diff --git a/packages/paperclip-runner/docs/architecture.md b/packages/paperclip-runner/docs/architecture.md
index 82f57155f5a..8eb1a35e547 100644
--- a/packages/paperclip-runner/docs/architecture.md
+++ b/packages/paperclip-runner/docs/architecture.md
@@ -196,6 +196,38 @@ lease expiry, storage pressure, drain, and revoke. The browser recovery page
 calls that peer through guarded package-local Vite middleware and renders only
 safe trace fields.
 
+## Phase 7 capability-model boundary
+
+```text
+Paperclip skill + 7 references        Paperclip Evals corpus (106 cases)
+            |                                      |
+            +-------------------+------------------+
+                                v
+                generated capability contract (258 rows, 41 MCP aliases)
+                                |
+        +-----------------------+-----------------------+
+        v                       v                       v
+ semantic tool catalog   authorization engine    eval conformance suite
+        \                       |                       /
+         \                      v                      /
+          +-----> in-process mock ControlPlanePort <--+
+                                |
+                                v
+                   read-only browser scenario explorer
+```
+
+Phase 7 is a package-local model of a native Paperclip run. It classifies every
+capability as control-plane-owned, always-agent-tool, or optional-agent-tool,
+exposes the always/optional set as a transport-neutral semantic tool catalog,
+gates optional tools behind grants, and proves 106 eval-derived cases against an
+in-process mock `ControlPlanePort`. The mock adapter is the only coupling point,
+so a real adapter can replace it later without touching the catalog,
+authorization rules, or conformance suite. Phase 7 contacts no Paperclip
+service, database, ACPX session, or provider credential; the
+[forbidden-imports checker](#forbidden-dependencies) keeps it that way. Real
+integration is Phase 8 (ACPX) and requires separate approval; see
+[the future binding boundary](phase-07-future-binding-boundary.md).
+
 ## Future integration rule
 
 The [implementation plan](../spec/paperclip-native-runner-implementation-plan.md)
diff --git a/packages/paperclip-runner/docs/index.md b/packages/paperclip-runner/docs/index.md
index 178856d60e1..28bf11705c4 100644
--- a/packages/paperclip-runner/docs/index.md
+++ b/packages/paperclip-runner/docs/index.md
@@ -14,6 +14,7 @@ Each tutorial starts with an explanation of what the phase is and what the phase
 - [Phase 4b: run the live Codex protocol console](tutorials/phase-04b-live-console.md) — chat with a live session in the browser, steer it, stop it three ways, answer its requests, change its goal, break its connection, and replay the record.
 - [Phase 5: run the SDK console and mini consumer](tutorials/phase-05-sdk-console.md) — exercise the versioned public browser/React surface in two independent consumers against fake and real drivers.
 - [Phase 6: run the thin Paperclip adapter](tutorials/phase-06-thin-paperclip-adapter.md) — prove mock/real conformance, run one feature-flagged local task, inspect replay/finalization, disable the kill switch, and verify legacy fallback.
+- [Phase 7: explore the capability contract and scenario explorer](tutorials/phase-07-scenario-explorer.md) — prove the 106-case conformance suite, open the read-only browser explorer over the mock control plane, and run the focused verification set, all from a clean checkout with no Paperclip service or external eval repository.
 - [Cumulative end-to-end tutorial](tutorials/end-to-end.md) — the shortest complete workflow available at the current phase.
 
 ## Reference
@@ -27,6 +28,16 @@ Each tutorial starts with an explanation of what the phase is and what the phase
 - [Phase 4b live console](phase-04b-live-console.md)
 - [Phase 5 browser SDK and reference console](phase-05-sdk.md)
 - [Phase 6 thin Paperclip adapter](phase-06-thin-paperclip-adapter.md)
+- [Phase 7 capability contract (generated)](phase-07-capability-contract.md)
+- [Phase 7 capability disposition](phase-07-capability-disposition.md)
+- [Phase 7 mock ControlPlanePort](phase-07-mock-control-plane-port.md)
+- [Phase 7 semantic tool catalog](phase-07-semantic-tools.md)
+- [Phase 7 authorization and exposure](phase-07-authorization-and-exposure.md)
+- [Phase 7 eval-derived conformance](phase-07-eval-conformance.md)
+- [Phase 7 browser scenario explorer](phase-07-scenario-explorer.md)
+- [Phase 7 future binding boundary (Phase 8 / ACPX)](phase-07-future-binding-boundary.md)
+- [Phase 7 verification commands](phase-07-verification-commands.md)
+- [Phase 7 scenario explorer UX interaction map](design/phase-7-scenario-explorer-ux.md)
 - [Engineering journal guide](journal.md)
 - [Dated shadcn/ui and AI Elements compatibility note](research/2026-08-07-ui-library-compatibility.md)
 - [Phase 4b live-console interaction map](design/phase-4b-interaction-map.md)
diff --git a/packages/paperclip-runner/docs/phase-07-authorization-and-exposure.md b/packages/paperclip-runner/docs/phase-07-authorization-and-exposure.md
new file mode 100644
index 00000000000..05e5c61e76e
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-authorization-and-exposure.md
@@ -0,0 +1,99 @@
+# Phase 7 Authorization and Exposure
+
+Two decisions gate every capability: **exposure** (is the tool even offered to
+this actor?) and **invocation** (may this specific call proceed?). The
+authorization engine answers both, produces typed denials that carry no
+protected state, and redacts secrets before any observable boundary.
+
+Sources: `src/tools/phase7-tool-authorization.ts`,
+`src/tools/phase7-semantic-tool-runtime.ts`, `src/tools/phase7-tool-bindings.ts`.
+
+## Three-way classification
+
+- **Control-plane-owned** — a frozen operation list (`checkout_task`,
+  `release_task`, `select_work`, `route_wake`, `enforce_budget`,
+  `append_audit_record`, `persist_run`, `replay_run`, `schedule_blocker_wake`,
+  `reconcile_run`). `authorizeInvocation` short-circuits these to outcome
+  `absent` with reason `control_plane_owned_operation`. No agent tool exists for
+  them and none can be invoked.
+- **Always-agent tool** — exposed to any actor whose task is active.
+- **Optional-agent tool** — exposed only when a grant unlocks it.
+
+## Exposure vs. invocation
+
+- `computeVisibleTools()` walks the whole catalog and records, per tool, an
+  exposure of `exposed` or `absent`, returning only the allowed tools.
+- `authorizeInvocation()` re-evaluates at call time, producing `allowed` or
+  `denied`.
+
+Invocation is evaluated in order and fails at the first rule that rejects:
+policy-denied operation IDs; task-mode membership
+(`operation_not_available_in_task_mode`); role check
+(`actor_role_not_authorized`); policy-denied claims
+(`required_claim_denied_by_policy`); missing grants
+(`required_claim_missing`); `read_secret_value` requires
+`policy.allowSecretValueAccess` (`secret_value_access_disabled`);
+`generic_api_request` requires the escape hatch to be enabled and the target to
+be allowlisted (`generic_escape_hatch_disabled` / `..._target_denied`); and the
+interaction-kind policy for `request_human_input`. Allowed calls carry the
+reason `task_scoped_always_tool` or `required_claims_granted`.
+
+Self-approval is prohibited even with the role and an explicit decision claim.
+
+## How grants unlock optional tools
+
+The actor's grants are the union of `capabilityGrants` (the actor's own) and
+`scenarioGrants` (seeded per scenario). An optional tool unlocks only when
+**every** entry in its `requiredClaims` is present in that union. Each
+authorization record stores the `consideredClaims`, `grantedClaims`, and
+`missingClaims`, so a denial explains exactly which claim was absent.
