Spike: rebuild M4 review screen in Streamlit (1-day time-box) #154

Spike

Rebuild the M4 review screen in Streamlit, as a learning exercise. Decide the real frontend direction based on what the spike teaches.

Resolves the "spike, then decide" plan from #152.

Why the M4 review screen

Most interactive screen in the current frontend (csv-table-hub-main/src/components/review-table.tsx):

• Paging through stored rows (100 at a time)
• Cell edits with validation errors shown inline
• Required-reason dialog, Save disabled until a reason is entered
• Role-based conditional UI (reviewers/tenant_admin see edit controls; others see read-only)
• Row history popup with who/when/why
• Red cell tinting for invalid, blue for edited, with original value in a tooltip

If Streamlit handles this comfortably, it'll handle every simpler screen. If it strains or needs ugly workarounds, Streamlit isn't the fit.

Time-box

1 day. If a working review page isn't up by end of day two, Streamlit isn't the right tool — stop and reconsider.

Scope

• One screen (M4 review), not the whole app
• Reads from the existing FastAPI API — no new backend code
• Must support: paging, cell edit dialog with reason, exclude/include, history view, role gating (uploader sees read-only)
• Does not need to: match the exact styling, support offline-only operation, replace the React screen in production

The React implementation stays live. This spike is a parallel exploration, not a migration.

Done when

The spike yields one of these three outcomes, documented in a comment on this issue:

1. "Streamlit was fine, I prefer this." Follow-up issue filed for staged migration plan.
2. "Streamlit was limiting but Python-native felt right." Follow-up issue filed for a NiceGUI spike (or commit to Reflex).
3. "Python-native felt wrong." This issue closed, Option 1 from Decide: frontend architecture after ending Lovable relationship #152 wins (keep React code), no follow-up.

In all three cases, this issue closes.

Pre-conditions

• Python 3.11+ already on dev machine (used for FastAPI)
• pip install streamlit on the dev machine (one-time)
• Local FastAPI running (same workflow as current frontend dev)

No air-gap concerns for the spike itself.

Timing

When the product owner picks it up. No fixed date. Current priorities (as of 2026-11-02): #60 offline install dry run due 15 Nov, then this.

Related

• Decide: frontend architecture after ending Lovable relationship #152 (closed): the decision to spike before committing
• Decide: Supabase data behind the Lovable frontend - migrate, retire or confirm none #69 (closed): ended Lovable platform dependency
• 4 Oct 2026 decision: adopted Lovable frontend (superseded on platform, pending on code)
• csv-table-hub-main/src/components/review-table.tsx: the thing being spiked against