@revturbine/sdk 0.10.9 → 0.11.0

package/CHANGELOG.md

@@ -47,6 +47,171 @@ also require a changelog entry.
 
 ---
 
+## 0.11.0
+
+### `Playbook` is the canonical name for every option that carries one (BL-0156)
+
+**What changed.** `ExportedConfig` is dead vocabulary. Plan 118 renamed the
+domain object to **Playbook** and plan 104 renamed the schema type to
+`RevTurbineConfig`, but the *option*, *keyword* and *method* names that carry a
+Playbook around the SDK still spelled it `exportedConfig` / `exported_config`.
+Every one of them now has a `Playbook`-named counterpart; every old name stays
+as a deprecated alias that still works.
+
+TypeScript:
+
+| deprecated | canonical |
+|---|---|
+| `localRuntime.exportedConfig` | `localRuntime.playbook` (already shipped; see below) |
+| `localRuntime.resolvers.resolveExportedConfig` | `localRuntime.resolvers.resolvePlaybook` |
+| `configProvider.getExportedConfig()` | `configProvider.getPlaybook()` |
+| `sdk.getExportedConfig()` | `sdk.getPlaybook()` |
+| `getPolicy().exportedConfigVersion` | `getPolicy().playbookVersion` |
+| `BrowserRuntimeOptions.exportedConfig` | `BrowserRuntimeOptions.playbook` |
+| `LocalEvaluationServerOptions.exportedConfig` | `LocalEvaluationServerOptions.playbook` |
+| `ExportedConfigProvider` (type) | `RevTurbineConfigProvider`, also exported as `PlaybookProvider` |
+| `ExportedConfig` (type) | `Playbook` |
+
+Python: `RevTurbineCustomerSdk(exported_config=…)`, `LocalRuntime(exported_config=…)`,
+`LocalRuntime.get_exported_config()`, `create_static_placement_resolver(exported_config=…)`,
+`derive_local_entitlement_from_configured_rules(exported_config=…)`,
+`configured_plan_name_from_exported_config`, `parse_exported_config_or_throw` and
+`static.ExportedConfig` all gained `playbook`-spelled counterparts.
+
+Rust: `configured_plan_name_from_exported_config` is now
+`configured_plan_name_from_playbook`, with the old name kept as the crate's
+first `#[deprecated]` item. Rust has no keyword arguments, so nothing else in
+that port was a caller-visible name.
+
+`RevTurbineConfigProvider`'s two accessors are both OPTIONAL on the interface so
+that an existing `implements RevTurbineConfigProvider` clause keeps compiling;
+the `configProvider` option intersects it with a union that still requires
+exactly one of them, so a provider implementing neither is a compile error where
+it is passed.
+
+Reading a deprecated name logs **one** development-build warning per runtime
+(one `DeprecationWarning` per process on Python) naming the canonical
+replacement — one flag for every alias, so an integration still spelling several
+of them gets a single line, not a wall. Production builds are silent. Requires
+`@revt-eng/core` ≥ 0.1.331, which made the same rename upstream (scaffold #380).
+
+Ruling: Kent, 2026-09-23 (BL-0156).
+
+**Landed in** `0.11.0`. **Fail-closed in** `0.12.0` for the `ExportedConfig`
+aliases — until then every old name works and warns once.
+
+**Proving test:** `web-sdk/playbook-option.test.ts` (canonical never warns, the
+alias still resolves, one warning across three read sites, both spellings reach
+the same Playbook at every public read site),
+`server-python/tests/test_playbook_option.py` (the same contract plus
+`get_exported_config()`), and `server-rust/src/user_context.rs`
+`playbook_rename_tests` (the alias returns exactly what the canonical returns).
+
+### `localRuntime.playbook` is the canonical local-mode config key
+
+**What changed.** The record this file owed and never carried: `localRuntime.playbook`
+shipped earlier as the canonical spelling of `localRuntime.exportedConfig`, with
+`resolveLocalPlaybook()` as the single resolver deciding precedence, but no
+changelog entry was ever written for it — so an integration could not date the
+change. It is recorded here rather than back-dated, because the version it
+landed in is not the version this entry appears in and pretending otherwise is
+exactly the confusion the "Landed in" / "Fail-closed in" split exists to
+prevent. As of `0.11.0` the alias also warns once (it did not before).
+
+**Landed in** a release before `0.11.0` (additive; the alias never stopped
+working). **Fail-closed in** `0.12.0`, with the rest of the `ExportedConfig`
+aliases.
+
+**Proving test:** `web-sdk/playbook-option.test.ts` — `resolveLocalPlaybook`
+accepts either key, prefers `playbook`, and warns only for the alias.
+
+### `publicKey` is the only browser credential; `ingestPublicKey` is gone (BL-0113)