+
+## Typed denials carry no protected state
+
+A denial is a `Phase7PolicyDenial`: `ok: false` with an `error.code` of
+`policy_denied`, `operation_absent`, `input_invalid`, or `operation_unsupported`,
+the `operationId`, a generic `reason`, and the authorization record. It carries
+no fixture state and no protected payload. If the underlying mock operation
+throws, the runtime emits `operation_unsupported` with reason
+`mock_operation_rejected` and swallows the underlying error, so protected state
+cannot leak through an exception.
+
+## Secret redaction (the PAP-16909 rule)
+
+Redaction is two layers:
+
+1. **Model-only delivery is separated from every observable boundary.**
+   `invoke()` returns a redacted `observableResult`. The only path to an
+   unredacted secret is `invokeForModel()`, which returns a frozen capsule whose
+   `toJSON()` and default serialization always yield the redacted form; only an
+   explicit `readModelResult()` opens the raw value. The capsule uses a distinct
+   schema (`paperclip.phase7.model-tool-result.v1`) so it cannot be assigned to
+   an observable sink by mistake.
+2. **Path redaction rules.** `read_secret_value` redacts `$.value` to
+   `[SECRET_VALUE]` across output, error, and authorization-record channels;
+   `generic_api_request` redacts `$.headers.authorization` and
+   `$.headers.cookie` to `[REDACTED]` across all channels. The scenario runner
+   applies the same redaction again defensively at the explorer artifact
+   boundary and injects only a placeholder secret value.
+
+This is the security-gate outcome (PAP-16902) and its remediation (PAP-16909):
+a real secret value reaches the model surface only, never a trace, artifact, or
+browser view.
+
+## Running the tests
+
+```sh
+pnpm --filter @paperclipai/paperclip-runner exec vitest run \
+  src/tools/phase7-semantic-tools.test.ts
+```
+
+The "exposure and authorization" and "security policy" describes cover typed
+denials without protected state, model-only secret delivery, the self-approval
+prohibition, and the test-only, explicitly granted, allowlisted escape hatch.
+
+## Related
+
+- [Capability disposition](phase-07-capability-disposition.md)
+- [Semantic tool catalog](phase-07-semantic-tools.md)
+- [Scenario explorer](phase-07-scenario-explorer.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-capability-disposition.md b/packages/paperclip-runner/docs/phase-07-capability-disposition.md
new file mode 100644
index 00000000000..bd9c3b72e1e
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-capability-disposition.md
@@ -0,0 +1,85 @@
+# Phase 7 Capability Disposition
+
+This page explains how Phase 7 classifies every Paperclip capability. The
+classification itself lives in a generated file,
+[the Phase 7A capability contract](phase-07-capability-contract.md), which
+carries a "DO NOT EDIT" header and is rewritten by `generate:phase7-inventory`.
+Read this page to understand what the generated table means; read the table for
+the authoritative rows.
+
+## Normative sources
+
+Only two sources are normative:
+
+1. The Paperclip skill and its seven references (`SKILL.md` plus
+   `references/*.md`), contributing **152 headings**.
+2. The Paperclip Evals corpus, contributing **106 cases across 16 groups**.
+
+Together these produce **258 normative rows**. The legacy Paperclip MCP tool
+surface (**41 tools**) is not a production capability surface; each MCP name is
+folded one-to-one into a normative eval row as a traceability alias and inherits
+that row's disposition. The contract prints the alias index only so the
+normative target is easy to audit.
+
+## The three dispositions
+
+Every capability is classified as exactly one of:
+
+- **`control_plane_owned`** — the control plane performs this; no agent tool
+  exists for it. Checkout, inbox resolution, blocker diagnostics, and status
+  arbitration are control-plane-owned. In the explorer these appear as
+  control-plane actions labelled "no agent tool exists for this," never as a
+  callable tool.
+- **`always_agent_tool`** — the agent always has this as a semantic tool. Adding
+  a comment, writing a document, opening an interaction, and registering a
+  deliverable are always-agent-tool capabilities.
+- **`optional_agent_tool`** — the agent may have this tool, but only when a
+  grant unlocks it. Absent the grant, the capability is not exposed and calling
+  it is denied. See [authorization and exposure](phase-07-authorization-and-exposure.md).
+
+The disposition is what makes the boundary legible: it states, per capability,
+whether an agent can act, must wait for the control plane, or needs a grant
+first.
+
+## Reading the generated contract
+
+The contract has three parts:
+
+- **Baseline counts** — the heading, case, row, and alias totals above, plus the
+  per-group case table.
+- **Skill / reference rows** — one row per heading, its primary disposition, and
+  its `file:line` source anchor.
+- **Legacy MCP alias index** — each MCP name, the normative row it folds into,
+  the inherited disposition, and its `packages/mcp-server/src/tools.ts` anchor.
+
+## Regenerating and checking
+
+Generation reads the live in-repo skill/reference sources, the legacy MCP tool
+source, and the Paperclip Evals corpus, so it **requires** the external eval
+repository (via `PAPERCLIP_EVALS_ROOT` or a known local path) and is not part of
+the offline path. Checking and testing read only the checked-in derivatives
+under `spec/phase-07/` and need no external repository.
+
+```sh
+# Rewrite every generated file. Requires the external Paperclip Evals corpus.
+pnpm --filter @paperclipai/paperclip-runner generate:phase7-inventory
+
+# Validate counts, uniqueness, normative dispositions, one-to-one MCP folds,
+# required fields, and generated-file drift. Offline; no external eval repo.
+pnpm --filter @paperclipai/paperclip-runner check:phase7-inventory
+
+# Prove the validator rejects an independent MCP classification and rejects
+# missing, duplicate, or unknown MCP folds. Offline.
+pnpm --filter @paperclipai/paperclip-runner test:phase7-inventory
+```
+
+`check:phase7-inventory` diffs the checked-in generated files against what the
+live in-repo sources imply and fails on any drift or stale anchor, so the
+contract cannot silently fall out of sync.
+
+## Related
+
+- [Semantic tool catalog](phase-07-semantic-tools.md)
+- [Authorization and exposure](phase-07-authorization-and-exposure.md)
+- [Eval conformance](phase-07-eval-conformance.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-eval-conformance.md b/packages/paperclip-runner/docs/phase-07-eval-conformance.md
new file mode 100644
index 00000000000..76f71ab8c57
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-eval-conformance.md
@@ -0,0 +1,83 @@
+# Phase 7 Eval-Derived Conformance
+
+Phase 7E turns the Paperclip Evals corpus into an executable, offline
+conformance suite. The suite derives its cases from the checked-in Phase 7
+traceability derivative (`spec/phase-07/eval-traceability.yaml`) — it does
+**not** clone or read an external eval repository, and it starts only the
+in-process mock control plane. It hard-fails unless the derivative declares
+schema version 2, exactly 106 rows, exactly 16 groups, and unique case IDs.
+
+Sources: `src/conformance/phase7-eval-suite.ts` and its test
+`src/conformance/phase7-eval-suite.test.ts`; the reporter
+`scripts/run-phase7-eval-suite.mjs`.
+
+## The corpus
+
+- **106 cases across 16 groups.** Per-group counts: hb 5, co 6, st 8, cm 6,
+  se 4, su 4, bl 5, dp 3, ix 9, ap 6, ar 4, er 9, rf 22, mh 4, rs 3, wk 8.
+- Each case declares an actor role, a task mode, an input scenario, and the
+  expected outcome, all bound to a row in
+  [the capability contract](phase-07-capability-contract.md).
