humanish 0.107.0 → 0.109.0

package/dist/routes/terminal/route.js

@@ -1,6 +1,6 @@
 // The terminal-product route: a real autonomous agent (Codex) studies a CLI or product from its
 // public surfaces only, inside an E2B shell. The flow:
-//   1. admit: planTerminalLab refuses what the plan cannot hold, then a live plan's machine checks
+//   1. admit: planTerminalStudy refuses what the plan cannot hold, then a live plan's machine checks
 //      (checkLiveTerminalMachine) run before any run scope opens, so the CLI can refuse before it
 //      loads a declared scorer;
 //   2. open the run scope;
@@ -8,7 +8,7 @@
 //      live-sandbox.ts and live-finish.ts); a dry plan publishes its contract bundle (dry-run.ts);
 //   4. automatic analysis of a run that published its final bundle.
 //
-// The safety contract (`docs/goals/terminal-product-lane/goal.md`) is enforced by construction in
+// The safety contract (`docs/history/goals/terminal-product-lane/goal.md`) is enforced by construction in
 // the files each item names, and checked by the verifier (verify/actor.ts
 // validateTerminalProductEvidence):
 //   1. Explicit key placement. openai-env (default) injects the raw runtime key command-scoped,
@@ -20,7 +20,7 @@
 //      (default/require 0 = no-spend) + maxMinutes (wall-clock kill of the codex command).
 //      Enforced in plan.ts (caps required), live-sandbox.ts and lifetime.ts (the wall clock).
 //   3. Public surfaces only. The mission references only subject.product.publicSurfaces + the
-//      author mission. No clone, no private-source access — nothing is git-cloned on this route.
+//      author mission. No clone and no private-source access: nothing is git-cloned on this route.
 //      Enforced in session.ts, which composes the prompt.
 //   4. Deny-by-default credentials. The command envs are built from an allowlist holding only the
 //      declared runtime key; GITHUB_TOKEN/GH_TOKEN/payment/deploy/db/media keys are excluded by
@@ -47,34 +47,34 @@ import { resolveAutomaticAnalysis } from "../../analysis/automatic-config.js";
 import { completeAutomaticAnalysis } from "../../analysis/automatic-completion.js";
 import path from "node:path";
 import { runScope } from "../../run/run.js";
-import { planTerminalLab } from "./plan.js";
-import { runDryTerminalLab } from "./dry-run.js";
+import { planTerminalStudy } from "./plan.js";
+import { runDryTerminalStudy } from "./dry-run.js";
 import { checkLiveTerminalMachine, runLiveTerminalSession } from "./session.js";
-import { terminalRouteScorer } from "../../lab/adapter-scorer-loader.js";
-import { withLateScorer } from "../../lab/route-inputs.js";
-import { TERMINAL_PRODUCT_LAB_SCHEMA, } from "./types.js";
+import { terminalRouteScorer } from "../../study/adapter-scorer-loader.js";
+import { withLateScorer } from "../../study/route-inputs.js";
+import {} from "./types.js";
+import { studyResultIdentity } from "../../run/study-result.js";
 /**
  * The config-taking entry point. It plans, returns a refusal with the same envelope and analysis
  * record the route has always returned, and otherwise runs the plan.
  */
-export async function runTerminalProductLab(options) {
+export async function runTerminalProductStudy(options) {
     const { config, dryRun, ...input } = options;
-    const planned = planTerminalLab(config, {
+    const planned = planTerminalStudy(config, {
         dryRun,
         hasCostProbe: input.deps?.costProbe !== undefined,
     });
     if (planned.ok)
         return runTerminalPlan(planned.plan, input);
-    return terminalLabRefusal(options, planned.refusal);
+    return terminalStudyRefusal(options, planned.refusal);
 }
-/** A refused terminal lab's result: the route's envelope, and the analysis record a refusal gets. */
-export function terminalLabRefusal(options, refusal) {
+/** A refused terminal study's result: the route's envelope, and the analysis record a refusal gets. */
+export function terminalStudyRefusal(options, refusal) {
     const { config, dryRun } = options;
     const refused = {
-        schema: TERMINAL_PRODUCT_LAB_SCHEMA,
+        ...studyResultIdentity("terminal", config.id),
         ok: false,
         cwd: path.resolve(options.cwd),
-        labId: config.id,
         actor: refusal.actor ?? config.actors[0]?.type ?? "",
         product: config.subject.product?.name ?? "",
         dryRun,
@@ -88,7 +88,7 @@ export function terminalLabRefusal(options, refusal) {
 }
 /**
  * Run a terminal plan. The run scope gives a direct library caller the same run lifetime the CLI
- * gets: whichever of its fail-closed exits the lab takes, the run it started is closed, and only a
+ * gets: whichever of its fail-closed exits the run takes, the run it started is closed, and only a
  * run that published its final bundle reaches automatic analysis.
  */
 export async function runTerminalPlan(plan, input) {
@@ -96,7 +96,7 @@ export async function runTerminalPlan(plan, input) {
     return admission.ok ? runAdmittedTerminalRun(admission.admitted, input) : admission.result;
 }
 /**
- * runLab's step for a terminal plan. It runs a live plan's local checks (checkLiveTerminalMachine)
+ * runStudyWith's step for a terminal plan. It runs a live plan's local checks (checkLiveTerminalMachine)
  * before any run scope opens, so the CLI can present their refusal before it loads a declared
  * scorer, and returns the run that continues from them with that scorer.
  */
@@ -141,10 +141,9 @@ function completeTerminalAnalysis(plan, input, result, finished) {
 function terminalFailure(plan, input, warnings) {
     const cwd = path.resolve(input.cwd);
     return (code, message) => ({
-        schema: TERMINAL_PRODUCT_LAB_SCHEMA,
+        ...studyResultIdentity("terminal", plan.studyId),
         ok: false,
         cwd,
-        labId: plan.labId,
         actor: plan.actor,
         product: plan.product.name,
         dryRun: plan.dryRun,
@@ -166,5 +165,5 @@ async function runTerminalPlanInScope(admitted, input, scope) {
         const { plan, runtimeEnv } = admitted;
         return runLiveTerminalSession({ plan, input, cwd, warnings, failed, scope, runtimeEnv });
     }
-    return runDryTerminalLab({ plan: admitted.plan, input, cwd, warnings, failed, scope });
+    return runDryTerminalStudy({ plan: admitted.plan, input, cwd, warnings, failed, scope });
 }

