formatools · stepango · Aug 2, 2026 · Aug 2, 2026
diff --git a/README.md b/README.md
@@ -64,7 +64,7 @@ systems like Buck and Bazel.
 15. [Bazel sample (experimental)](bazel-sample/) — minimal `kt_jvm_*` workspace exercising forma-core matrix + `impl` ↛ `impl` (F-042)
 16. [**Progressive examples + agent skills**](docs/PROGRESSIVE-EXAMPLES.md) — feature-by-feature ladders (`examples/`) + coding-agent skills (F-050; KMP skill `forma-kmp-targets`)
 17. [**Test coverage**](docs/TEST-COVERAGE.md) — JaCoCo on `plugins/`, happy-path LINE ≥60% gate
-18. [**Dogfood: Now in Android**](docs/DOGFOOD-NIA.md) — real-world multimodule OSS migration map + matrix findings (F-115)
+18. [**Dogfood: Now in Android**](docs/DOGFOOD-NIA.md) — real-world multimodule OSS migration map + matrix findings (F-115) · [**case study**](docs/DOGFOOD-NIA-CASE-STUDY.md) (Phase D before/after)
 
 Configuration made easy:
 

diff --git a/TICKETS.md b/TICKETS.md
@@ -171,12 +171,12 @@ composition stays at Android/JVM roots. iOS/JS/Wasm = later phase.
 External multimodule OSS validation of meta-build axioms (structure over configuration,
 type-owned plugins, attrs-only call sites, closed matrix). **Not** another in-repo sample.
 
-Open coding: **F-115** (`in_progress`, **`priority: now`**, **cron may continue**). **F-094** Portal remains `blocked` (human).
-4h workers: pick F-115 **Phase D** case study (or residual dogfood polish) while gated; if no ticket has `priority: now` / `cron may continue`, respond `[SILENT]` (audit 2026-07-29 gate — interactive “Go ahead” alone is not enough).
+Open coding: **none gated** after F-115 done. **F-094** Portal remains `blocked` (human).
+4h workers: if no ticket has `todo`/`in_progress` with **`priority: now`** / **`cron may continue`**, respond `[SILENT]` (audit 2026-07-29 gate — interactive “Go ahead” alone is not enough). Do not invent tickets.
 
 | ID | Status | Title | Notes |
 |----|--------|-------|-------|
-| F-115 | in_progress · **priority: now** · **cron may continue** | Dogfood Now in Android under Forma | Primary OSS target [`android/nowinandroid`](https://github.com/android/nowinandroid). Design: [`docs/DOGFOOD-NIA.md`](docs/DOGFOOD-NIA.md). **Phase A+B+C done through F27 library flavors:** Hilt/Room/Firebase/Proto DS + features + WM + binary flavors + designsystem + core:ui + network + analytics/notifications + **library `productFlavors` + `.forProductFlavor` deps** (F27). Findings F16–F31 (F27 closed). **Next:** Phase D case study. No `androidLibrary`. |
+| F-115 | done | Dogfood Now in Android under Forma | Primary OSS target [`android/nowinandroid`](https://github.com/android/nowinandroid). Design: [`docs/DOGFOOD-NIA.md`](docs/DOGFOOD-NIA.md). Case study: [`docs/DOGFOOD-NIA-CASE-STUDY.md`](docs/DOGFOOD-NIA-CASE-STUDY.md). **Phase A–D complete:** external mavenLocal spike green (`assembleDemoDebug`/`ProdDebug`); Hilt/Room/Firebase/Proto DS + features + WM + flavors (F7/F27) + designsystem/core.ui/network/analytics/notifications; findings F1–F31; module scripts **−59%** LOC vs upstream. No `androidLibrary`. Optional parity leftovers = new tickets only. |
 
