Four ways to build the same Model & performance screen, all showing the same fictional data in the same unedited state. I recommend A. Open each one, compare them, then make your selection at the end of this page.
Use the arrow keys on the letters to switch directions.
Organizing ideaRecommended
The chart is the workspace, and everything else docks to it.
From miss to activation
Read the miss in the table under the chart.
Adjust Base with the stepper beside its line end.
Follow Draft, Saved, Captured and Active along the bottom row to Activate.
Three design decisions
Hub tasks sit in the top bar, leaving the full width for analysis.
Only the editing case has a live control; the others need an explicit Edit.
Miss evidence and the lifecycle share one fixed band.
Tradeoff
The bottom band takes height, so exact weekly forecasts need a drill-down.
Labels specific to A
Miss against previous issued, Edit, All settings
Still imperfect
Case rows line up only roughly with the line ends.
The first 5 in “5 of 5” is faint.
Hub reads like a tab.
Generated type is about 16–18px, not the 13px requested.
Inputs & packages isn't shown.
Organizing idea
Each week is one column running through the chart and the table.
From miss to activation
Read the Error row directly under the gap in the chart.
Edit the 0% in the Base row.
Use the Record column and the bottom buttons to reach Activate.
Three design decisions
The left outline lists every task at all times.
The case setting lives in its own table row.
Blank cells stay blank instead of showing zero.
Tradeoff
It has the best exact values but the smallest chart, and weekly columns won't stretch to views through 2029.
Labels specific to B
Record, Previous issued, Inputs & packages
Still imperfect
The chart and table are an exact local render, at 14px, so they look smaller than the text around them.
The four chart families aren't shown.
The Record column has unused height.
Organizing idea
The process is the interface: four numbered steps above four lifecycle stations.
From miss to activation
Open 1 Diagnose.
Set the case in 2 Adjust.
Move along the Draft, Saved, Captured and Active tiles.
Three design decisions
Each lifecycle state owns its own action button.
Arrows make the required order visible.
The rail keeps Hub and Spokes visible, with shortened labels.
Tradeoff
The save-to-activate order is unmistakable, but the heavy styling competes with the data and the tiles take 30% of the height.
Labels specific to C
Full names for the rail: Model & performance, Reporting, Forecast review, v1 change visibility, Version history, Business inputs, Versions & scenarios, Approval & delivery
Still imperfect
The zero in Active V0 looks like the letter O.
Generated type is about 16–18px, not the 13px requested.
The four chart families aren't shown.
Inputs & packages isn't shown.
Organizing idea
It answers “what needs recalibration or activation now?” directly, with a queue that drives the screen.
From miss to activation
Pick an item in Needs attention.
Read the evidence and headline, then edit tokens in the sentence.
Follow Path to active.
Three design decisions
A written finding leads the screen.
Settings read as one sentence.
The evidence table follows the selected queue item.
Tradeoff
It's the clearest for a first read, but a sentence can't hold the many advanced controls, and the headline must stay factual.
Labels specific to D
Needs attention, Repeated miss, Largest weekly misses, New actuals available, Selected evidence, Path to active
Still imperfect
Green ticks reuse Upside's colour.
Generated type is about 16–18px, not the 13px requested.
The four chart families aren't shown.
Inputs & packages isn't shown.
Compare
Pick any two directions to see them side by side.
How the four directions differ
Measure
A, recommended
B
C
D
Organizing idea
Chart first; everything else docks to it
Every week is one column through chart and table
Numbered process over lifecycle stations
A queue of issues drives the screen
Navigation
Top tabs for Hub, Spokes menu
Left outline, always fully listed
Black rail with shortened labels
Two switchers plus task search
Model control
Case rows beside line ends, stepper on the editing case
Input inside the case's table row
Form inside the Adjust step
Editable tokens in a sentence
Evidence lead
Chart, then miss table
Exact values, then chart
Process state, then chart
Written finding, then chart
Type and character
Neo-grotesque, hairlines, no cards
Serif and monospaced, ruled paper
Condensed capitals, heavy rules
Humanist, soft columns, sentence case
Chart share of screen
41%
26%
39%
32%
Chart families shown
Yes
No
No
No
Weekly forecast values on screen
No, totals only
Yes
No, totals only
No, totals only
Main risk
Case rows drift from line ends as values move
Weekly columns break down on long horizons
Heavy styling competes with data at density
Sentence doesn't scale to advanced controls
Why I recommend A
It has the largest chart, at 41% of the screen.
It's the only study showing miss evidence, case settings, the four lifecycle states and all four chart families together.
Its right dock can grow into the many advanced settings without changing the layout.
Extra window width goes straight to the chart, which suits 2560×1440.
I'd also borrow two things. From B, align the W38–W42 miss table under the chart's weeks. From D, add a collapsible Needs attention list, without accept or decline actions, because those would imply a Hub approval gate on v1.
A's weakest point is exact weekly values. If those matter most on first view, B is the stronger choice. This is my judgment from static images, with no user testing.
Shared content
These are the correct labels and values for all four directions, whatever the image shows.
Context
Forecast Workspace, USD m · synthetic, Source DEMO-SRC-42, Actuals through W42, Demo as-of W42
Scope
Editing target All models, Editing case Base, Chart scope All forecast nodes (view only)
Cases
Base 0%, 630.00. Upside +3%, 648.90. Downside −3%, 611.10.
State
No unsaved changes. Save and Run disabled. Inspect enabled. Activate disabled because the capture is already active.
Actuals, W38–W42
96, 99, 97, 101, 100
Previous issued, W38–W42
98, 100, 101, 103, 104
Errors, W38–W42
−2, −1, −4, −2, −4
Added by me, not from your brief
IDs CFG-B-01 and RUN-B-01. Method Seasonality indexing. Momentum window Recent 8 weeks. Baseline-trend window 26 weeks. Derived: net −13, −2.57% (−13 ÷ 506 issued), and “5 of 5 weeks below issued”.
What I inspected and corrected
None of these is pure Higgsfield output. Every chart point comes straight from the fixture, as does B's table. I checked the remaining generated text by eye at 1728×960 and found no wrong numbers.
Defects found in generated output and how each was fixed
Study
Problem in the generated output
Fix
A
Upside and Downside steppers looked live while the editing case was Base
Regenerated with Edit links; Base alone keeps the stepper
A, B, D
The Previous issued label collided with the data lines (my reference chart caused this)
Exact chart pasted in with the label moved
A, C, D
The charts had no direct or correctly placed case labels: A had none, and C and D were redrawn with drift
Exact chart pasted in, with case labels at the line ends
B
Weeks in the chart didn't sit above the table columns
The correction attempt failed and turned W48 into W46, so I replaced the chart and table with an exact aligned block
C
Past weeks were about 83px apart and future weeks about 107px, and the label covered the W39 actual
Exact chart pasted in
D
Past weeks about 81px apart, future about 86px
Exact chart pasted in
Public references
The TradingView capture timed out and the Observable Plot page blocked my request, so neither informed these concepts.
Public references, how each was inspected and what transfers