package/dist/routes/terminal/sandbox.d.ts

@@ -25,7 +25,7 @@ export declare function teardownSandbox(args: {
     warnings: string[];
 }): Promise<TerminalLedgers["cleanup"]>;
 /**
- * Race a commands.run promise against the maxMinutes wall-clock (safety contract item 2). The E2B
+ * Race a commands.run promise against the maxMinutes wall-clock. The E2B
  * commands.run timeoutMs is the primary kill; this injected-clock guard is the belt-and-suspenders
  * backstop so a mock CLI (which ignores timeoutMs) is still bounded and fails closed in CI.
  */

package/dist/routes/terminal/sandbox.js

@@ -98,7 +98,7 @@ export async function teardownSandbox(args) {
     }
 }
 /**
- * Race a commands.run promise against the maxMinutes wall-clock (safety contract item 2). The E2B
+ * Race a commands.run promise against the maxMinutes wall-clock. The E2B
  * commands.run timeoutMs is the primary kill; this injected-clock guard is the belt-and-suspenders
  * backstop so a mock CLI (which ignores timeoutMs) is still bounded and fails closed in CI.
  */

package/dist/routes/terminal/session.d.ts

@@ -1,4 +1,4 @@
-import type { LiveTerminalAuth, LiveTerminalPlan, RunLiveTerminalSessionArgs, TerminalProductLabErrorCode, TerminalProductLabResult, TerminalRunInput } from "./types.js";
+import type { LiveTerminalAuth, LiveTerminalPlan, RunLiveTerminalSessionArgs, TerminalProductStudyErrorCode, TerminalProductStudyResult, TerminalRunInput } from "./types.js";
 /**
  * The checks of this machine a live run makes before its run starts: the runtime key and
  * E2B_API_KEY. They read only the plan and the environment, so the CLI makes them before it loads
@@ -9,17 +9,17 @@ export declare function checkLiveTerminalMachine(plan: LiveTerminalPlan, input:
     readonly runtimeEnv: LiveTerminalAuth;
 } | {
     readonly ok: false;
-    readonly code: TerminalProductLabErrorCode;
+    readonly code: TerminalProductStudyErrorCode;
     readonly message: string;
 };
 /**
- * The live in-sandbox agent session orchestrator (mirror of runCuaActorLab's E2B branch). Enforces
+ * The live in-sandbox agent session orchestrator (mirror of runCuaActorStudy's E2B branch). Enforces
  * the 8-point safety contract by construction; fails closed before any sandbox/key/spend on any
  * precondition miss. Persists the substrate-lifecycle/command-log/interventions/cleanup ledgers,
  * the redacted terminal event stream + normalized transcript, the agent report, and the
  * provider-neutral actor trace; tears the sandbox down in a finally and proves the teardown.
  */
-export declare function runLiveTerminalSession(args: RunLiveTerminalSessionArgs): Promise<TerminalProductLabResult>;
-/** The default mission when the lab omits one. Public-safe, product-neutral author text. */
+export declare function runLiveTerminalSession(args: RunLiveTerminalSessionArgs): Promise<TerminalProductStudyResult>;
+/** The default mission when the study omits one. Public-safe, product-neutral author text. */
 export declare function defaultMission(productName: string): string;
 export declare function makeTerminalRunId(): string;

