[REFACTOR]: Add shared linear trace runner

docs/api/core-types.md

@@ -1,6 +1,8 @@
 # API Reference — Core Types
 
-Data types shared across the entire framework. All importable from `rampart` directly.
+Data types shared across the entire framework. Stable execution vocabulary is
+available from `rampart.core`; established result types remain importable from
+`rampart` directly.
 
 ## Data Types
 
@@ -14,6 +16,8 @@ Data types shared across the entire framework. All importable from `rampart` dir
         - ToolCall
         - SideEffect
         - Turn
+        - EvaluationPurpose
+        - TraceEndReason
         - EvalOutcome
         - EvalResult
         - EvalContext
@@ -30,6 +34,8 @@ Data types shared across the entire framework. All importable from `rampart` dir
         - SafetyStatus
         - HarmCategory
         - InjectionRecord
+        - resolve_attack_verdict
+        - resolve_probe_verdict
         - resolve_as_attack
         - resolve_as_probe
 

docs/attacks/xpia.md

@@ -159,12 +159,12 @@ Place the cheaper evaluator on the left side of `|` — it short-circuits if the
 The `&` above asks whether both happened, so one condition that definitively did not happen settles the result even if the adapter could not observe the other. Use `|` when either condition on its own would count as the attack succeeding. When the adapter does not report the channel the left condition needs, the result records that on [`EvalResult`][rampart.core.types.EvalResult]. Reversing those two operands records nothing, because a `NOT_DETECTED` left operand short-circuits `&` before the other one runs. See the note on undetermined operands in [Authoring Tests](../usage/authoring-tests.md#composing-evaluators).
 
 !!! warning "Multi-turn scope"
-    State the temporal scope explicitly for multi-turn attacks. The complete
-    positive and negated mapping is maintained in the
+    `ResponseContains` requires an explicit temporal scope, even for a
+    single-turn attack. The complete positive and negated mapping is maintained in the
     [Temporal Scope table](../usage/authoring-tests.md#temporal-scope).
-    Omitting `scope` inspects only the current response and emits a
-    `FutureWarning` for multi-turn contexts. Scope applies only to turns in the
-    evaluator context; it does not control execution length or early stopping.
+    Use `CURRENT_TURN` only when earlier responses should be ignored. Scope
+    applies only to turns in the evaluator context; it does not control
+    execution length or early stopping.
 
 ### LLMDriver for Adaptive Triggers
 
@@ -233,7 +233,7 @@ See [`Attacks.xpia()`][rampart.attacks.Attacks.xpia] for the full API reference.
 | `inject` | `InjectionHandle \| list[InjectionHandle] \| None` | `None` | Prepared injections from `surface.inject()`. `None` for inline XPIA. |
 | `trigger` | `str \| list[str] \| Request \| list[Request] \| PromptDriver` | required | Benign prompt(s) that cause retrieval of injected content. |
 | `evaluator` | [`Evaluator`][rampart.core.evaluator.Evaluator] | required | What attack condition to detect. |
-| `max_turns` | `int` | `5` | Maximum prompt-response exchanges before `ERROR`. |
+| `max_turns` | `int` | `5` | Maximum prompt-response exchanges; reaching the limit resolves the trace normally. |
 | `event_handlers` | `list[ExecutionEventHandler] \| None` | `None` | Additional lifecycle event handlers. |
 
 ---
@@ -249,5 +249,3 @@ This only fires when all three conditions hold:
 3. Zero tool calls were observed
 
 It is a backstop for evaluators that cannot say up front what evidence they need, such as `LLMJudge`, where the answer depends on the objective. `ToolCalled` and `SideEffectOccurred` return `UNDETERMINED` themselves, so on their own they do not reach this check as `SAFE`. A composition still can, so the backstop stays.
-
-

