:

Per-variable overrides (optional)

4Specify input forecasts

Choose how input values for the forecast horizon will be provided

METHOD 1

Auto-generate

Use methods from Step 3 to generate inputs automatically

METHOD 2

Import from file

Upload a CSV/Excel; optionally seed from Method 1 results

Inputs will be auto-generated using the forecast methods configured in Step 3. You can switch to Method 2 to import custom values, or seed a template from these results.

Seed template from Method 1

Pre-fills the template with model-generated values so you can override specific cells

📂

Drop CSV or Excel file here

File must contain a date column plus one or more input variable columns

Select events defined in the Event Manager to include as exogenous variables. They will be materialised over the data range plus the forecast horizon and merged with the event columns above.

Loading events…

Ready to export

Reconciled Forecasting

Forecast across the full hierarchy — Bottom-up, Top-down, or Middle-out — then browse and override any node.

↑ Upload and confirm a hierarchical dataset in Tab 1 first.

Model Family

Hierarchy Level

MAPE ≤

MAPE ≥

MAE ≤

RMSE ≤

Min

Max

Sort By

Run a reconciled forecast
to browse the series hierarchy

Allow negative forecast

Excel

CSV

🏗

Configure strategy and run to see reconciled forecasts

Node timeline, family context and override table will appear here

Override Forecast

Selected node override editor

🔴 Highlighted dates indicate overrides already applied.

Select baseline or reconciled cells

Ready to export

Future Forecast Preview

Reconciliation Report

Superimpose all series at a chosen hierarchy level. Analyse stability across time intervals — Day · Week · Month · Quarter · Semi-year · Year — and detect seasonal patterns, cross-series divergence, and year-over-year consistency.

⬆ Upload and confirm a hierarchical dataset in Tab 1 first.

Multi-resolution sliding-window analysis and weighted combination forecasting for sub-hourly data (10-second · 1-min · 5-min · 15-min · 30-min). Requires upload and schema confirmation first.

⚡

No session data yet

Upload a sub-hourly dataset and confirm the schema in Tab 1 first.

Column Selector

Quantity Type

Window Anchor

Zero Threshold

? Quantity type will be detected after HF scan

Frame Specifiers (number of bars per accumulation level)

Accumulation method

Forecast mode

suitable general

ROLLING WINDOW SIZE (lightweight mode only)

2 3 6 12 24

FORECAST HORIZON (STEPS)

LIGHTWEIGHT METHOD

CONFIDENCE LEVEL

Negative Forecast Allowed

ROLLING WINDOW SIZE (observations for rolling_mean)

2 3 6 12 24

FORECAST HORIZON (FRAMES)

FORECAST METHOD

CONFIDENCE LEVEL

Negative Forecast Allowed

RECONCILIATION METHOD

OLS — Ordinary Least Squares

W = I · No variance info needed · Best default when σ estimates are noisy · Fastest

MinT-Variance

W = diag(σ²) · Higher weight to historically precise levels · Uses per-level σ from Multi-Interval

MinT-Shrink (coming soon)

Shrunk sample covariance · Requires residual arrays · Not yet available

Negative Forecast Allowed

⚖️ What reconciliation does

Combination blends level forecasts with fixed weights. Reconciliation instead finds the closest coherent forecast — one where every coarse-level value equals exactly the sum of the fine-level values it covers.

              OLS:  min ‖ŷ - Sb‖² s.t. Sŷ = ŷ

              MinT: min ‖ŷ - Sb‖²_W s.t. Sŷ = ŷ

              S = summing matrix, b = base forecasts

🪜 Frequency Ladder & Reconcilability select levels ▼

—

Total Events

—

Active

—

Inactive

—

Date List

	Event Name	Type	Granularity	Start Date	End Date	Status	Actions
Loading events…

0 events selected

📊 Preview Panel Select an event from the list to preview its graph below.

From

Select an event to preview its activation graph.

Upload & Parse

Validate

Confirm & Import

📄 Download Sample File — Use this as a template. Contains one GREGORIAN and one DATE_LIST example event.

📂

Drag & drop or click to upload

.csv or .json files accepted

📦 Format

CSV — Human-friendly

Best for Excel/Sheets. Pipe-separated for multi-value fields.

JSON — Machine-friendly

Full fidelity; preferred for programmatic re-import.

🎯 Scope

All Events

Export all event definitions including inactive.

Active Only

Only events with is_active = true.

Selected Events

Only events checked in the Event List tab.

🗓 Binary Matrix — optional

Include materialised binary matrix for a date range

📄 CSV Column Reference

# events_export.csv
event_id,name, event_type,
calendar_type, granularity,
occurrence_json, recurrence_json,
date_list, before_intervals,
after_intervals, decay_intervals,
tags, description,
is_active, created_at, updated_at

🔑 Required on Import

nameUnique per tenant

event_typePOINT|LEVEL_SHIFT|TEMPORARY_CHANGE

calendar_typeGREGORIAN | DATE_LIST

granularityD | W | MS | QS | h …

occurrence_jsonJSON-encoded rule

✦ Forecast Advisor

Ready-to-use recommendations based on the ingested dataset.

No advisor run yet

Upload and confirm a dataset to generate guidance.

Frequency Comparison

Frequency	Avg obs	Avg train	Status

Recommendations

Suggested Settings

No dataset is available yet. Upload data, confirm the schema, and the advisor will populate automatically.

Advisor Unavailable