 ## Backlog (lower priority / historical GitHub)
 

diff --git a/docs/DOGFOOD-NIA-CASE-STUDY.md b/docs/DOGFOOD-NIA-CASE-STUDY.md
@@ -0,0 +1,262 @@
+# Case study: Now in Android → Forma (F-115 Phase D)
+
+**Date:** 2026-08-02  
+**Ticket:** F-115  
+**Upstream:** [android/nowinandroid](https://github.com/android/nowinandroid) @ `7d45eae` (local pin)  
+**Forma consumer:** `/Users/claw/work/nowinandroid-forma/forma-spike`  
+**Forma plugins:** `mavenLocal` **`0.1.3-NIA`** (external — **no** `includeBuild` of Forma plugins, **no** NiA `build-logic`)  
+**Living design / findings:** [`DOGFOOD-NIA.md`](DOGFOOD-NIA.md)
+
+This write-up answers: *does Forma’s meta-build model hold on a real multi-module Google reference app, and what does the call-site surface look like compared with convention-plugin Gradle?*
+
+---
+
+## 1. Executive summary
+
+| Claim | Evidence |
+|-------|----------|
+| **Structure over configuration works at NiA scale** | 36 role-typed modules + binary; every module is one DSL entry (`hiltImpl`, `roomAndroidUtil`, `uiLibrary`, …). Zero raw `android { }` blocks and zero `.withPlugin` in the spike. |
+| **Convention soup collapses to types** | NiA **15** convention plugin classes + **~1.9k LOC** `build-logic` Kotlin → spike **5** thin Path A/B files (**401 LOC**) in local `forma-defs` + project-global classpath. |
+| **Module build files shrink hard** | All module `build.gradle.kts` **1476 → 600 LOC** (~**59%** fewer lines), same order of module count (36 upstream product modules vs 36 spike targets + root build). |
+| **Closed matrix forced better boundaries** | Findings F1–F3, F5, F8, F19, F29, F31 closed *by re-slicing* (nav ports, res modules, domain-facing features, `composeWidget` DTOs) — not by weakening `impl` ↛ `impl` or restoring `androidLibrary`. |
+| **Type-owned plugins scale** | Hilt, Room, Firebase, Protobuf, flavors are type/rule-owned; call sites pass attrs (`packageName`, deps, `productFlavors`, versions). |
+| **External consumer path is real** | `:binary:assembleDemoDebug` + `:binary:assembleProdDebug` **BUILD SUCCESSFUL** (945 tasks); APKs ~22 MB each (`0.17.0-nia-forma-library-flavors`). |
+
+**Not claimed:** full NiA feature/test/tooling parity (Roborazzi, baseline profiles, dependency-guard, managed devices, Spotless, Retrofit prod network, screenshot tests). Phase B–C deliberately scoped a vertical product graph that exercises the meta-build axioms.
+
+**Upstream PR:** **No** — dogfood stays on the fork unless Google asks.
+
+---
+
+## 2. Method
+
+1. **Phase A** — inventory upstream modules, convention plugins, project edges; map roles → Forma types; record matrix stress findings (`DOGFOOD-NIA.md`).
+2. **Phase B** — external spike from empty tree; publish Forma `0.1.3-NIA` to mavenLocal; attrs-only vertical slice (`assembleDebug`).
+3. **Phase C** — expand cores/features/plugins/flavors against the live matrix; engine gaps fixed only when dogfood proved them (KSP companions, `androidUtilTarget` / `libraryTarget`, library product flavors, …).
+4. **Phase D** — this document: LOC / plugin / call-site metrics from the pinned trees on the worker host (2026-08-02).
+
+Commands used for metrics (re-runnable):
+
+```bash
+# Module build scripts (exclude build-logic / build dirs)
+find "$UP" -name 'build.gradle.kts' -not -path '*/build/*' -not -path '*/build-logic/*' \
+  -not -path '*/.gradle/*' | xargs wc -l | tail -1
+find "$SP" -name 'build.gradle.kts' -not -path '*/build/*' -not -path '*/.gradle/*' \
+  | xargs wc -l | tail -1
+
+# Convention vs forma-defs
+find "$UP/build-logic" -name '*.kt' -not -path '*/build/*' | xargs wc -l | tail -1
+find "$SP/forma-defs" -name '*.kt' | xargs wc -l | tail -1
+```
+
+`$UP` = `/Users/claw/work/nowinandroid`, `$SP` = `/Users/claw/work/nowinandroid-forma/forma-spike`.
+
+---