docs/concepts/probes.md

@@ -38,11 +38,11 @@ All probes are created through the [`Probes`][rampart.probes.Probes] class:
 
 ```python
 from rampart import Probes
-from rampart.evaluators import ResponseContains
+from rampart.evaluators import ResponseContains, ResponseScope
 
 execution = Probes.behavior(
     prompt="What is 2 + 2?",
-    evaluator=ResponseContains("4"),
+    evaluator=ResponseContains("4", scope=ResponseScope.ALL_TURNS),
 )
 
 result = await execution.execute_async(adapter=my_adapter)
@@ -60,5 +60,3 @@ Provide exactly one of `prompt`, `prompts`, or `driver`.
 | [Behavioral](../probes/behavioral.md) | `Probes.behavior(...)` | Verify the agent produces expected responses or behaviors |
 
 More probe types will be added. Each new probe is a new factory method on `Probes`.
-
-

docs/concepts/trace-schema.md

@@ -49,10 +49,22 @@ IDs must be recorded, not generated during deserialization. These boundary
 rules do not replace the normal dataclass constructors used during execution.
 
 When recorded, `result_index` must be a nonnegative integer. Population references
-require a positive `size`, a zero-based `index` less than `size`, and a `threshold`
-in `[0, 1]`. These invariants are enforced by the canonical record boundary and
-adapter, without changing live `PopulationRef` construction or other adapters.
-Scalar bounds are included in the generated JSON Schema.
+require a nonempty string `id`, a positive `size`, a zero-based `index` less than
+`size`, and a finite `threshold` in `[0, 1]`. Live `PopulationRef` constructors
+enforce these invariants, and the canonical adapter revalidates existing instances
+at the write boundary as well as decoded records. Scalar bounds, including the
+nonempty ID, are included in the generated JSON Schema.
+
+`Result.final_trace_evaluation` records evaluation of the trace when execution
+stops, including when the turn budget is reached.
+`Turn.eval_result` remains separate online evidence; `Turn.eval_purpose` records
+why that online evaluation ran. A non-null purpose requires an evaluation on the
+same turn. `Result.trace_end_reason` records why turn production stopped. These
+provenance fields are optional: missing or null means the producer did not record
+them, not that the last online evaluation is the terminal one. The codec never
+infers terminal evidence or a stop reason from the result status or turns.
+Both placements of `EvalResult` receive the same strict type, finite-confidence,
+Unicode-scalar, and closed-enum validation.
 
 Body encoding validates the live result and uses Pydantic's JSON-mode
 serialization, with adapter-local Unicode validation and Python ISO datetime
@@ -66,18 +78,24 @@ re-encoded when Pydantic's JSON-mode writer has a lower nesting limit.
 These failures raise `SchemaError`; successful decoding alone does not guarantee
 that an unusually deep record can be emitted again.
 
-These policies belong to the cached canonical adapter, not to the public
-dataclass annotations or configuration. Fields remain `dict[str, Any]` and
+Canonical serialization policies belong to the cached adapter, not to the public
+dataclass annotations or configuration. Shared dataclass definitions reuse the
+adapter's configured copy at every occurrence. Fields remain `dict[str, Any]` and
 `datetime | None`. Independently constructed Pydantic adapters retain their
-normal behavior, including live binary payload support.
+normal behavior, including live binary payload support and ordinary instance
+validation. Constructor invariants still apply when those adapters construct a
+new instance.
 The canonical adapter supplies its own `datetime` / `Path` resolution namespace;
 the shared types module keeps those imports under `TYPE_CHECKING`.
 
 `ResultRecord.json_schema()` returns the adapter-derived body schema plus the
 versioned envelope. Small schema customizations describe the trace-only payload
 restrictions and the request invariant (a prompt or at least one attachment).
+It also describes the dependency between a turn's purpose and evaluation.
 `JsonSchemaValue` is the return type, not a separate model or validator.
-The open Draft 2020-12 contract is committed at `schemas/trace.v1.schema.json`.
+The active open Draft 2020-12 contract is committed at
+`schemas/trace.v2.schema.json`; `schemas/trace.v1.schema.json` is retained unchanged
+as the historical description of v1.
 
 Regenerate it with `uv run python scripts/generate_trace_schema.py`.
 The generator selects the filename from `TRACE_SCHEMA_VERSION`. CI runs the same
@@ -88,7 +106,8 @@ command with `--check` to detect drift.
 Schema drift checking alone does not establish compatibility. A separate CI gate
 requires a checked-in decision in `schemas/trace-compatibility.json`, bound to the
 contract content by SHA-256 fingerprints. The inputs are `result.py`, `types.py`,
-`serialization.py`, `_schema.py`, and all published `trace.v*.schema.json` files.
+`serialization.py`, `_schema.py`, `_population.py`, and all published
+`trace.v*.schema.json` files.
 Watching the models and codec policies also catches changes that do not appear
 in JSON Schema. This is deliberately conservative: even a nonsemantic edit to
 these inputs needs a compatibility rationale.
@@ -100,7 +119,8 @@ For a contract change, update the declaration:
   compatibility, such as an additive-optional field with a defined absence behavior.
 - **`new-major`** increments the major by one, retains earlier published schema
   files, and references a nonempty repository migration document in `migration_note`.
-  The migration obligations below still apply, including an upcaster and API/CLI.
+  The note explains the break and the actual reader/migration support shipped;
+  it does not require an upcaster or dual reader.
 
 Historical schema files remain unchanged in subsequent same-major PRs, not just
 during a major bump. A `compatible` decision may update the active major's schema;
@@ -122,7 +142,7 @@ Without `--base-ref`, including on main-branch pushes, the command checks the
 declaration's version and current content fingerprint only.
 
 **The declaration is a review gate, not proof of compatibility.** Reviewers must
-assess the rationale, semantic behavior, and required migration implementation.
+assess the rationale, semantic behavior, and any claimed migration support.
 A regenerated schema or a `compatible` assertion does not make a breaking change
 safe. Keep the input list current if contract policy moves to additional modules.
 
@@ -190,10 +210,10 @@ does not make currently rejected formats readable by older readers.
 ## Versioning
 
 - Every serialized record carries one root `version` field. The current schema
-  is **`rampart.trace.v1`**.
+  is **`rampart.trace.v2`**.
 - The record version is **independent** of transport or projection versions,
-  including the existing xdist envelope version (`rampart.xdist.v2`). Each
-  version describes its own layer and may evolve separately.
+  including xdist's versioned envelope. A transport's cadence or version number
+  does not select the canonical trace major.
 - There is a **single root version** — nested types (`Turn`, `Payload`,
   `EvalResult`, …) do not carry their own versions.
 
@@ -210,13 +230,14 @@ does not make currently rejected formats readable by older readers.
   defaults and does not retain which fields were absent. For example, omitted
   `turns` becomes `[]` and is emitted when re-encoded.
 - **Structural change = major bump.** Removing, renaming, or retyping a field,
-  or changing its meaning or nesting, bumps `vN → vN+1` with a changelog and a
-  migration note.
+  changing its meaning or nesting, or narrowing its accepted value domain bumps
+  `vN → vN+1` with a changelog and a migration note. Rejecting previously accepted
+  empty population IDs is a domain-narrowing change, not an optional addition.
 