+
+## Assertion classes
+
+Every case belongs to one assertion class, and each class checks a different
+kind of invariant:
+
+- **`agent_tool_contract`** — a semantic tool call produces the expected typed
+  effect and state change.
+- **`authorization_policy`** — a capability is exposed, denied, or unlocked by a
+  grant exactly as its disposition requires.
+- **`control_plane_invariant`** — a control-plane-owned action happens without
+  any agent tool, and no tool can perform it.
+- **`combined_multi_hop`** — a sequence of operations across turns produces the
+  expected cumulative state and respects forbidden-operation rules.
+- **`restraint_no_call`** — the correct behavior is to make **no** call; the
+  case passes only if the agent deliberately does nothing further.
+
+## Fake-agent matrix and bounded Codex sample
+
+The suite executes each case with a deterministic fake agent whose plan is
+fixed by a fixture seed, so repeat runs are byte-identical. Optional-tool rows
+are run twice — once with the unlocking grants (must be allowed and must mutate
+state) and once ungranted (must be absent or denied with no state change);
+control-plane-owned rows remain absent in every configuration; and
+`restraint_no_call` rows must produce an empty state diff.
+
+The fake-agent surface is 14 always-agent tools plus 4 optional tools unlocked
+by four seed grants (`discovery:tasks:read`, `discovery:agents:read`,
+`delegation:tasks:create`, `governance:approvals:request`) — **18 operations**.
+The suite binds that surface through both the fake-agent and Codex bindings and
+asserts the two operation lists are byte-identical (**18/18**).
+
+A **bounded Codex binding sample** picks one representative case from nine
+groups (`hb`, `dp`, `bl`, `ap`, `ar`, `ix`, `mh`, `rs`, `wk`) and checks that
+`checkout_task` is absent from the Codex surface while every other sampled
+operation is present. It is an offline parity check; no real Codex or network is
+contacted, and the browser explorer holds no credential.
+
+## Running it
+
+```sh
+# Run the 106-case suite in-process (one vitest file drives all cases).
+pnpm --filter @paperclipai/paperclip-runner test:phase7-evals
+
+# Build the public surface and write the parity report with per-group counts,
+# assertion classes, the fake-agent matrix, the bounded Codex sample, and the
+# semantic-operation execution counts.
+pnpm --filter @paperclipai/paperclip-runner report:phase7-evals
+```
+
+The reporter writes `knowledge/evidence/phase-07/eval-parity-report.{json,md}`.
+Each failure carries its case ID, assertion class, semantic operation,
+authorization decision, and final state diff. The report is generated on demand
+and is not committed; delete it before running `docs:validate` (it carries no
+OKF frontmatter). See the
+[verification commands reference](phase-07-verification-commands.md).
+
+## Related
+
+- [Capability disposition](phase-07-capability-disposition.md)
+- [Semantic tool catalog](phase-07-semantic-tools.md)
+- [Scenario explorer](phase-07-scenario-explorer.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-future-binding-boundary.md b/packages/paperclip-runner/docs/phase-07-future-binding-boundary.md
new file mode 100644
index 00000000000..772cb35bd38
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-future-binding-boundary.md
@@ -0,0 +1,52 @@
+# Phase 7 Future Binding Boundary (Phase 8 / ACPX)
+
+Phase 7 is a package-local model. It does not integrate the runner into
+Paperclip, and it must not. This page states exactly what Phase 7 defers to
+Phase 8 and what clean seam it preserves so the deferral stays cheap.
+
+## What Phase 7 does not do
+
+Nothing in Phase 7 contacts a real Paperclip control plane, database, ACPX
+session, or provider credential. The 106-case conformance suite and the browser
+explorer run entirely against the in-process
+[mock ControlPlanePort](phase-07-mock-control-plane-port.md) and checked-in
+fixtures. The [forbidden-imports boundary](architecture.md) rejects any import
+that crosses into `server/`, `ui/`, `cli/`, `@paperclipai/db`, or other
+Paperclip workspace internals.
+
+## The preserved seam
+
+The dependency arrow always points from an implementation toward a contract.
+The package owns two contracts a future consumer implements:
+
+- **`ControlPlanePort`** — the narrow surface through which a runner opens a
+  run, appends ordered events, records semantic operations, and submits a
+  terminal result. Phase 7 injects the *mock* adapter behind this port; Phase 8
+  injects a real one. The catalog, authorization engine, and conformance suite
+  bind to the port, not to any adapter.
+- **`NativeSessionBackend`** — the normalized session surface for a future
+  control-plane consumer.
+
+Because the port is the only coupling point, Phase 8 replaces the adapter
+without touching the tool catalog, the authorization rules, or the eval-derived
+conformance suite. The package remains independently buildable, testable, and
+runnable against the mock adapter after the real one exists.
+
+## What Phase 8 (ACPX) will bind
+
+Phase 8 binds a real Paperclip `ControlPlanePort` implementation behind the same
+seam so the semantic tools and authorization engine act against a live control
+plane instead of the mock. That work is out of scope here and requires separate
+CTO approval at the Phase 7 checkpoint (`PAP-16908`). Until then:
+
+- ACPX is Phase 8, not Phase 7.
+- No Phase 7 documentation claims Phase 8 capability.
+- Real integration is a separately reviewed phase, consistent with the
+  [future integration rule](architecture.md) recorded for every prior phase.
+
+## Related
+
+- [Architecture and dependency boundary](architecture.md)
+- [Mock ControlPlanePort](phase-07-mock-control-plane-port.md)
+- [Capability disposition](phase-07-capability-disposition.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-mock-control-plane-port.md b/packages/paperclip-runner/docs/phase-07-mock-control-plane-port.md
new file mode 100644
index 00000000000..3add9e4526f
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-mock-control-plane-port.md
@@ -0,0 +1,95 @@
+# Phase 7 Mock ControlPlanePort
+
+`Phase7MockControlPlaneAdapter` is a deterministic, serializable, in-memory
+model of Paperclip's control-plane **semantics** — not its transport. It imports
+no server route, service, database, or provider binding, and it holds no
+credential.
+
+Sources: `src/mock-core/phase7-mock-control-plane-adapter.ts`,
+`src/mock-core/phase7-control-plane-types.ts`. Shared contract:
+`src/conformance/control-plane-port.ts`.
+
+## Entity domains
+
+The adapter models ten renderable entity domains, which are exactly the domains
+the [scenario explorer](phase-07-scenario-explorer.md) diffs:
+
+`tasks`, `comments`, `documents`, `interactions`, `approvals`, `artifacts`
+(artifacts and work products), `blockers`, `workspace` (workspace services),
+`budget`, and `run`.
+
+The fixture state also carries supporting records — company, actors, wakes,
+audit, decisions, idempotency, faults, counters, and a fixture clock — that back
+behavior but are not diff domains.
+
+## Operations
+
+- **Lifecycle:** `start()` / `stop()` only change local object state.
+- **Runs:** `openRun()` / `openFixtureRun()` return a run context with atomic
+  checkout; `context(runId)` reads it back.
+- **Events:** `appendEvent()` accepts both canonical PRP events and native run
+  events with deduplication, replay-conflict detection, and contiguous-sequence
+  tracking; `replayEvents()` replays from an exclusive cursor.
+- **Commands:** `applyCommand(envelope)` executes a semantic command with
+  idempotency and an optimistic `expectedRevision`.
+- **Sessions:** `loadSessionCheckpoint()` / `checkpointSession()`.
+- **Terminal:** `completeRun()` reconciles a terminal disposition — `done`,
+  `blocked`, `needs_review`, or `yielded` — and fans out blocker wakes.
+- **Introspection:** `snapshot()`, `decisionRecords()`, `serialize()`, and
+  static `restore()`.