+
+## 3. Before / after numbers
+
+### 3.1 Graph size
+
+| Metric | Upstream NiA | Forma spike |
+|--------|--------------|-------------|
+| Product `build.gradle.kts` files | **36** | **37** (36 targets + root `build.gradle.kts`; +`forma-defs` composite separate) |
+| Settings LOC | 91 | 105 (flat `include` recipe — finding **F9**) |
+| Gradle wrapper | 9.4.0 | 9.6.1 (Forma consumer pin) |
+| Project-edge refs | ~106 `projects.*` | ~111 `target(` (explicit; no convention-injected hidden edges) |
+
+Module counts are intentionally comparable: spike covers the main product graph (6 features × api/res/impl, cores, sync, root, binary), not benchmarks/lint/catalog apps.
+
+### 3.2 Configuration surface (LOC)
+
+| Surface | Upstream | Spike | Δ |
+|---------|----------|-------|---|
+| All module `build.gradle.kts` | **1476** | **600** | **−59%** |
+| Org build logic (`build-logic` `*.kt` / `forma-defs` `*.kt`) | **1917** (28 files, 15 `*ConventionPlugin*`) | **401** (5 Kotlin files) | **−79%** |
+| APK root module | `app` **152** LOC | `binary` **45** LOC | **−70%** |
+| Feature For You impl | **43** LOC | **32** LOC | still smaller; spike lists deps explicitly (no convention force-deps) |
+| `core:data` | **41** LOC | **17** LOC | **−59%** |
+| `core:database` | **36** LOC (4 plugin aliases + `android {}`) | **11** LOC (`roomAndroidUtil` only) | **−69%** |
+| `core:designsystem` | **48** LOC | **26** LOC | **−46%** |
+| `core:analytics` | **31** LOC | **20** LOC (+ flavor source sets outside this file) | **−35%** |
+
+Notes:
+
+- Upstream LOC includes AOSP license headers (~15–20 lines/file). Even stripping those, module scripts remain longer because each file still has `plugins { }`, `android { namespace = … }`, and often test/tooling blocks.
+- Spike **never** opens `plugins { }` or `android { }` in module scripts — **0** raw `android {` blocks, **0** `.withPlugin` occurrences under the spike tree.
+- Spike **27** single-method target declarations across module builds (one role entry per module). Upstream **~113** `alias(`/`id(` plugin-apply lines across product modules alone (before counting what each convention applies internally).
+
+### 3.3 What replaced `build-logic`
+
+| NiA convention theme | Forma home in spike |
+|----------------------|---------------------|
+| `android.library` / feature api/impl / compose | Role types: `api`, `hiltImpl`, `androidRes`, `uiLibrary`, `composeWidget`, `hiltAndroidUtil`, JVM `library` |
+| `hilt` | Path A `forma-defs/HiltTargets.kt` → `hiltImpl` / `hiltApp` / `hiltAndroidUtil` / `hiltBinary` |
+| `android.room` | Path B `RoomAndroidUtil.kt` → `roomAndroidUtil` (re-registers Hilt — **F14**) |
+| `android.application.firebase` | Path A `FirebaseBinary.kt` → `hiltFirebaseBinary` (stacked registers — **F16**) |
+| flavors demo/prod | Shared `FormaProductFlavor` on binary **and** libraries (**F7/F27**); `NiaFlavors.kt` list helper |
+| protobuf / datastore-proto | Path B `ProtobufLibrary.kt` → `protobufLibrary` + `libraryTarget` |
+| jacoco / roborazzi / baseline / spotless / oss / dependency-guard | **Out of scope** (non-structure tooling) |
+
+Classpath plugins live **once** in settings/`androidProjectConfiguration(extraPlugins=…)` — never as per-module identity.
+
+### 3.4 Verify bar (host, 2026-08-01 cascade tip)
+
+```text
+forma-spike$ ./gradlew :binary:assembleDemoDebug :binary:assembleProdDebug
+BUILD SUCCESSFUL in … (945 tasks)
+APK demo ~21.8 MB  binary/build/outputs/apk/demo/debug/binary-demo-debug.apk
+APK prod ~21.9 MB  binary/build/outputs/apk/prod/debug/binary-prod-debug.apk
+versionName 0.17.0-nia-forma-library-flavors
+```
+
+Plugins published: `bash scripts/publish-local.sh 0.1.3-NIA` from the Forma repo.
+
+### 3.5 Configuration-time performance
+