package/dist/routes/terminal/session.js

@@ -1,5 +1,5 @@
 import { randomBytes, randomUUID } from "node:crypto";
-import { missingKeys } from "../../lab/requirements.js";
+import { missingKeys } from "../../study/requirements.js";
 import { declaredRuntimeProvenance } from "./runtime.js";
 import { digestText, redactText, scrubLiterals } from "../../evidence/redaction.js";
 import { describeMissingKeys } from "../../keys/key-resolution.js";
@@ -18,11 +18,11 @@ import { finishLiveTerminalSession } from "./live-finish.js";
 export function checkLiveTerminalMachine(plan, input, warnings) {
     const env = input.env ?? process.env;
     const { maxUsd } = plan.caps;
-    // planTerminalLab refuses a positive maxUsd without a costProbe, so one is present here.
+    // planTerminalStudy refuses a positive maxUsd without a costProbe, so one is present here.
     if (maxUsd > 0) {
         warnings.push(`scenario.caps.maxUsd=${maxUsd} is checked after the session against the lines the costProbe measures; lines it leaves null (unmeasured) never trip it. scenario.caps.maxMinutes bounds the run while it runs.`);
     }
-    // --- Safety contract item 4: deny-by-default credentials; build the command-scoped allowlist. ---
+    // --- Deny-by-default credentials: build the command-scoped allowlist. ---
     const runtimeEnv = buildRuntimeAuth({ runtimeAuth: plan.runtime.auth, env });
     if (!runtimeEnv.ok)
         return runtimeEnv;
@@ -31,14 +31,14 @@ export function checkLiveTerminalMachine(plan, input, warnings) {
     if (missingKeys(plan.requirements, env).length > 0) {
         return {
             ok: false,
-            code: "HUMANISH_TERMINAL_LAB_KEYS_MISSING",
-            message: `Live terminal-product labs need E2B_API_KEY in the environment (values are never persisted). ${describeMissingKeys(["E2B_API_KEY"], env)}`,
+            code: "HUMANISH_TERMINAL_KEYS_MISSING",
+            message: `Live terminal-product studies need E2B_API_KEY in the environment (values are never persisted). ${describeMissingKeys(["E2B_API_KEY"], env)}`,
         };
     }
     return { ok: true, runtimeEnv };
 }
 /**
- * The live in-sandbox agent session orchestrator (mirror of runCuaActorLab's E2B branch). Enforces
+ * The live in-sandbox agent session orchestrator (mirror of runCuaActorStudy's E2B branch). Enforces
  * the 8-point safety contract by construction; fails closed before any sandbox/key/spend on any
  * precondition miss. Persists the substrate-lifecycle/command-log/interventions/cleanup ledgers,
  * the redacted terminal event stream + normalized transcript, the agent report, and the
@@ -61,7 +61,7 @@ export async function runLiveTerminalSession(args) {
         mintRunId: makeTerminalRunId,
         // This entry point is the live terminal route; its dry-run sibling is a separate function.
         mode: "live",
-        lab: plan.lab,
+        study: plan.study,
         renderReview: renderTerminalReviewMarkdown,
         observer: { open: input.open === true, render: deps.renderObserver },
         now,
@@ -86,7 +86,7 @@ export async function runLiveTerminalSession(args) {
         modelSource,
         ...(reasoningEffort === undefined ? {} : { reasoningEffort }),
     });
-    recorder.recordLifecycle("terminal-lab.run.created", `Created live terminal-product run ${runId} (actor ${actor}, product ${product.name}). Caps: maxUsd=${maxUsd}, maxMinutes=${maxMinutes}. Subject provenance UNPINNED (public surfaces only).`);