 ```mermaid
 flowchart TD
     change([proposed schema change]) --> q1{"adds a field only?"}
-    q1 -- no --> struct["structural:<br/>remove / rename / retype /<br/>change meaning or nesting"]
+    q1 -- no --> struct["structural:<br/>remove / rename / retype /<br/>change meaning, nesting, or accepted values"]
     q1 -- yes --> q2{"optional with a<br/>well-defined default?"}
     q2 -- no --> struct
     q2 -- yes --> add["additive-optional"]
@@ -227,17 +248,20 @@ flowchart TD
 
 ## Reader posture
 
-- Readers tolerate unknown fields and **fail closed on an unknown major** — a
+- Readers tolerate unknown fields and **fail closed on an unsupported major** — a
   record is never best-effort parsed across a major boundary.
+- This reader supports only v2. Retaining the v1 schema does not register a v1
+  decoder; v1 records raise `UnsupportedSchemaVersionError`, as do future majors.
 - Forward compatibility is **additive-only within a major**. A newer major read
   by an older framework fails closed by design.
 - Schema descriptions and validators derived from this format must remain open
   to unknown properties within a major version.
 
 ## Enum posture
 
-- The closed enums — `SafetyStatus`, `EvalOutcome`, `ObservabilityLevel`, and
-  `PayloadFormat` — **fail closed** on an unknown value. A serialized safety
+- The closed enums — `SafetyStatus`, `EvalOutcome`, `EvaluationPurpose`,
+  `TraceEndReason`, `ObservabilityLevel`, and `PayloadFormat` — **fail closed**
+  on an unknown value. A serialized safety
   result must never silently misread one; there is no warn-and-degrade path.
 - `HarmCategory` is the sole exception: it travels as a **passthrough string**
   and is never coerced, so a new harm label from a future producer round-trips
@@ -256,26 +280,37 @@ flowchart TD
 - Timestamps retain Python's ISO 8601 representation, including naive datetimes
   and UTC offsets. The schema describes strings rather than RFC 3339
   `date-time`, which would exclude some supported Python datetimes.
-- `rampart.trace.v1` does not define a durable representation for binary or
+- `rampart.trace.v2` does not define a durable representation for binary or
   opaque payload artifacts. Encoding or decoding one fails closed rather than
   coercing it to text.
 - Encoding and decoding preserve supported metadata, including keys used for
   transport bookkeeping or loss/truncation markers. Unsupported values are
   rejected regardless of the key name. Metadata hygiene belongs to consumer
   preparation, not to the canonical codec.
 