+No same-machine cold `--profile` pair was locked for upstream NiA vs spike in this Phase D pass (different wrappers, SDKs, and composite `forma-defs` confound absolute seconds).
+
+**Recipe** (from [`CONFIGURATION-PERFORMANCE.md`](CONFIGURATION-PERFORMANCE.md)) if a future worker wants a pair:
+
+```bash
+cd /Users/claw/work/nowinandroid-forma/forma-spike
+./gradlew --stop
+rm -rf .gradle/configuration-cache
+./gradlew help --no-configuration-cache --profile --offline
+# build/reports/profile/profile-*.html → "Configuring Projects"
+```
+
+Qualitative: spike configuration is dominated by AGP per Android module + Hilt/KSP/Room on annotated types — the same class of work NiA pays — not by Forma validator overhead on this graph size. Prefer same-host before/after if quoting wall-clock.
+
+---
+
+## 4. Side-by-side call sites
+
+### 4.1 Feature impl (For You)
+
+**Upstream** — pick convention plugins, then deps (force-deps from `feature.impl` / compose conventions are invisible here):
+
+```kotlin
+plugins {
+    alias(libs.plugins.nowinandroid.android.feature.impl)
+    alias(libs.plugins.nowinandroid.android.library.compose)
+    alias(libs.plugins.roborazzi)
+}
+android {
+    namespace = "com.google.samples.apps.nowinandroid.feature.foryou.impl"
+    testOptions.unitTests.isIncludeAndroidResources = true
+}
+dependencies {
+    implementation(projects.core.domain)
+    implementation(projects.feature.foryou.api)
+    // …
+}
+```
+
+**Spike** — one type entry; Hilt/Compose owned by type + attrs; project edges explicit; **no** other feature `impl`:
+
+```kotlin
+hiltImpl(
+    packageName = "com.google.samples.apps.nowinandroid.feature.foryou.impl",
+    compose = true,
+    productFlavors = niaContentTypeFlavors,
+    dependencies = transitiveDeps(/* compose + lifecycle + hilt-compose */) + deps(
+        target(":feature:foryou:api"),
+        target(":feature:foryou:res"),
+        target(":feature:interests:api"),
+        // … other feature api only …
+        target(":core:domain:android-util"),
+        target(":core:designsystem:ui-library"),
+        target(":core:ui:compose-widget"),
+        target(":core:navigation:api"),
+    ),
+)
+```
+
+### 4.2 Room database core
+
+**Upstream:** four plugin aliases + `android { namespace }`.  
+**Spike:**
+
+```kotlin
+roomAndroidUtil(
+    packageName = "com.google.samples.apps.nowinandroid.core.database",
+    compose = false,
+    dependencies = transitiveDeps("org.jetbrains.kotlinx:kotlinx-coroutines-core:1.10.2") + deps(
+        target(":core:model:library"),
+    ),
+)
+```
+
+### 4.3 APK root
+
+**Upstream `app`:** 10 plugin aliases (application, compose, flavors, jacoco, firebase, hilt, oss, baseline, roborazzi, serialization) + large `android { }` + long dependency list.  
+**Spike `binary`:** single `hiltFirebaseBinary(…)` with `versionCode`/`versionName`, `productFlavors`, and composition deps (features + sync + analytics/notifications leaves). UI composition stays in `root-app` (`hiltApp`).
+
+---
+
+## 5. What the matrix taught us (high-signal findings)
+
+Full table: [`DOGFOOD-NIA.md` § Findings log](DOGFOOD-NIA.md). Phase D highlights:
+
+| ID | Lesson for Forma product |
+|----|---------------------------|
+| **F1 / F2 / F22** | Feature `api` must stay ports-only (nav keys, no domain/android navigation). Upstream convention force-deps fight this; Forma makes the illegal edge a hard error. |
+| **F3 / F28 / F29** | Shared UI is not a generic library bucket: `uiLibrary` (designsystem) + `composeWidget` (cards) + presentation DTOs beat `uiLibrary`→`library` matrix weaken. |
+| **F5 / F20 / F21** | Multi-feature composition only at roots; feature→feature is **api** only at runtime. |
+| **F8 / F23 / F31** | Strings/icons belong on `androidRes`; `androidUtil` content rule is load-bearing. |