+    recorder.recordLifecycle("terminal-lab.run.created", `Created live terminal-product run ${runId} (actor ${actor}, product ${product.name}). Caps: maxUsd=${maxUsd}, maxMinutes=${maxMinutes}. Subject provenance: unpinned (public pages only).`);
     const session = new LiveTerminalSandbox({
         plan,
         cwd,
@@ -101,7 +101,7 @@ export async function runLiveTerminalSession(args) {
         maxMinutes,
         e2bApiKey,
         runPaths,
-        metadata: buildSandboxMetadata({ labId: plan.labId, simId: "sim-001", runId }),
+        metadata: buildSandboxMetadata({ studyId: plan.studyId, recordId: "sim-001", runId }),
         warnings,
         recorder,
     });
@@ -146,7 +146,7 @@ export async function runLiveTerminalSession(args) {
 async function prepareLivePrompt(args) {
     const { plan, cwd, runtimeEnv, env, warnings } = args;
     const { product } = plan;
-    // Compose the prompt from public surfaces and the author mission alone (safety contract item 3).
+    // Compose the prompt from public surfaces and the author mission alone.
     // Inject a per-run verdict nonce: the agent echoes HUMANISH_ACTOR_VERDICT=<status>
     // HUMANISH_ACTOR_NONCE=<nonce>; the scorer verifies the nonce so replayed text cannot forge it.
     const mission = plan.mission ?? defaultMission(product.name);
@@ -160,10 +160,10 @@ async function prepareLivePrompt(args) {
         verdictNonce,
     });
     const promptDigest = digestText(composedPrompt);
-    // --- Safety contract item 5: literal-scrub every known value, then pattern-redact, at the source. ---
+    // --- Literal-scrub every known value, then pattern-redact, at the source. ---
     // The runtime key value (+ any other provisioned value) is scrubbed by literal match before
     // anything persists (a key has no detectable "shape" if it is an arbitrary token); redactText is
-    // the second pass for secret-SHAPED content. Applied PRE-truncation so a cut can never split a
+    // the second pass for secret-shaped content. Applied pre-truncation so a cut can never split a
     // value past the scrubber.
     const knownSecretValues = [runtimeEnv.keyValue, env.E2B_API_KEY?.trim() ?? ""].filter((v) => v.length >= 4);
     const scrubKnownValues = scrubLiterals(knownSecretValues);
@@ -181,6 +181,7 @@ async function prepareLivePrompt(args) {
 }
 /** Compose the live prompt: public surfaces + author mission + the verdict-nonce marker contract. */
 function composeLivePrompt(args) {
+    // prose-check: model prompt (the terminal agent reads this, not a person)
     return [
         args.personaLine,
         `product: ${args.productName}`,
@@ -191,7 +192,7 @@ function composeLivePrompt(args) {
         `When finished, print exactly one final machine-readable line in this format: HUMANISH_ACTOR_VERDICT=<status> HUMANISH_ACTOR_NONCE=${args.verdictNonce} where <status> is passed, blocked, or failed.`,
     ].join("\n");
 }
-/** The default mission when the lab omits one. Public-safe, product-neutral author text. */
+/** The default mission when the study omits one. Public-safe, product-neutral author text. */
 export function defaultMission(productName) {
     return `You are an autonomous agent. Discover ${productName} from its public surfaces and determine whether it can help with a durable real task. Stay within the declared no-spend caps. Leave feedback if the workflow is confusing.`;
 }

