2fe423cf47
PersonalPredictionEngine keeps a discrete probability distribution over
candidate start dates rather than a date with a margin bolted on. Everything the
product needs falls out of that one structure: the most likely date is its mode,
the window is the narrowest span holding 80% of the mass, and a "Not yet" is the
distribution conditioned on what the user just said — which is what §13 asks for
and what a date-plus-margin design cannot express at all.
It is better, and that is a number rather than an opinion. EngineComparisonTest
scores both engines over the §51 fixtures on every build:
engine MAE mean window within +/-2 window covered
baseline 1.00 2.67 7/9 7/9
personal 0.67 4.56 9/9 9/9
COVERAGE IS THE MEASURE, NOT WIDTH
The first version of that test asserted the new windows must not be wider, and
it failed. Measuring showed why the assertion was wrong: the fixtures where the
personal engine is wider are the ones that are genuinely less certain — a
history with a suspected missing period, and one with a 45-day outlier — and the
baseline answers both with a two-day window and misses. What a window promises
is that the period starts inside it. An engine keeping that promise 7 times in 9
has a broken promise, not a tight forecast. The test now asserts coverage, with
a ceiling so "some time this month" still fails.
THREE MODELLING BUGS THE TESTS FOUND
Each was found by a test failing, not by reading the code:
- Median absolute deviation alone reads a user alternating 25 and 37 as
perfectly consistent, because half her deviations are zero. Twenty
disagreeing cycles came back High, breaking §15's rule that volume alone
must never buy High confidence. Spread is now the larger of MAD and mean
absolute deviation; robustness comes from IntervalAnalysis down-weighting
what is questionable, which is a better place for it.
- Recency weighting assumes the recent past predicts the near future. For a
variable user that is false — her latest cycle is a draw from a wide
distribution, not a signal — and weighting it equally cost three days on the
§51 variable fixture. Recency is now trusted in proportion to how much her
cycles actually agree.
- A fixed one-day floor on trend detection fired on a 42-day-cycle history
whose medians differed by a single day, turning an exact forecast into a
wrong one. One day is a real trend at 28 and rounding error at 42, so the
floor is relative to the user's own spread.
WIRED THROUGH, NOT JUST TESTED
PredictionInput carries recentAbsoluteErrors, and CycleRepository feeds the
scored errors back in. Without that the app stores every error it makes and
never reads one back — measuring accuracy rather than learning from it, with the
widening happening only in a unit test. A repository test asserts the errors
actually reach the engine.
BaselinePredictionEngine stays as the control, and both engines run the same
§51 acceptance suite, so the next engine's improvement is measurable too.
108 tests, all passing. ./gradlew check green. Verified on a device.
closes #10
closes #11
closes #12
closes #14
|
||
|---|---|---|
| .githooks | ||
| app | ||
| core | ||
| docs | ||
| domain | ||
| gradle | ||
| scripts | ||
| .gitignore | ||
| README.md | ||
| build.gradle.kts | ||
| gradle.properties | ||
| gradlew | ||
| gradlew.bat | ||
| settings.gradle.kts | ||
README.md
Period
Status: Draft
Owner: _null
Last reviewed: 2026-08-18
Governs: README.md as the project-facing overview for Period
Review trigger: The first buildable feature release; any change to the stack, the
privacy promise, or how the project is built and run
A private Android period tracker that learns your cycle rather than the average person's — and never sells your data.
What it is | Status | Build and run | Repository map | Project docs | Agent notes
{ Android / Google Play | Kotlin + Jetpack Compose | Room, offline-first | Personalized prediction with an honest window | Discreet notifications | No account, no server, no data sale }
The core loop is:
Track → Learn → Predict → Remind → Learn again
What it is
A focused tracker that answers one question well: when is my next period likely to start? It records confirmed period dates, learns the individual's pattern from them, and produces a most-likely date with a window and a confidence label — never a bare date presented as fact. From that it estimates ovulation and the fertile window, and it asks discreetly whether the period started, using both yes and not yet to improve the next forecast.
Two promises hold the product up:
Your period. Better predicted. — once there is enough personal history, the app must not fall back to a generic 28-day cycle. A user recording 34, 35, 36, 34, 35 who is predicted 28 is a core product defect, not a tuning issue.
Your cycle belongs to you. We will never sell period history, fertility information, ovulation estimates, cycle predictions or personally identifiable information — to advertisers, data brokers or anyone else.
The second is enforced structurally, not remembered: health data cannot reach the advertising subsystem, because the module that will hold ad code declares no dependency on the cycle database or the prediction domain, and a guard proves it.
What it deliberately is not — no community, no pregnancy mode, no chatbot, no article feed, no symptom encyclopedia, and not contraception — is in docs/planning/PROJECT_PLAN.md.