+| **F9 / F10** | External consumers need published **includer** or a documented flat-name `include` recipe (`target(":a:b:c")` → `:a-b-c`). |
+| **F11** | Project-global Compose ≠ every target; non-UI `androidUtil` / Application-only binary need `compose = false`. |
+| **F12** | Type-owned KSP companions require `processorConfigurationFeatures` on the apply path (engine fix). |
+| **F14 / F15 / F25** | Path B `deriveTargetType` does **not** inherit base plugins; need `*Target(type=)` helpers (`androidUtilTarget`, `libraryTarget`). |
+| **F16 / F17** | Multiple Path A plugins accumulate per type; Firebase SDK companions must be `transitiveDeps`. |
+| **F18 / F30** | Some “plugins” are library stacks only (DataStore Preferences, demo serialization JSON) — do not invent structure plugins. |
+| **F19** | Keep features on **domain** use cases; don’t punch `impl`→`data` for convenience. |
+| **F7 / F27** | One `FormaProductFlavor` model for binary **and** libraries; flavor-scoped deps via `.forProductFlavor` → `prodImplementation`, not raw AGP DSL. |
+
+---
+
+## 6. Product bar checklist (F-115)
+
+| Bar item | Status |
+|----------|--------|
+| Structure over configuration | **Met** — role DSL only at call sites |
+| Closed matrix / no `androidLibrary` | **Met** — findings re-sliced graph |
+| Composition only at roots | **Met** — binary/root-app compose features; no feature `impl`→`impl` |
+| Type-owned plugins | **Met** — Hilt/Room/Firebase/Protobuf/flavors |
+| Measurable external spike | **Met** — mavenLocal consumer, dual-flavor debug APKs green |
+| Case study metrics | **Met** — this doc |
+
+---
+
+## 7. Residual gaps (explicit non-goals / follow-ups)
+
+Not failures of the dogfood bar — candidates if Stepan opens a new ticket:
+
+1. **Publish `tools.forma.includer`** to mavenLocal/Portal (closes F9 properly).  
+2. **Retrofit/prod network** flavor source sets (demo JSON path already proves network util + F30).  
+3. **Test graph** — `androidTestUtil` / fakes without production `impl`→`impl` (F5 tests).  
+4. **Non-structure tooling** — Jacoco/Roborazzi/baseline/dependency-guard as fleet or root config, not target identity.  
+5. **Config-time profile pair** — optional same-host numbers for marketing/CI budgets.  
+6. **F-094** Portal org — still human-blocked for public plugin coordinates (spike used mavenLocal).
+
+---
+
+## 8. Conclusion
+
+On a ~36-module slice of Now in Android, Forma replaced **per-module plugin shopping + convention-plugin indirection** with **Bazel-like target types**: configure behavior on the type once (Path A/B + project-global classpath), keep call sites as **attributes and deps**.
+
+The strongest product signal is not only LOC reduction (**~59%** module scripts, **~79%** org build-logic Kotlin) but **enforced structure**: illegal edges and content rules surfaced real design fixes (nav ports, res splits, UI DTOs, domain façades) that upstream’s generic `android.library` graph leaves optional.
+
+Dogfood **Phase A–D complete** for F-115. Engine lessons already upstreamed on `v2` where needed (KSP apply path, util/library target helpers, library product flavors). Further NiA parity is optional product work, not a blocker on the meta-build thesis.
+
+---
+
+## Cross links
+
+- Design + findings: [`DOGFOOD-NIA.md`](DOGFOOD-NIA.md)  
+- Matrix: [`DEPENDENCY-MATRIX.md`](DEPENDENCY-MATRIX.md)  
+- Target plugins: [`TARGET-PLUGINS.md`](TARGET-PLUGINS.md)  
+- Call-site surface: [`CALL-SITE-SURFACE.md`](CALL-SITE-SURFACE.md)  
+- Nav ports: [`NAVIGATION-ABSTRACTION.md`](NAVIGATION-ABSTRACTION.md)  
+- Config performance recipe: [`CONFIGURATION-PERFORMANCE.md`](CONFIGURATION-PERFORMANCE.md)  
+- Local publish: [`PLUGIN-PUBLISH.md`](PLUGIN-PUBLISH.md)  
diff --git a/docs/DOGFOOD-NIA.md b/docs/DOGFOOD-NIA.md
@@ -1,6 +1,7 @@
 # Dogfood: Now in Android → Forma (F-115)
 
