JSON vs SQLite vs MongoDB Scale Advice

JSON vs SQLite vs MongoDB Scale Advice | Generated by AI

Home 2026.05

Question: Should I use MongoDB or SQLite (instead of JSON files) for the tt timesheet verification tool?

Answer:

Neither. Keep the JSON cache. Your own README says it best: “no database needed at this scale.”

Here’s why, specifically for tt:

What `tt` actually stores

contractors: ~1-20
weeks per contractor: ~4-5
punch days per contractor: ~20-31

That’s maybe 500 records total per month, read once per run, written once per extraction. This is config-file scale, not database scale.

Why JSON wins here

No relations to enforce — client_weeks and punch_days are denormalized per contractor. No cross-contractor joins ever happen.
The “foreign key problem” doesn’t apply — each contractor_a.json is self-contained. You’re not linking rows across files.
Human-editable — you explicitly use jq to inspect/fix extracted data. A database breaks this workflow.
LLM extraction output is naturally JSON — you’d just serialize to DB and deserialize back out for zero gain.
--force re-extraction replaces the whole record — atomic file replace is simpler than DELETE + INSERT.

When you’d actually need a DB

Condition	SQLite	Mongo
50+ contractors, querying across all	✅	✅
Historical month-over-month trend analysis	✅	✅
Multiple users writing concurrently	✅	✅
Flexible schema per contractor (nested, variable)	—	✅
You need `GROUP BY month, contractor` in SQL	✅	—

None of these apply to tt right now.

If you did outgrow JSON → SQLite, not MongoDB

MongoDB would be overkill (separate server process, no ACID by default pre-4.0, operational overhead). If you ever needed querying across contractors or historical runs, SQLite is the right step:

CREATE TABLE contractor_weeks (
    contractor TEXT,
    week_start DATE,
    total_hours REAL,
    project_hours REAL,
    status TEXT,
    extracted_at TIMESTAMP
);

Then SELECT contractor, AVG(project_hours/total_hours) FROM contractor_weeks GROUP BY contractor — actually useful for trend reports.

Verdict

tt scale:  JSON ✅   SQLite (overkill)   MongoDB (way overkill)

The JSON design is correct for this tool. The previous conversation about JSON-as-DB applies when you have relational data with integrity requirements — tt’s data is flat, isolated per contractor, and LLM-generated. JSON is the right call.

Back Donate