+
+The eighteen semantic command kinds (`report_progress`, `write_document`,
+`request_human_input`, `resolve_human_input`, `register_deliverable`,
+`set_dependencies`, `create_task`, `request_approval`, `decide_approval`,
+`comment_on_approval`, `control_workspace_service`, `record_budget_usage`,
+`finish_task`, `block_task`, `request_review`, `release_task`, `schedule_wake`)
+are the mock side of the [semantic tool catalog](phase-07-semantic-tools.md).
+
+## Determinism
+
+- **Fixture clock.** Every timestamp derives from `clock.epochMs + tick*1000`
+  and the tick increments on each read. The default epoch is
+  `2026-08-09T00:00:00Z`. There is no RNG anywhere.
+- **Deterministic IDs.** Every generated ID is `fixture-<kind>-<nnnn>` from a
+  monotonic per-kind counter.
+- **Immutability.** Inputs and outputs are `structuredClone`d and every returned
+  value is deep-frozen. Canonical JSON with sorted keys backs every dedupe and
+  replay comparison.
+- **Serializable.** `serialize()` / `restore()` round-trip under the schema tag
+  `paperclip.phase7.mock-state.v1`.
+- **Scripted faults.** A fault rule injects a bounded number of
+  `retryable_error` or `lost_ack` effects; a lost-ack command commits once and
+  returns its stored result on retry.
+
+## Boundary
+
+The adapter is a pure state machine: no network, no database, no provider SDK,
+no credentials. Secrets appearing in wake payloads are redacted against
+`/authorization|credential|api.?key|secret|token/i`; the adapter test seeds
+`apiKey` and `Bearer` values and asserts they never surface. See
+[authorization and exposure](phase-07-authorization-and-exposure.md).
+
+## Shared conformance
+
+The adapter satisfies the same `runControlPlanePortConformance` contract used by
+earlier phases: it rejects open-binding violations, recovers from source gaps,
+treats duplicates idempotently, and rejects event-id, sequence, replay-binding,
+and result mutations (failing closed with `native_event_replay_conflict`).
+
+## Running the tests
+
+```sh
+pnpm --filter @paperclipai/paperclip-runner exec vitest run \
+  src/conformance/control-plane-port.test.ts \
+  src/mock-core/phase7-mock-control-plane-adapter.test.ts
+```
+
+Nine tests, no npm alias of their own — they also run inside
+`pnpm --filter @paperclipai/paperclip-runner test`.
+
+## Related
+
+- [Semantic tool catalog](phase-07-semantic-tools.md)
+- [Authorization and exposure](phase-07-authorization-and-exposure.md)
+- [Future binding boundary](phase-07-future-binding-boundary.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-scenario-explorer.md b/packages/paperclip-runner/docs/phase-07-scenario-explorer.md
new file mode 100644
index 00000000000..4e5fe85a0bc
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-scenario-explorer.md
@@ -0,0 +1,86 @@
+# Phase 7 Browser Scenario Explorer
+
+The scenario explorer is a read-only browser surface over all 106 conformance
+cases. It runs each scenario in the page against the
+[mock control plane](phase-07-mock-control-plane-port.md) through the
+[semantic tool runtime](phase-07-semantic-tools.md) and renders the resulting
+run artifact. It never re-judges parity, never leaves its own origin, and holds
+no credential.
+
+Sources: `src/phase7/` (scenario index, runner, parity, state diff) and
+`examples/phase7-explorer/`. Interaction contract:
+[Phase 7 scenario explorer UX](design/phase-7-scenario-explorer-ux.md).
+
+## The run artifact
+
+Running a scenario produces one immutable artifact carrying:
+
+- **Tool exposure** — which semantic tools were available, and for optional
+  tools, the grant that unlocked each one.
+- **Control-plane actions** — control-plane-owned steps (for example checkout),
+  each labelled "no agent tool exists for this."
+- **Authorization records** — allow and typed-deny decisions; a denial carries
+  the missing claim and no protected task data.
+- **State diff** — an immutable diff over the ten entity domains, with unchanged
+  domains collapsed.
+- **Parity verdict** — the runtime's own verdict, plus Phase 7E's per-case
+  result carried through in a separately labelled block. The explorer displays
+  these; it does not recompute them.
+
+## Read-only and frozen SDK
+
+The explorer imports the frozen **0.1.2** public SDK through the package-local
+alias `@paperclip-runner-local/phase7`, which is deliberately not the published
+package name. It adds no SDK export and changes no published surface. Fake mode
+runs entirely in the page against checked-in fixtures and renders fixture time
+only, so two loads of the same route produce identical settled DOM.
+
+## What the explorer shows
+
+- **Home** — 16 group facets whose counts sum to 106, corpus stats, and example
+  deep links.
+- **Picker** — a listbox with live filter chips, disabled zero-count values, and
+  a clear-filters control.
+- **Transcript** — both channels (agent-visible and control-plane).
+- **Inspector** — four tabs: context (exposure and grants), authorization
+  (allow/deny), state diff, and traceability.
+
+## Boundary
+
+- No network request leaves the explorer's own origin.
+- `localStorage` holds no run artifact, grant, or fixture payload.
+- Codex mode is a disabled option with a stated reason; wiring it to the Phase 5
+  relay is deferred, and the browser holds no provider credential either way.
+- Secrets are redacted before display; redaction chips name the rule and never
+  the value. See [authorization and exposure](phase-07-authorization-and-exposure.md).
+
+## Accessibility
+
+The 7F browser suite asserts: listbox arrow-key navigation with
+`aria-activedescendant`, WAI-ARIA tab arrow-key activation with exactly one
+tablist, every interactive control named, a polite live region announcing the
+settled verdict, three landmarks, a single `h1`, and — at 390px — one segment at
+a time with zero horizontal overflow.
+
+## Running it
+
+```sh
+# Open the explorer on 127.0.0.1:4183.
+pnpm --filter @paperclipai/paperclip-runner demo:phase7
+
+# Scenario runtime, explorer components, and route determinism (45 tests).
+pnpm --filter @paperclipai/paperclip-runner test:phase7
+
+# Browser IA, determinism, evidence routes, boundary, a11y, responsive (19).
+pnpm --filter @paperclipai/paperclip-runner test:browser:phase7
+
+# Deterministic 24-image acceptance set (12 routes x 2 viewports).
+pnpm --filter @paperclipai/paperclip-runner record:phase7
+```
+
+## Related
+
+- [Eval conformance](phase-07-eval-conformance.md)
+- [Verification commands](phase-07-verification-commands.md)
+- [Phase 7F explorer verification](../knowledge/evidence/2026-08-09-phase-07f-explorer-verification.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-semantic-tools.md b/packages/paperclip-runner/docs/phase-07-semantic-tools.md
new file mode 100644
index 00000000000..0578be33296
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-semantic-tools.md
@@ -0,0 +1,87 @@
+# Phase 7 Semantic Tool Catalog
+
+The semantic tool catalog is the transport-neutral surface an agent may call. It
+is a frozen set of JSON-schema descriptors — no HTTP method, path, or provider
+detail. Provider-specific shapes are produced only by *bindings*, so every
+provider sees the same operation set.
+
+Sources: `src/tools/phase7-semantic-tool-catalog.ts`,
+`src/tools/phase7-semantic-tool-types.ts`, `src/tools/phase7-tool-bindings.ts`,
+barrel `src/tools/index.ts`.
+
+## Two dispositions, plus a separate control-plane list
+
+A tool descriptor's `disposition` is either **`always_agent_tool`** or
+**`optional_agent_tool`**. Control-plane-owned operations are a separate frozen
+list, not a disposition — they are never exposed as tools. See
+[capability disposition](phase-07-capability-disposition.md) and
+[authorization and exposure](phase-07-authorization-and-exposure.md).