+
+**What changed.** `0.10.0` made `publicKey` the browser credential and kept
+`apiKey` and `ingestPublicKey` working as aliases "for one minor". This is that
+minor:
+
+- **`ingestPublicKey` is removed from `RevTurbineInitOptions`.** Passing it is
+  now a **type error**, including when the options are built in a variable
+  (`ExactInitOptions` rejects it there too).
+- **`apiKey` is no longer read as a browser credential.** It still exists, and
+  still means exactly one thing: the secret **server key**, for
+  `@revturbine/sdk/server` and for the headless SDK on a backend. A browser init
+  that supplies only `apiKey` now resolves **no** browser credential and fails
+  at init, instead of silently sending a server key as the bearer.
+
+`resolveBrowserPublicKey()` reads `publicKey` and nothing else, and
+`resetBrowserKeyAliasWarning()` is gone with the warning it reset. Keyless
+local-only init is unchanged, and the plan 95 anonymous `sdk_init` beacon now
+keys off `publicKey` alone.
+
+Ruling: Kent, 2026-09-23 (BL-0113), closing the window plan 257 opened.
+
+**Landed in** `0.11.0`. **Fail-closed in** `0.11.0` — the same release. This IS
+the fail-closed half of the `0.10.0` entry below; there is no further tolerance
+window.
+
+**Proving test:** `web-sdk/public-key-option.test.ts` (`publicKey` is the bearer
+on ingest and every control-plane fetch; `apiKey` resolves nothing; nothing
+warns), `web-sdk/customer-side-ingest.test.ts` (the bearer no longer carries the
+server key), and `web-sdk/init-options-exactness.test-d.ts` (`ingestPublicKey`
+is a type error inline AND in a variable — if the key is ever re-added, `tsc`
+fails on the unused `@ts-expect-error`).
+
+---
+
+## 0.10.10
+
+### Trial tokens now reach placement content, not only `getPersonalizationTokens()` (BL-0169)
+
+**What changed.** A placement whose copy authors `{{trial_days_remaining}}` or
+`{{trial_days_total}}` now renders the live values. Content interpolation sources
+its token map from the **output content itself**, so a provider-derived token
+only reaches the copy if the resolver writes it onto that content first. The
+usage lane always did (`usage_current`, `usage_limit`, `usage_percent`,
+`usage_remaining`, `reset_date`); the trial lane never did, so a `trial_ending`
+body reading `"{{trial_days_remaining}} days left"` shipped the literal
+`{{trial_days_remaining}}` to the end user — on every port. The two names were
+derived for `getPersonalizationTokens()` all along, which is why the gap survived:
+a host that read the token map saw the right numbers and a host that authored the
+token into copy did not.
+
+| | before | now |
+|---|---|---|
+| `getPersonalizationTokens().trial_days_remaining` | live value | live value (unchanged) |
+| content `{{trial_days_remaining}}` / `{{trial_days_total}}` | literal `{{…}}` | live value |
+| `outputContent.trial_days_remaining` / `_total` | absent | written when the plan provider carries a finite number |
+
+Identical on all three ports (`ts:local-resolver.ts`,
+`py:core/placements/local_resolver.py`, `rs:placements/static_resolver.rs`): the
+same two token names, a `Number.isFinite` guard, provider state winning over an
+authored content value of the same key, and the value passed through **without
+widening** — `day_number` 7 + `days_remaining` 3 is `trial_days_total` `10`, never
+`10.0` (BL-0155).
+
+**Who this reaches.** Any host whose placement copy authors either token. Copy
+that hard-coded one of those keys as a decorative literal now has it overwritten
+by live state — the same precedence the usage lane and
+`derivePlacementPersonalizationTokens` have always had. `{{trial_plan_name}}` and
+`{{trial_features_used}}` remain **not yet wired** by the placement-studio-ui
+spec's own note and are untouched.
+
+**Landed in.** `0.10.10`.
+
+**Fail-closed in.** `0.10.10` — absence still leaves the raw token standing
+rather than rendering `0`, so a host with no trial provider state sees exactly
+what it saw before.
+
+**Proving test.** Parity scenario `trial_tokens_in_content` under
+`"normalize": "preserve-representation"` (byte-locked ts ≡ py ≡ rs), plus
+`py:tests/placements/test_local_resolver.py` and `rs:tests/static_resolver.rs`
+`trial_tokens_are_injected_from_plan_provider_state` and its two neighbours,
+mirroring the scaffold-side `local-resolver.test.ts` cases. Needs
+`@revt-eng/core` 0.1.331 or later (scaffold #382) for the TS half; this release
+pins 0.1.332.
+
+---
+
 ## 0.10.9
 
