[GH#262] [Future] Define opt-in retention/export policy for derived local training datasets #24

New issue

Open

opened 2026-05-19 22:15:42 +02:00 by Max · 0 comments

Max commented

2026-05-19 22:15:42 +02:00

Owner

Migrated from GitHub #262
Originally created by @Bio1988 on 2026-04-23T09:43:00Z

Context

Post-MVP follow-up from the retention discussion.

The current agreed direction is:

raw/high-volume telemetry should rotate locally by default
90-day retention is a sensible default for raw telemetry/replay-heavy data
if the product ever wants local LM-assisted coaching or training-ready derived datasets, that should be explicit and opt-in

Problem

There is not yet a policy for what should happen if the app later derives compact training-oriented artifacts from local telemetry, such as:

lap summaries
section facts
coaching facts
selected replay segments
local embeddings or future LM-ready derived records

Scope

define what counts as a derived training-oriented artifact vs raw telemetry
define whether such artifacts should be retained longer than the default raw-telemetry retention window
define export/consent/visibility requirements for any local training-oriented dataset retention
ensure the default remains privacy-preserving and low-maintenance

Acceptance criteria

policy exists for opt-in retention of derived training-oriented artifacts
raw telemetry rotation remains the default behavior
no future training retention happens implicitly without clear user intent
if the decision materially affects architecture or storage policy, add an ADR

Notes

This is intentionally a Future Features issue. It should not expand the MVP retention scope beyond the agreed 90-day rotating default for raw telemetry.

Migrated from [GitHub #262](https://github.com/Bio1988/strategy-desktop/issues/262) Originally created by @Bio1988 on 2026-04-23T09:43:00Z --- ## Context Post-MVP follow-up from the retention discussion. The current agreed direction is: - raw/high-volume telemetry should rotate locally by default - 90-day retention is a sensible default for raw telemetry/replay-heavy data - if the product ever wants local LM-assisted coaching or training-ready derived datasets, that should be explicit and opt-in ## Problem There is not yet a policy for what should happen if the app later derives compact training-oriented artifacts from local telemetry, such as: - lap summaries - section facts - coaching facts - selected replay segments - local embeddings or future LM-ready derived records ## Scope - define what counts as a derived training-oriented artifact vs raw telemetry - define whether such artifacts should be retained longer than the default raw-telemetry retention window - define export/consent/visibility requirements for any local training-oriented dataset retention - ensure the default remains privacy-preserving and low-maintenance ## Acceptance criteria - policy exists for opt-in retention of derived training-oriented artifacts - raw telemetry rotation remains the default behavior - no future training retention happens implicitly without clear user intent - if the decision materially affects architecture or storage policy, add an ADR ## Notes This is intentionally a `Future Features` issue. It should not expand the MVP retention scope beyond the agreed 90-day rotating default for raw telemetry.

Max added the

labels

2026-05-19 22:15:42 +02:00