+
+The catalog holds **36 tools**: 14 always-agent tools and 22 optional tools
+across 10 groups.
+
+### Always-agent tools (14)
+
+`get_task_context`, `get_task_history`, `list_documents`, `read_document`,
+`list_document_revisions`, `report_progress`, `answer_status_question`,
+`finish_task`, `block_task`, `request_review`, `write_document`,
+`request_human_input`, `register_deliverable`, `inspect_operation_result`.
+
+### Optional tools (22), by group
+
+| Group | Tools |
+| --- | --- |
+| discovery | `search_tasks`, `list_agents`, `list_projects`, `list_goals` |
+| delegation_dependencies | `create_task`, `set_dependencies` |
+| governance | `list_approvals`, `request_approval`, `decide_approval`, `comment_on_approval` |
+| cases | `list_cases`, `upsert_case` |
+| workspace_runtime | `get_workspace_runtime`, `control_workspace_service` |
+| routines | `list_routines`, `manage_routine` |
+| company_skills | `list_company_skills`, `sync_company_skills` |
+| secrets | `list_secret_metadata`, `read_secret_value` |
+| portability_admin | `export_company`, `administer_company` |
+| test_escape_hatch | `generic_api_request` |
+
+## Descriptor shape
+
+Each descriptor carries `operationId`, `version` (1), `title`, `description`,
+input/output JSON schemas, `disposition`, an optional `optionalGroup`,
+`requiredClaims`, and optionally `allowedRoles`, `taskModes`, a `sideEffectClass`
+(`read`, `task_write`, `company_write`, `governance`, `workspace_control`,
+`secret_read`, `admin`, `test_escape_hatch`), an `idempotency` level, `redaction`
+rules, and an abstract `mockCommandMapping`.
+
+The `mockCommandMapping` is one of `context_read`, `snapshot_read`,
+`semantic_command`, `operation_result`, or `mock_extension` — describing *what*
+the tool does against the mock, never *how* a transport would carry it.
+
+## Transport neutrality
+
+Bindings, not descriptors, produce provider shapes, both derived from the same
+`visibleTools.tools` array:
+
+- `Phase7FakeAgentToolBinding` emits `{operationId, description, inputSchema,
+  outputSchema}`.
+- `Phase7CodexToolBinding` emits `{type: "function", name, description, strict:
+  true, parameters}`.
+
+Because both bindings derive from one array, the fake-agent and Codex operation
+surfaces are byte-identical. The [conformance suite](phase-07-eval-conformance.md)
+asserts this parity (the 18/18 fake-agent/Codex operation matrix).
+
+## Running the tests
+
+```sh
+pnpm --filter @paperclipai/paperclip-runner exec vitest run \
+  src/tools/phase7-semantic-tools.test.ts
+```
+
+The "catalog" describe asserts a unique, versioned, provider-neutral catalog and
+that every optional group is present.
+
+## Related
+
+- [Authorization and exposure](phase-07-authorization-and-exposure.md)
+- [Mock ControlPlanePort](phase-07-mock-control-plane-port.md)
+- [Eval conformance](phase-07-eval-conformance.md)
+</content>
diff --git a/packages/paperclip-runner/docs/phase-07-verification-commands.md b/packages/paperclip-runner/docs/phase-07-verification-commands.md
new file mode 100644
index 00000000000..054125b4318
--- /dev/null
+++ b/packages/paperclip-runner/docs/phase-07-verification-commands.md
@@ -0,0 +1,42 @@
+# Phase 7 Verification Commands
+
+Every command here runs from the repository root, is offline and deterministic,
+starts no Paperclip service, and holds no credential. All are prefixed
+`pnpm --filter @paperclipai/paperclip-runner`. The observed results for the
+acceptance record are in
+[the Phase 7 verification evidence](../knowledge/evidence/2026-08-09-phase-07-verification.md).
+
+| Surface | Command | Expected result |
+| --- | --- | --- |
+| Capability contract completeness and drift | `check:phase7-inventory` | `Phase 7 inventory completeness and generated-output checks passed.` |
+| Contract validator negatives | `test:phase7-inventory` | 4 tests pass |
+| Mock control plane + shared port | `exec vitest run src/conformance/control-plane-port.test.ts src/mock-core/phase7-mock-control-plane-adapter.test.ts` | 9 tests pass |
+| Semantic tools + authorization/redaction | `exec vitest run src/tools/phase7-semantic-tools.test.ts` | 9 tests pass |
+| 106-case conformance | `test:phase7-evals` | 1 file (all 106 cases) passes |
+| Fake-agent matrix + bounded Codex + parity report | `report:phase7-evals` | `Phase 7 eval conformance passed: 106 cases across 16 groups.` |
+| Scenario runtime + explorer + routes | `test:phase7` | 45 tests pass |
+| Browser IA, accessibility, determinism, boundary | `test:browser:phase7` | 19 Playwright tests pass |
+| Screenshot acceptance set | `record:phase7` | 24 deterministic images |
+| Doc links + OKF bundle | `docs:validate` | 62 files, 26 OKF concepts, 5 indexes |
+
+## Notes
+
+- **No Rust toolchain is needed** for any Phase 7 command. The full
+  `verify` target still builds the Rust workspace, but nothing above does.
+- **Determinism.** Fake-mode runs render fixture time only, so repeat runs and
+  repeat screenshots are byte-identical.
+- **The parity report is generated on demand.** `report:phase7-evals` writes
+  `knowledge/evidence/phase-07/eval-parity-report.{json,md}`, which are not
+  committed. Because the `.md` carries no OKF frontmatter, run `docs:validate`
+  **before** generating the report, or delete the report first. This dirty-tree
+  interaction is tracked for the Phase 7H QA gate (`PAP-16907`).
+- **Browser libraries.** On minimal or rootless hosts, extract Playwright's
+  browser libraries first; the package's `verify:rootless` target shows the
+  pattern.
+
+## Related
+
+- [Clean-start tutorial](tutorials/phase-07-scenario-explorer.md)
+- [Eval conformance](phase-07-eval-conformance.md)
+- [Scenario explorer](phase-07-scenario-explorer.md)
+</content>
diff --git a/packages/paperclip-runner/docs/tutorials/phase-07-scenario-explorer.md b/packages/paperclip-runner/docs/tutorials/phase-07-scenario-explorer.md
new file mode 100644
index 00000000000..0b566d53e43
--- /dev/null
+++ b/packages/paperclip-runner/docs/tutorials/phase-07-scenario-explorer.md
@@ -0,0 +1,188 @@
+# Phase 7 Clean-Start Tutorial: Capability Contract, Mock Control Plane, and Scenario Explorer
+
+**Time to first success: about 5 minutes.** Two commands take you from a clean
+checkout to 106 passing conformance cases and a browser you can click through.
+The whole tutorial runs from the repository root. It starts no Paperclip
+service, contacts no Paperclip control plane, clones no external eval
+repository, and holds no provider credential. Everything it needs is checked in
+under `packages/paperclip-runner/`.
+
+Phase 7 does not integrate the runner into Paperclip. It builds a package-local
+model of what a native Paperclip run *would* do — a deterministic mock control
+plane, a transport-neutral semantic tool catalog, an authorization engine, a
+106-case conformance suite derived from the Paperclip Evals corpus, and a
+read-only browser explorer over all of it. Real integration is Phase 8 and
+requires separate approval. See
+[the future binding boundary reference](../phase-07-future-binding-boundary.md).
+
+## What you need
+
+- Node.js 20 or newer and pnpm 9 or newer. This tutorial was verified with Node
+  22.22.2 and pnpm 9.15.4.