Status
Skeleton. There is no usable app yet, and this section says so rather than listing features that do not exist.
| Surface | Status | Evidence |
|---|---|---|
| Documentation tree and tracker convention | Adopted | this tree; milestones and issues in the tracker |
| Gradle / Kotlin / Compose project | Builds | ./gradlew assembleDebug and assembleRelease both pass |
| Material 3 theme and tokens | Built | core/designsystem |
| Four-tab navigation shell | Built, placeholder screens | app/src/main/kotlin/dev/privacyllc/period/navigation/PeriodApp.kt |
| Cycle model and interval derivation | Built | domain/cycle, 5 tests |
| Prediction engine | Baseline only | domain/prediction, 12 acceptance tests from PRODUCT_PLAN §51 |
| Room persistence, DataStore | Not built | Batch 01 |
| Onboarding, Today, Calendar, Insights | Not built | Batch 03 |
| Fertility, notifications, privacy features, monetization | Not built | Batches 04–07 |
| QA | No round run | docs/qa/ClaudeReport.md |
BaselinePredictionEngine is a robust-median prototype and is not the
product. PRODUCT_PLAN §11 names a plain average as an acceptable prototype and
an unacceptable final engine; Batch 02 replaces it with the recency-weighted,
trend-aware, "not yet"-conditioned engine §12 specifies. The acceptance tests
exist now so that replacement can be shown to be better rather than merely
different.
Build and run
Prerequisites: a JDK 21 and the Android SDK with platform 37 and
build-tools 37.0.0 installed, ANDROID_HOME set.
git config core.hooksPath .githooks # per clone, every clone — see below
./gradlew test # JVM suites, no emulator, ~1s
./gradlew assembleDebug # app/build/outputs/apk/debug/
./gradlew assembleRelease # minified, unsigned
compileSdk is 37 because the current AndroidX libraries require it;
targetSdk is 36, Google Play's floor for new apps from 2026-08-31. They are
deliberately different — raising targetSdk opts the app into runtime behaviour
changes and is a tested decision, not a build fix.
git config core.hooksPath .githooks is per clone. Without it the hooks are
not installed and fail silently, which is the failure mode they exist to
prevent. post-commit pushes; see
docs/architecture/githooks/README.md.
Repository map
app/ application module, MainActivity, navigation shell
core/designsystem/ Material 3 theme and colour tokens
domain/cycle/ pure Kotlin — period records, cycle derivation
domain/prediction/ pure Kotlin — the forecast, window and confidence
scripts/ the six template scripts this project adopted
.githooks/ pre-commit, commit-msg, post-commit
docs/ product, architecture, design, QA, security, history
domain/* are kotlin("jvm") modules and cannot see the Android SDK. That is
on purpose: the prediction engine is the product, so it needs the most tests,
and tests that need an emulator are tests that do not get run.
Project docs
Detailed procedures belong in docs; open work belongs in the tracker, not in this README.
| Doc | Purpose |
|---|---|
| docs/DOC_TRUST_MAP.md | Which document owns which answer, and which source wins when records disagree. Read this first. |
| docs/planning/PRODUCT_PLAN.md | The full V1 specification — prediction requirements, screens, copy, compliance |
| docs/planning/PROJECT_PLAN.md | The short vision, the stack and its reasons, and what this is deliberately not |
| docs/WORK_CYCLE.md | What happens at the end of a piece of work |
| docs/architecture/README.md | Modules, boundaries, data shapes, migrations |
| docs/design/README.md | Tone, and the four rules that settle design arguments |
| docs/security/SECURITY.md | Threat model, the advertising boundary, logging rules |
| docs/TOOLS.md | Which script to run, and which ones can stop you |
Where the tracker is
Milestones are batches, issues are deliverables, and severity labels are exactly
P0, P1, P2 and release-blocker. Credentials come from the environment,
never from this repository:
set -a; . ~/.openclaw/docker-registry.env; set +a
python3 scripts/forgejo-issue.py list
Two traps that cost an hour each otherwise: Cloudflare 1010-blocks clients that
do not look like a browser or curl, so every request needs
User-Agent: curl/8.5.0 — forgejo-issue.py sends it and anything new must
too. And /issues returns pull requests unless type=issues is passed.
Agent notes
- Product truth comes from the code and the tracker before prose.
- Do not keep a work list in this README or anywhere else in
docs/. - Finish with docs/WORK_CYCLE.md, every time: close what you finished with the evidence, close the milestone if the batch landed, update the documents the change triggered in the same commit, push, log the entry, then reconcile and write the summary and next action.
- Nothing on privacyllc.dev writes itself except the tracker counts and the pushed docs.
- Do not claim a feature is built unless you can cite the file, test or screenshot that proves it. The status table above is the standard.
- Verify current stable versions before changing the toolchain. The ones in
gradle/libs.versions.tomlwere checked against the official sources on 2026-08-18; PRODUCT_PLAN.md asks for that check rather than for its own numbers to be trusted. - Never put a cycle date, a prediction or a fertility state into a log, an analytics event, or an ad request.