Lattice

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build the review and one-to-one tooling I actually need instead of Lattice — and understand what makes this ALMOST.

The forms are a fortnight. **The chasing is the product**: a cycle that opens, reminds, escalates and closes without anybody administering it. And the benchmarks — how your engagement compares to other companies — need other companies, which you do not have.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode
- A worker for the reminders
- Caddy in front

THE DATA MODEL
- people: id, name, email, manager_id, team_id, started_at, is_active
- cycles: id, name, kind, opens_at, closes_at, status, template_json — self, manager, peer, upward
- participants: id, cycle_id, subject_id, reviewer_id, relationship, status, submitted_at
- responses: id, participant_id, question_key, value_text, value_num, submitted_at
- one_to_ones: id, manager_id, report_id, scheduled_for, agenda_json, notes_private, notes_shared, completed_at
- talking_points: id, one_to_one_id, author_id, body, carried_from_id, resolved_at
- goals and check-ins, if you want them here rather than in a separate tool
- surveys, survey_responses — with the anonymity rules below

THE CYCLE, WHICH IS THE PART THAT MUST RUN ITSELF
- Opens on a date, closes on a date, and between them the worker chases: a reminder at the start, one at the halfway point, one before closing, and one to the manager listing who has not submitted
- **At most one message per person per day.** A cycle that nags four times is a cycle people resent and complete carelessly
- Nobody sees anybody else's response until they have submitted their own. Without that, the first opinion anchors the rest
- A manager cannot close a review until every required reviewer has submitted or been explicitly excused, and the excuse is recorded
- Extensions are a decision with a name on them, not a silent slip

ONE-TO-ONES, WHICH ARE THE PART PEOPLE ACTUALLY USE
- A shared agenda both sides can add to before the meeting, and private notes each side keeps to themselves
- **An unresolved talking point carries to the next meeting automatically.** That single behaviour is what stops a one-to-one becoming a status update, and it is a foreign key
- A history of what was discussed, which is what makes a review write itself rather than being invented from memory in an afternoon

ANONYMITY, WHICH IS EARNED RATHER THAN IMPLEMENTED
- If a survey is anonymous, **do not store the respondent identifier at all** — not encrypted, not hashed, not in a column somebody could join. Store only that the person responded, in a separate table, so you can chase without being able to link
- Never show a breakdown for a group small enough to identify somebody. Set a minimum — five is common — and suppress below it, visibly, with an explanation rather than a blank
- Free-text answers are where anonymity actually breaks, because a phrase identifies a person. Say so where somebody is typing
- People will assume you can see it anyway. The only thing that changes that is the design being explainable in one sentence

CALIBRATION
- Ratings across a department shown as a distribution per manager, before anything is final
- Not a forced curve — that is a policy decision and a contested one — but the distribution visible, so a conversation can happen

OPERATIONS
- .env: DATABASE_PATH, BASE_URL, SESSION_SECRET, SMTP_URL, OIDC_*
- Migrations on boot, each once; nightly backup off the machine
- Access is sensitive: a manager sees their own reports, an administrator sees everything, and every read of a review is logged

WHAT MATTERS MOST
The chaser and the anonymity design. Build the worker that runs a cycle to completion, and make the anonymous survey genuinely unlinkable — because the first time somebody believes it is not, nobody answers honestly again.