+- No Rust toolchain. Every command in this tutorial is TypeScript/Node only. (A
+  full `verify` still builds Rust, but nothing here does.)
+- No network access after `pnpm install`.
+
+Install the package workspace from the repository root:
+
+```sh
+pnpm install --filter @paperclipai/paperclip-runner --lockfile=false --offline --ignore-scripts --dev
+```
+
+## 1. Prove the 106-case conformance suite (about 1 minute)
+
+```sh
+pnpm --filter @paperclipai/paperclip-runner test:phase7-evals
+```
+
+This runs the eval-derived conformance suite entirely in-process against the
+mock control plane. Expected final line:
+
+```
+ Test Files  1 passed (1)
+      Tests  1 passed (1)
+```
+
+The single test file drives all 106 cases across the 16 eval groups. To see the
+per-group counts, assertion classes, the fake-agent operation matrix, and the
+bounded Codex binding sample, generate the parity report:
+
+```sh
+pnpm --filter @paperclipai/paperclip-runner report:phase7-evals
+```
+
+Expected final line:
+
+```
+Phase 7 eval conformance passed: 106 cases across 16 groups.
+```
+
+The report is written to
+`knowledge/evidence/phase-07/eval-parity-report.{json,md}`. It is a
+generated-on-demand artifact, not a committed file; a clean checkout does not
+contain it. Delete it before you run `pnpm --filter @paperclipai/paperclip-runner docs:validate`
+(the report carries no OKF frontmatter and would otherwise fail the knowledge
+bundle check — see [Known gaps](#known-gaps-and-boundaries)).
+
+The 16 groups and case counts are fixed by the capability contract:
+
+| Group | Cases | Group | Cases | Group | Cases | Group | Cases |
+| --- | ---: | --- | ---: | --- | ---: | --- | ---: |
+| hb | 5 | co | 6 | st | 8 | cm | 6 |
+| se | 4 | su | 4 | bl | 5 | dp | 3 |
+| ix | 9 | ap | 6 | ar | 4 | er | 9 |
+| rf | 22 | mh | 4 | rs | 3 | wk | 8 |
+
+## 2. Open the scenario explorer (about 2 minutes)
+
+```sh
+pnpm --filter @paperclipai/paperclip-runner demo:phase7
+```
+
+Open `http://127.0.0.1:4183/phase7-explorer/`. The explorer is read-only. It
+runs each scenario in the browser against the same mock control plane and
+renders the run artifact; it never re-judges parity and never leaves its own
+origin. Walk this path:
+
+1. **Home.** Confirm 16 group facets whose counts sum to 106.
+2. **Pick a scenario.** Filter to `ap` and open `ap-mcp-gate-01`. The
+   authorization view shows a deny row and a "1 deny" tab chip; the typed
+   denial card carries no task data.
+3. **Read a heartbeat.** Open `hb-context-01`. The control-plane checkout entry
+   is labelled "no agent tool exists for this" — it is control-plane-owned, not
+   a semantic tool.
+4. **Inspect a manager scenario.** Open `rf-api-mgr-heartbeat-01`. Optional
+   tools are listed with the grant that unlocked them, alongside the
+   control-plane "no tool" list and a traceability panel.
+5. **Confirm the credential boundary.** Codex mode is a disabled option with a
+   stated reason. `localStorage` holds no run artifact or grant, and the page
+   makes no request beyond its own assets.
+
+The explorer imports the frozen `0.1.2` SDK through the package-local
+`@paperclip-runner-local/phase7` alias — deliberately not the published package
+name. See [the browser explorer reference](../phase-07-scenario-explorer.md).
+
+## 3. Read what each surface guarantees
+
+The explorer renders four surfaces produced by the runtime. Each has a
+package-local reference page:
+
+- [Capability disposition](../phase-07-capability-disposition.md) — how every
+  skill/reference behavior and eval case is classified as control-plane-owned,
+  always-agent-tool, or optional-agent-tool, and how the generated contract is
+  produced and checked.
+- [Mock ControlPlanePort](../phase-07-mock-control-plane-port.md) — the
+  deterministic in-memory adapter, its entity domains, and its boundary.
+- [Semantic tool catalog](../phase-07-semantic-tools.md) — the
+  transport-neutral tools an agent may call.
+- [Authorization and exposure](../phase-07-authorization-and-exposure.md) —
+  grants, typed denials, and secret redaction.
+- [Eval conformance](../phase-07-eval-conformance.md) — how the 106 cases are
+  derived from the checked-in traceability derivative with no external repo.
+
+## 4. Run the full focused verification set
+
+Every command below is offline, deterministic, and package-local. The
+[verification commands reference](../phase-07-verification-commands.md) lists
+each one with its purpose and expected result.
+
+```sh
+# Capability contract: completeness, uniqueness, one-to-one MCP folds, no drift.
+pnpm --filter @paperclipai/paperclip-runner check:phase7-inventory
+pnpm --filter @paperclipai/paperclip-runner test:phase7-inventory
+
+# Mock control plane adapter and shared ControlPlanePort conformance.
+pnpm --filter @paperclipai/paperclip-runner exec vitest run \
+  src/conformance/control-plane-port.test.ts \
+  src/mock-core/phase7-mock-control-plane-adapter.test.ts
+
+# Semantic tool catalog and authorization/redaction engine.
+pnpm --filter @paperclipai/paperclip-runner exec vitest run \
+  src/tools/phase7-semantic-tools.test.ts
+
+# 106-case conformance and the parity/fake-agent/bounded-Codex report.
+pnpm --filter @paperclipai/paperclip-runner test:phase7-evals
+pnpm --filter @paperclipai/paperclip-runner report:phase7-evals
+
+# Scenario runtime, explorer components, and route determinism (45 tests).
+pnpm --filter @paperclipai/paperclip-runner test:phase7
+
+# Browser information architecture, accessibility, determinism, and boundary.
+pnpm --filter @paperclipai/paperclip-runner test:browser:phase7
+
+# Deterministic 24-image screenshot acceptance set.
+pnpm --filter @paperclipai/paperclip-runner record:phase7
+
+# Documentation links and the OKF knowledge bundle (run with no generated
+# report present).
+pnpm --filter @paperclipai/paperclip-runner docs:validate
+```
+
+The exact results captured for the acceptance record are in the
+[Phase 7 verification evidence](../../knowledge/evidence/2026-08-09-phase-07-verification.md).
+
+## Known gaps and boundaries
+
+- **The parity report breaks the OKF bundle if left in place.**
+  `report:phase7-evals` writes `eval-parity-report.md` into the knowledge
+  bundle without OKF frontmatter, so `docs:validate` fails while it is present.
+  Delete it before validating. A clean checkout never contains it. This is
+  recorded for the Phase 7H QA gate.
+- **Codex mode is disabled.** The explorer offers a bounded real-Codex binding
+  sample only through `report:phase7-evals`; the browser holds no credential
+  and the in-page Codex mode is inert with a stated reason.
+- **No Paperclip integration.** Nothing here touches a real control plane,
+  database, or provider. Phase 8 (ACPX) binds the real adapter behind the same
+  seam and requires separate CTO approval.
+
+## Where this sits
+
+- Plan: Phase 7 plan, issue `PAP-16897` (see its `#document-plan` document).
+- Child issues: 7A capability inventory (PAP-16898), 7B UX map (PAP-16899),
+  7C mock adapter (PAP-16900), 7D semantic tools and authorization (PAP-16901,
+  security gate PAP-16902, remediation PAP-16909), 7E conformance suite
+  (PAP-16903), 7F browser explorer (PAP-16904, UX gate PAP-16905), 7G docs and
+  evidence (PAP-16906), 7H clean-room QA (PAP-16907), Phase 7 checkpoint
+  (PAP-16908).