 ### The trial-status overlay no longer changes a number's type (BL-0155)
@@ -87,11 +252,21 @@ representation simply changes.
 `server-python/tests/test_trial_overlay_upsert.py::test_overlay_preserves_integer_numeric_representation`
 and `server-rust/src/sdk.rs::overlay_preserves_integer_numeric_representation`
 — both assert the *type*, not just the value, because `100 == 100.0` in Python.
-The parity corpus cannot see this class of divergence (`tests/parity/normalize.*`
-rule 4 collapses integral floats by design); scenario
-`trial_overlay_integer_fields` locks that the three ports still decide
+Scenario `trial_overlay_integer_fields` locks that the three ports still decide
 identically from an all-integer trial status.
 
+> **Correction, 2026-09-23 (BL-0158, harness-only — no release).** This entry
+> originally said the parity corpus *cannot* see this class of divergence,
+> because `tests/parity/normalize.*` rule 4 collapses integral floats by design
+> and nothing in the corpus returned the overlaid provider state. Both halves
+> are now closed and the sentence no longer holds:
+> `trial_overlay_integer_fields` drives `resolveProviders` (already shipped on
+> all three runtimes) under a new per-scenario
+> `"normalize": "preserve-representation"` flag that turns the collapse off, so
+> `14` and `14.0` are different bytes and the cross-language byte-diff fails on
+> a widened field. No public API changed; the harness gained the ability to see
+> provider state.
+
 ---
 
 ## 0.10.8
@@ -483,8 +658,9 @@ and the plan 95 anonymous `sdk_init` beacon still fires only when no public key
 
 Ruling: Kent, 2026-09-21, devkit PR #808 (closed).
 
-**Landed in** `0.10.0`. **Fail-closed in** not yet — the aliases are removed one
-minor later; until then the old shape works and warns.
+**Landed in** `0.10.0`. **Fail-closed in** `0.11.0` — `ingestPublicKey` was
+removed from the options type and `apiKey` stopped being read as a browser
+credential (BL-0113); see the `0.11.0` entry above.
 
 **Proving test:** `web-sdk/public-key-option.test.ts` — precedence, the single
 bearer across ingest and control-plane fetches, the browser-only warning, and