-**Status:** `in_progress` — phase C through **library product flavors (F27)** green; **Phase D** case study next  
+**Status:** **Phase A–D complete** (2026-08-02) — case study [`DOGFOOD-NIA-CASE-STUDY.md`](DOGFOOD-NIA-CASE-STUDY.md); spike remains external reference  
+
 **Upstream:** [android/nowinandroid](https://github.com/android/nowinandroid) (Apache-2.0)  
 **Pinned checkout (local):** `/Users/claw/work/nowinandroid` @ `7d45eae` (main tip when cloned 2026-07-29)  
 **Dogfood fork:** `/Users/claw/work/nowinandroid-forma`  
@@ -431,13 +432,18 @@ Workspace: same **`forma-spike`** external tree.
 | Version | `0.17.0-nia-forma-library-flavors` |
 | **F27** | **closed** |
 
-**Still open:** Phase D case-study write-up (optional remaining Retrofit/prod network out of scope).
+**Still open (optional, not F-115 bar):** Retrofit/prod network flavor path; test graph; non-structure tooling (Roborazzi/baseline/…).
 
-### Phase D — Case study write-up
+### Phase D — Case study write-up ✅ (2026-08-02)
 
-- Before/after build files (LOC, plugin lines)  
-- Config time optional (`CONFIGURATION-PERFORMANCE.md` recipe)  
-- Upstream PR? **No** — dogfood stays fork unless Google interest  
+| Step | Result |
+|------|--------|
+| Doc | [`DOGFOOD-NIA-CASE-STUDY.md`](DOGFOOD-NIA-CASE-STUDY.md) — before/after LOC, call-site pairs, findings digest, residual gaps |
+| Module scripts | **1476 → 600** LOC (−59%); org logic **1917 → 401** LOC build-logic vs forma-defs (−79%) |
+| Call sites | **0** raw `android {}`, **0** `.withPlugin`; single-method target DSLs |
+| Config time | Recipe only (no locked same-host pair) — see case study §3.5 |
+| Upstream PR | **No** — fork-only unless Google interest |
+| F-115 bar | **Met** — structure / matrix / roots / type-owned plugins / measurable spike / case study |
 
 ## Findings log (living)