Problem / use case

With grouped CSV/JSON, every numeric column becomes its own stat type, and the UI creates one chart per stat.

For wide “competitor” tables (e.g. framework benchmarks), columns are series of the same metric, not separate charts. Example (concurrency.csv):

load,default,chi,echo,gin,goframe,httpz
100,3103.62,3103.73,...
1000,32423.67,32421.28,...

vizb bar concurrency.csv -g load -p y -o out.html

Today this yields 6 independent charts (one per framework). The desired shape is one chart where frameworks share an axis — same idea as merge’s --tag-axis / -A placing tags on n/x/y/z instead of splitting views.

Proposed solution

Flag

Target UX:

vizb bar concurrency.csv -o out.html -g load -p y -A x
# or long form:
vizb bar concurrency.csv -o out.html -g load -p y --col-axis x

• load → group dimension y (series / variants)
• numeric column names → x categories
• all values share a single stat → one chart

Naming: --col-axis (“put column names on this axis”), parallel to merge’s --tag-axis. Avoids --stat-axis because --stat already means the analytics panel.

Data transform

After parse + aggregate, when --col-axis is set, expand multi-stat points into long form:

Before (-g load -p y):

{ "yAxis": "100", "stats": [
  { "type": "default", "value": 3103.62 },
  { "type": "chi", "value": 3103.73 }
]}

After (-A x / --col-axis x):

{ "yAxis": "100", "xAxis": "default", "stats": [{ "value": 3103.62 }] }
{ "yAxis": "100", "xAxis": "chi",     "stats": [{ "value": 3103.73 }] }

Rules:

1. One output point per (input point × stat).
2. Write the column label onto the chosen dimension (Name / XAxis / YAxis / ZAxis).
3. Replace Stats with a single stat: Type left empty so JSON omits type (json:"type,omitempty" on shared.Stat). Wire format has no type key → UI sees undefined. Same Value.
4. Ensure Dataset.Axes includes the inject axis (reuse merge’s ensureInjectAxis pattern).

Chart title (UI)

Empty/omitted stats[].type still yields one chart signature. The card header must not stay blank:

• When type is empty/undefined, show Dataset.Name (from -n / default "Comparisons") as the chart title.
• Accept that the dataset name may also appear elsewhere in the UI.

Validation

Same-axis conflict must fail hard (do not overwrite or append like merge tag injection):

vizb bar concurrency.csv -g load -p x -A x -o out.html
# Error: --col-axis "x" conflicts with group dimension (already used by --group-pattern)

Implementation sketch

Aggregation stays before expand so duplicate rows still sum per column first.

Acceptance / verification

# One chart: frameworks on X, load on Y series
vizb bar concurrency.csv -g load -p y -A x -n "HTTP frameworks" -o /tmp/comp.json
# assert: stats have no type key, xAxis in {default,chi,...}, yAxis in {100,1000,5000}
# HTML: chart card title shows "HTTP frameworks"

# Classic grouped bar: load on X, frameworks as Y series
vizb bar concurrency.csv -g load -p x -A y -o /tmp/comp2.html

# Same axis as group → fatal exit
vizb bar concurrency.csv -g load -p x -A x -o /tmp/bad.html

# No group → warn + skip
vizb bar concurrency.csv -A x -o /tmp/skip.json

Out of scope

• Changing default multi-chart behavior without the flag
• UI-side “merge stats into series” without reshape
• Auto-detecting competitor tables (all-numeric wide files still need explicit -g + -A)
• Renaming merge’s -A or sharing one flag across merge and chart commands