+</content>
+</invoke>
diff --git a/packages/paperclip-runner/knowledge/evidence/2026-08-09-phase-07-verification.md b/packages/paperclip-runner/knowledge/evidence/2026-08-09-phase-07-verification.md
new file mode 100644
index 00000000000..fbc0cbaef1e
--- /dev/null
+++ b/packages/paperclip-runner/knowledge/evidence/2026-08-09-phase-07-verification.md
@@ -0,0 +1,105 @@
+---
+type: Verification Evidence
+title: Phase 7 documentation, tutorial, and evidence manifest
+description: Package-local acceptance manifest cross-linking every focused Phase 7 verification command, its observed result, the parity/screenshot artifacts, and the security, UX, and QA gate issues.
+tags: [native-runner, phase-7, verification, conformance, evidence, manifest]
+status: stable
+generated: { by: anthropic/claude-opus-4-8, at: 2026-08-09T13:42:50Z }
+---
+
+# Scope
+
+This record is the Phase 7G evidence manifest (PAP-16906). It links each
+package-local, offline, deterministic verification command to the result
+observed on 2026-08-09, and to the reference pages, tutorial, parity report,
+and screenshot set that describe the Phase 7 surfaces.
+
+Phase 7 contacts no Paperclip service, ACPX session, database, or provider
+credential. Every command below runs from the repository root against
+checked-in fixtures and the in-process mock control plane.
+
+# Environment
+
+- Node.js 22.22.2, pnpm 9.15.4. No Rust toolchain is required for any Phase 7
+  focused command.
+- vitest 4.1.10; Playwright with its bundled Chromium.
+- No network access after `pnpm install`.
+
+# Verification matrix
+
+All commands are prefixed `pnpm --filter @paperclipai/paperclip-runner`.
+Results were captured on 2026-08-09 between 13:37Z and 13:39Z.
+
+| Surface | Command | Observed result |
+| --- | --- | --- |
+| Capability contract completeness and drift | `check:phase7-inventory` | `Phase 7 inventory completeness and generated-output checks passed.` (exit 0) |
+| Contract validator negatives | `test:phase7-inventory` | 4 tests, 4 pass, 0 fail |
+| Mock control plane + shared port | `exec vitest run src/conformance/control-plane-port.test.ts src/mock-core/phase7-mock-control-plane-adapter.test.ts` | 2 files, 9 tests pass |
+| Semantic tools + authorization/redaction | `exec vitest run src/tools/phase7-semantic-tools.test.ts` | 1 file, 9 tests pass |
+| 106-case conformance | `test:phase7-evals` | 1 file, 1 test (drives all 106 cases) pass |
+| Fake-agent matrix + bounded Codex sample + parity report | `report:phase7-evals` | `Phase 7 eval conformance passed: 106 cases across 16 groups.` |
+| Scenario runtime + explorer components + routes | `test:phase7` | 3 files, 45 tests pass |
+| Browser IA, accessibility, determinism, boundary | `test:browser:phase7` | 19 Playwright tests pass (6.8s) |
+| Screenshot acceptance set | `record:phase7` | 24 deterministic images (12 routes × 2 viewports) |
+| Doc links + OKF bundle | `docs:validate` | `Documentation link validation passed (62 files).` and `OKF v0.2 validation passed (26 concepts, 5 indexes).` |
+
+# Conformance shape
+
+- 106 cases across 16 groups: hb 5, co 6, st 8, cm 6, se 4, su 4, bl 5, dp 3,
+  ix 9, ap 6, ar 4, er 9, rf 22, mh 4, rs 3, wk 8.
+- Assertion classes: `agent_tool_contract`, `authorization_policy`,
+  `combined_multi_hop`, `control_plane_invariant`, `restraint_no_call`.
+- Fake-agent/Codex operations: 18/18 executed.
+- Capability contract baseline: 152 skill/reference headings, 106 eval cases,
+  258 normative rows, 41 legacy MCP aliases folded one-to-one into normative
+  rows. See [the generated contract](../../docs/phase-07-capability-contract.md).
+
+# Artifacts
+
+- Parity report: generated on demand by `report:phase7-evals` into
+  `knowledge/evidence/phase-07/eval-parity-report.{json,md}`. It is not a
+  committed file; a clean checkout does not contain it, and the explorer renders
+  "Not run" when it is absent.
+- Screenshots: the committed 24-image acceptance set and its manifest are under
+  [Phase 7 evidence](phase-07/index.md).
+- Prior surface evidence:
+  [Phase 7F browser scenario explorer verification](2026-08-09-phase-07f-explorer-verification.md).
+
+# Documentation delivered
+
+- Tutorial: [Phase 7 clean-start tutorial](../../docs/tutorials/phase-07-scenario-explorer.md).
+- Reference pages:
+  [capability disposition](../../docs/phase-07-capability-disposition.md),
+  [mock ControlPlanePort](../../docs/phase-07-mock-control-plane-port.md),
+  [semantic tools](../../docs/phase-07-semantic-tools.md),
+  [authorization and exposure](../../docs/phase-07-authorization-and-exposure.md),
+  [eval conformance](../../docs/phase-07-eval-conformance.md),
+  [scenario explorer](../../docs/phase-07-scenario-explorer.md),
+  [future binding boundary](../../docs/phase-07-future-binding-boundary.md),
+  [verification commands](../../docs/phase-07-verification-commands.md).
+
+# Gate issues
+
+- Security gate: `PAP-16902` (authorization and credential isolation review),
+  with remediation `PAP-16909` (redact secret invocation results before trace
+  or browser exposure). The `rs-secret-redaction` case and the
+  `src/tools/phase7-semantic-tools.test.ts` redaction assertions exercise the
+  remediated path.
+- UX gate: `PAP-16905` (explorer acceptance and screenshots), against the
+  approved interaction map §9 twelve-route matrix.
+- QA gate: `PAP-16907` (clean-room QA and human checkpoint), which consumes this
+  manifest and the tutorial from a clean checkout.
+
+# Known gaps
+
+- `report:phase7-evals` writes `eval-parity-report.md` into the knowledge bundle
+  without OKF frontmatter, and `evidence/phase-07/index.md` does not link it, so
+  `docs:validate` fails while the generated report is present. Delete the report
+  before validating, or run validation first. Recorded for the Phase 7H QA gate
+  as a candidate to relocate the generated report outside the bundle or exclude
+  it from the OKF scan.
+- Codex mode in the explorer is disabled with a stated reason; the only real
+  provider exercise is the bounded binding sample in `report:phase7-evals`.
+- No Paperclip control plane, database, or provider is contacted anywhere in
+  Phase 7. Real integration is Phase 8 (ACPX) and requires separate approval.
+</content>
diff --git a/packages/paperclip-runner/knowledge/evidence/index.md b/packages/paperclip-runner/knowledge/evidence/index.md
index 2e427681a9b..11cef188db5 100644
--- a/packages/paperclip-runner/knowledge/evidence/index.md
+++ b/packages/paperclip-runner/knowledge/evidence/index.md
@@ -14,4 +14,5 @@
 * [Phase 5 browser SDK verification](2026-08-08-phase-05-verification.md) - Targeted and package-acceptance evidence for the versioned SDK, both public consumers, fake/real drivers, accessibility, reconnect, replay, and screenshots.
 * [Phase 6 thin Paperclip adapter verification](2026-08-09-phase-06-verification.md) - Mock and database-backed conformance, authoritative completion, restart recovery, migration repair, atomic liveness, internal canary, and deterministic post-kill-switch legacy evidence.