-## Migration mechanics
-
-Only `rampart.trace.v1` exists today. No upcaster or persisted-data migration
-tooling is implemented. If a later structural change introduces a new major,
-the migration policy requires:
-
-- writers emit the latest supported major;
-- each major bump ships an adjacent upcaster (`vN-1 → vN`) and an explicit
-  migration API/CLI;
-- migrating persisted data is an explicit operation; reading never rewrites an
-  artifact in place; and
-- encountering an unsupported major fails closed.
+## v1 to v2 migration note
+
+V2 narrows population IDs to nonempty strings. It also records optional terminal
+evaluation, trace-end reason, and online evaluation purpose, and consistently
+applies canonical validation to both terminal and online evaluations. The new
+optional fields alone would not require a major bump; the narrowed ID domain
+does. The field is named `final_trace_evaluation` in the Python API, canonical
+records, JSON reports, and xdist transport. The earlier `terminal_evaluation`
+spelling is removed without an alias.
+
+Writers emit v2, and this reader accepts only v2. No v1 reader, adjacent upcaster,
+or persisted-data migration API/CLI is shipped. Historical v1 schema files are
+retained for consumers that need to inspect old records, not as a support-window
+promise.
+
+Persisted v1 data must not be silently relabeled or parsed through the v2 reader.
+Consumers choosing to migrate it must perform an explicit, application-owned
+conversion into a separate v2 record and validate the result with
+`deserialize_record()`. An empty population ID requires a legitimate identifier
+from the producer's provenance or regeneration of the record; do not invent one.
+Leave unrecorded terminal evaluation, stop reason, and turn purpose absent or
+null rather than inferring them from the last online evaluation. Preserve the
+original artifact; reading never rewrites persisted data in place.
 
 ## Future extensions
 
@@ -289,12 +324,11 @@ default do not require a major bump. Structural changes do. Apply the compatibil
 review and declaration requirements to each extension rather than promising
 compatibility for an unimplemented representation.
 
-## Support window
-
-This is a release-support commitment; the current reader supports only
-`rampart.trace.v1`.
+## Pre-1.0 support policy
 
-Starting with the first release that writes durable trace records by default,
-RAMPART supports reading `vN` and `vN-1` for **two subsequent framework
-releases** (one deprecation cycle). The window is keyed on releases, not time.
-Any major bump includes a changelog entry and migration note.
+RAMPART does not promise deprecation periods, compatibility aliases, a two-release
+support window, dual readers, or mandatory upcasters. Breaking changes may replace
+old APIs directly. Every canonical major change still requires an explicit
+version/compatibility decision, unchanged historical schema descriptions, and a
+changelog entry and migration note describing actual support. Unsupported
+versions always fail closed.

docs/contributing/release-process.md

@@ -26,11 +26,16 @@ RAMPART follows [Semantic Versioning](https://semver.org/) (`MAJOR.MINOR.PATCH`)
 !!! note "Pre-1.0 stability"
     While RAMPART is below `1.0`, minor version bumps may include breaking changes. The API is stabilizing but not yet frozen. The first stable release will be `1.0.0`.
 
-## 3. Remove Deprecated Functionality
+## 3. Review Breaking Changes
 
-If you are incrementing the minor version, search the codebase for the new minor version (no leading `v`) to find occurrences where functionality was deprecated and announced for removal in this version. Typically, functionality is deprecated and stays for two minor versions before being removed.
+During pre-1.0 development, remove obsolete APIs directly rather than maintaining
+deprecated aliases, warnings, or a fixed support window. Migrate in-tree callers,
+tests, and documentation together, and explain required caller changes in the
+release notes.
 
-If you find functionality to remove, merge the removal PR to `main` before proceeding.
+Breaking persisted-data changes still require an explicit schema-version and
+compatibility decision; see [Trace Schema](../concepts/trace-schema.md).
+Merge the completed changes to `main` before proceeding.
 
 ## 4. Prepare Release Metadata