package/dist/routes/terminal/token-usage.d.ts

@@ -3,7 +3,7 @@ import type { ActorTokenUsage } from "../../actors/contract.js";
  * Accumulate runtime-turn usage from a captured `codex exec --json` stream. A Codex turn may
  * include multiple provider requests; these records cannot establish per-request pricing tiers.
  *
- * Returns undefined when the stream carried no usage record at all, which is the honest "no
+ * Returns undefined when the stream carried no usage record at all, which is the "no
  * signal" case and must stay distinguishable from a measured zero. Per-turn records are preserved
  * in `turns` because long-context pricing can only be computed from per-request sizes
  * (src/run/pricing.ts), and totals cannot say which requests crossed a threshold.

package/dist/routes/terminal/token-usage.js

@@ -1,10 +1,9 @@
-// Provider token accounting for the terminal route (#531).
+// Provider token accounting for the terminal route.
 //
-// The gap this closes: a live terminal run reported `cost.lines.provider = null` with source
-// "unmeasured", and the no-spend proof then read `satisfied: true` for `maxUsd: 0` on a run that
-// had demonstrably consumed hundreds of thousands of provider tokens. The statement was honest
-// (it listed provider as unmeasured) but it conflated two very different states: having no signal
-// at all, and knowing exactly how many tokens were spent while lacking a rate to price them.
+// Without these counts, a provider line of `null` with source "unmeasured" would let the no-spend
+// proof read `satisfied: true` for `maxUsd: 0` on a run that consumed hundreds of thousands of
+// provider tokens, conflating two different states: having no signal at all, and knowing exactly
+// how many tokens were spent while lacking a rate to price them.
 //
 // The counts were already in the bundle. `codex exec --json` emits one usage record per turn:
 //   {"type":"turn.completed","usage":{"input_tokens":201536,"cached_input_tokens":170558,
@@ -12,7 +11,7 @@
 //
 // So the route can record a measured token fact even when it cannot record a priced one. Rates stay
 // out of this module deliberately: pricing lives in src/run/pricing.ts, and the terminal route has no
-// real model id to price against (it records `model: "codex"`), so the honest output is
+// real model id to price against (it records `model: "codex"`), so the output is
 // tokens-known-rate-unknown rather than a guessed dollar figure.
 // Bounded to a single line on purpose. `[^}]*` would run past a truncated record's missing brace
 // and swallow the next, valid record's body, silently dropping a real turn from the count.
@@ -24,7 +23,7 @@ function num(value) {
  * Accumulate runtime-turn usage from a captured `codex exec --json` stream. A Codex turn may
  * include multiple provider requests; these records cannot establish per-request pricing tiers.
  *
- * Returns undefined when the stream carried no usage record at all, which is the honest "no
+ * Returns undefined when the stream carried no usage record at all, which is the "no
  * signal" case and must stay distinguishable from a measured zero. Per-turn records are preserved
  * in `turns` because long-context pricing can only be computed from per-request sizes
  * (src/run/pricing.ts), and totals cannot say which requests crossed a threshold.