 * [Phase 7F browser scenario explorer verification](2026-08-09-phase-07f-explorer-verification.md) - Scenario index, run artifacts, exposure, authorization, state diff, parity, route determinism, accessibility, and the 24-image screenshot acceptance set.
+* [Phase 7 documentation, tutorial, and evidence manifest](2026-08-09-phase-07-verification.md) - Command→result matrix for every focused Phase 7 verification, artifact links, and the security, UX, and QA gate issues.
 * [Phase 7 evidence](phase-07/) - Screenshot acceptance captures and the eval conformance parity report.
diff --git a/packages/paperclip-runner/knowledge/journal/2026-08-09-phase-07.md b/packages/paperclip-runner/knowledge/journal/2026-08-09-phase-07.md
new file mode 100644
index 00000000000..43ba3b4dafd
--- /dev/null
+++ b/packages/paperclip-runner/knowledge/journal/2026-08-09-phase-07.md
@@ -0,0 +1,102 @@
+---
+type: Engineering Journal Entry
+title: Phase 7 reference docs, tutorial, evidence, and OKF
+description: Decisions, observed facts, gaps, and readiness evidence for the package-local Phase 7 documentation set and clean-start tutorial.
+tags: [native-runner, phase-7, documentation, tutorial, evidence, okf]
+status: stable
+generated: { by: anthropic/claude-opus-4-8, at: 2026-08-09T13:42:50Z }
+entry_kind: phase
+phase: "7"
+---
+
+# Context
+
+Phase 7 (PAP-16897) built a package-local model of a native Paperclip run:
+a generated capability contract (7A), a UX interaction map (7B), a deterministic
+mock `ControlPlanePort` (7C), a transport-neutral semantic tool catalog and
+authorization engine (7D, security gate 7D/PAP-16902, remediation PAP-16909),
+an eval-derived 106-case conformance suite (7E), and a read-only browser
+scenario explorer (7F, UX gate PAP-16905). Phase 7G (PAP-16906) produces the
+reference documentation, the clean-start tutorial, the evidence manifest, and
+this OKF entry. It was blocked until 7E and 7F were done because the tutorial
+and evidence must describe and exercise the final conformance and browser
+surfaces.
+
+# Decisions
+
+1. Keep the reference set flat under `docs/phase-07-*.md`, matching the existing
+   per-phase reference convention, rather than introducing a new folder. Eight
+   reference pages: capability disposition, mock control plane, semantic tools,
+   authorization/exposure, eval conformance, scenario explorer, future binding
+   boundary, and verification commands.
+2. Treat the hand-written capability-disposition page as a reader's guide to the
+   generated `docs/phase-07-capability-contract.md`; never restate the generated
+   table, which carries a "DO NOT EDIT" header and is rewritten by
+   `generate:phase7-inventory`.
+3. Write the tutorial to run from the repository root with two commands to first
+   success, no Rust toolchain, no Paperclip service, and no external eval
+   repository. State time-to-first-success and exact expected output.
+4. Make the verification evidence file the single manifest: a command→result
+   matrix, artifact links, and the security/UX/QA gate issues. Capture real
+   observed output rather than asserting expected output.
+5. Do not commit the generated parity report. It is an on-demand artifact; a
+   clean checkout does not contain it, and the tracked
+   `evidence/phase-07/index.md` already documents it as generated.
+6. Update `docs/index.md`, `docs/architecture.md`, `README.md`, and the two
+   knowledge indexes additively. Keep ACPX as Phase 8 and state that real
+   Paperclip integration requires separate approval.
+
+# Observed facts
+
+All commands ran from the repository root, offline, on Node 22.22.2 / pnpm
+9.15.4, between 13:37Z and 13:39Z on 2026-08-09. Full matrix in
+[the verification evidence](../evidence/2026-08-09-phase-07-verification.md).
+
+- `check:phase7-inventory` and `test:phase7-inventory` pass; the contract holds
+  152 skill/reference headings, 106 eval cases, 258 normative rows, and 41
+  legacy MCP aliases folded one-to-one.
+- Mock control plane (9 tests), semantic tools and authorization/redaction
+  (9 tests), scenario runtime and explorer (45 tests), and the 106-case
+  conformance suite all pass in-process.
+- `report:phase7-evals` reports 106 cases across 16 groups, 18/18 fake-agent
+  operations, and the bounded Codex binding sample.
+- `test:browser:phase7` passes 19 Playwright checks (6.8s) for information
+  architecture, route determinism, evidence routes, the credential boundary,
+  accessibility, and responsive layout.
+- On a clean tree `docs:validate` passes 62 documentation files and an OKF
+  bundle of 26 concepts across 5 indexes.
+
+# Evidence
+
+- [Phase 7 verification manifest](../evidence/2026-08-09-phase-07-verification.md)
+- [Phase 7F explorer verification](../evidence/2026-08-09-phase-07f-explorer-verification.md)
+- [Clean-start tutorial](../../docs/tutorials/phase-07-scenario-explorer.md)
+- [Verification commands reference](../../docs/phase-07-verification-commands.md)
+
+# Failures
+
+- With the generated parity report present, `validate-okf` fails:
+  `eval-parity-report.md` has no OKF frontmatter and is not linked from
+  `evidence/phase-07/index.md`. Deleting the on-demand report restored a clean
+  pass. This is a real dirty-tree gap, not a documentation error; it is recorded
+  below and handed to the QA gate.
+
+# Known gaps
+
+- The generated parity report is incompatible with the OKF bundle scan when left
+  in the knowledge tree. Candidate fixes for Phase 7H (PAP-16907): relocate the
+  generated report outside `knowledge/`, add OKF frontmatter plus an index link
+  during generation, or exclude generated reports from the OKF scan.
+- Codex mode is disabled in the explorer; the only real provider exercise is the
+  bounded binding sample in `report:phase7-evals`.
+- Phase 7 adds no real Paperclip integration. Phase 8 (ACPX) binds the real
+  adapter behind the preserved seam and requires separate CTO approval
+  (PAP-16908).
+
+# Follow-up questions
+
+- Should the generated parity report move under `knowledge/evidence/phase-07/`
+  with generator-authored OKF frontmatter, or out of the bundle entirely?
+- Which of the eight reference pages should the Phase 8 binding work extend
+  first when the real adapter lands behind the seam?
+</content>
diff --git a/packages/paperclip-runner/knowledge/journal/index.md b/packages/paperclip-runner/knowledge/journal/index.md
index 7d4133ba5ce..bd0abc52457 100644
--- a/packages/paperclip-runner/knowledge/journal/index.md
+++ b/packages/paperclip-runner/knowledge/journal/index.md
@@ -16,3 +16,4 @@
 * [Phase 6 thin Paperclip adapter design](2026-08-09-phase-06-design.md) - Proposed public-port seam, feature flag, kill switch, company/auth/governance boundary, durable replay, conformance matrix, and tutorial contract.
 * [Phase 6 feature-flagged Paperclip adapter implementation](2026-08-09-phase-06-implementation.md) - Persisted selection, production port, public-session task, canonical event allocation, server-owned finalization, cancellation, and kill-switch evidence.
 * [Phase 6 thin Paperclip adapter implementation](2026-08-09-phase-06-implementation.md) - Default-off native selection, package-owned sessions, canonical persistence/replay, server-owned finalization, recovery, and compatibility evidence.
+* [Phase 7 reference docs, tutorial, evidence, and OKF](2026-08-09-phase-07.md) - Documentation decisions, observed verification facts across every focused command, the parity-report OKF gap, and the deferred Phase 8 binding boundary.