package/dist/routes/terminal/trace.d.ts

@@ -1,5 +1,5 @@
 import type { ActorCompletionReason, ActorPersonaRef, ActorRuntimeProvenance, ActorStatus, ActorTokenUsage, ActorTrace } from "../../actors/contract.js";
-import type { LabRuntimeAuth } from "../../lab/types.js";
+import type { StudyRuntimeAuth } from "../../study/types.js";
 import type { TerminalParticipantText } from "./participant-text.js";
 import { type CommandLogRecord, type TerminalEventRecord } from "./types.js";
 /**
@@ -12,7 +12,7 @@ export declare function scrubSplitKnownValues(events: TerminalEventRecord[], kno
 export declare function tailOf(text: string): string;
 /**
  * Project the live terminal session into the provider-neutral humanish.actor-trace.v1 (`lane`
- * "terminal", protocol "terminal-exec"). counts.actions/messages drive the no-engagement honesty
+ * "terminal", protocol "terminal-exec"). counts.actions/messages drive the no-engagement
  * guard (a real run bumps them; a no-op is caught). No screenshots on this route.
  */
 export declare function buildTerminalActorTrace(args: {
@@ -27,9 +27,9 @@ export declare function buildTerminalActorTrace(args: {
     terminalEvents: TerminalEventRecord[];
     commandLog: CommandLogRecord[];
     transcriptTail: string;
-    runtimeAuth: LabRuntimeAuth;
+    runtimeAuth: StudyRuntimeAuth;
     runtime: ActorRuntimeProvenance;
-    /** Runtime-turn aggregate usage parsed from the exec stream (#531). Absent when the stream
+    /** Runtime-turn aggregate usage parsed from the exec stream. Absent when the stream
      *  carried no usage record, which stays distinct from a measured zero. */
     tokenUsage?: ActorTokenUsage;
     /** The agent's own text and activity, read from the raw stdout as it arrived. */

package/dist/routes/terminal/trace.js

@@ -57,7 +57,7 @@ export function tailOf(text) {
 }
 /**
  * Project the live terminal session into the provider-neutral humanish.actor-trace.v1 (`lane`
- * "terminal", protocol "terminal-exec"). counts.actions/messages drive the no-engagement honesty
+ * "terminal", protocol "terminal-exec"). counts.actions/messages drive the no-engagement
  * guard (a real run bumps them; a no-op is caught). No screenshots on this route.
  */
 export function buildTerminalActorTrace(args) {
@@ -88,7 +88,7 @@ export function buildTerminalActorTrace(args) {
         redaction: {
             status: "passed",
             screenshots: "n/a",
-            notes: "Terminal exec output captured via commands.run onStdout/onStderr, scrubbed (literal known values) then redacted (shape patterns) AT THE SOURCE before persisting; no screenshots on this lane.",
+            notes: "Terminal output captured through commands.run onStdout and onStderr; known values are scrubbed and shape patterns redacted before anything is stored. This route takes no screenshots.",
         },
         startedAt: args.createdAt,
         completedAt: args.completedAt,

package/dist/routes/terminal/types.d.ts

@@ -1,16 +1,17 @@
-import type { LabEvent } from "../../lab/run-lab-events.js";
+import type { StudyEvent } from "../../study/run-study-events.js";
 import { type AutomaticAnalysisResult } from "../../analysis/automatic-completion.js";
 import type { ActorCompletionReason, ActorRuntimeProvenance, ActorStatus, ActorTrace } from "../../actors/contract.js";
 import type { CostCategory } from "../../run/terminal-contract.js";
 import type { RunScope } from "../../run/run.js";
 import type { buildRuntimeAuth } from "./credentials.js";
-import type { TerminalPlan } from "../../lab/plan-types.js";
-import type { LabDeps } from "../../lab/lab-deps.js";
-import type { LabConfig } from "../../lab/types.js";
+import type { TerminalPlan } from "../../study/plan-types.js";
+import type { StudyDeps } from "../../study/study-deps.js";
+import type { StudyConfig } from "../../study/types.js";
 import type { ObserverResult } from "../../observer/render.js";
 import { type RunAdapterScore, type RunBundle, type RunFeedbackCandidate, type RunScorerProvenance } from "../../run/bundle.js";
-/** Provider-neutral metadata constant: the route's non-secret tag (mirrors CUA_ACTOR_LAB_PROVIDER_METADATA). */
-export declare const TERMINAL_PRODUCT_LAB_PROVIDER_METADATA: {
+import { type StudyResultIdentity } from "../../run/study-result.js";
+/** Provider-neutral metadata constant: the route's non-secret tag (mirrors CUA_ACTOR_STUDY_PROVIDER_METADATA). */
+export declare const TERMINAL_PRODUCT_STUDY_PROVIDER_METADATA: {
     readonly mode: "terminal-product-lab";
     readonly tool: "humanish";
 };
@@ -22,7 +23,7 @@ export declare const UPLOAD_MAX_BYTES: number;
 /**
  * Server-side reclamation buffer past the codex command's own wall-clock (caps.maxMinutes) kill.
  * Shorter than the desktop routes' SANDBOX_TIMEOUT_BUFFER_MS (src/substrates/e2b/lifetime.ts):
- * when #158 set it, the terminal sandbox ran nothing but the codex command, and after its kill only
+ * when this buffer was set, the terminal sandbox ran nothing but the codex command, and after its kill only
  * teardown remained.
  */
 export declare const TERMINAL_SANDBOX_TIMEOUT_BUFFER_MS: number;
@@ -34,13 +35,12 @@ export declare const TEXT_ITEMS_MAX = 200;
 export declare const TEXT_ITEMS_BYTES: number;
 export declare const PENDING_LINE_CHARS: number;
 export declare const MAX_TRANSCRIPT_BYTES: number;
-export declare const TERMINAL_PRODUCT_LAB_SCHEMA = "humanish.terminal-lab-result.v1";
 /**
- * The read-only evidence a thin adapter's scorer/feedback hook sees (the layer-6 extension seam,
- * issue #154 acceptance #8). It is the FULLY-ASSEMBLED, redacted, verifiable evidence — the live run
+ * The read-only evidence a thin adapter's scorer/feedback hook sees (the adapter extension seam).
+ * It is the fully assembled, redacted, verifiable evidence: the live run
  * bundle, the provider-neutral actor trace, and the persisted ledgers (substrate/command/
  * interventions/cleanup/cost/no-spend). Every member is an exported public type, so a thin adapter
- * types against `import("humanish")` alone — never a deep `src/` import. The adapter reads this
+ * types against `import("humanish")` alone and never needs a deep `src/` import. The adapter reads this
  * to score the product attempt and derive feedback; it cannot mutate core's evidence (the route
  * attaches only the namespaced `RunAdapterScore` it returns + the feedback candidates it derives).
  */
@@ -56,30 +56,30 @@ export interface TerminalProductScoringContext {
      * values) then redacted (shape patterns) at the source, capped at MAX_TRANSCRIPT_BYTES, and
      * byte-identical to the persisted terminal-transcript.txt artifact. The trace's transcriptTail
      * is a ~2KB projection of this; a scorer needs the whole session so a rubric can find
-     * command-tier evidence anywhere in it, not only in the tail window (#341).
+     * command-tier evidence anywhere in it.
      */
     transcript: string;
     /** The studied product name (public-safe). */
     product: string;
-    /** The lab id (the run's scenario scope). */
-    labId: string;
+    /** The study id (the run's scenario scope). */
+    studyId: string;
     /** The run id (for building namespaced idempotency keys + evidence pointers). */
     runId: string;
 }
 /**
  * Known spend lines for the terminal cost ledger. Core has no product, media or payment spend
  * signal, so it fills only the provider line from trace tokenUsage when present. A test injects
- * known lines through `LabDeps.costProbe`; absent signals keep the null-discipline default.
+ * known lines through `StudyDeps.costProbe`; absent signals keep the null-discipline default.
  */
 export type TerminalCostProbe = (context: {
     tokenCostUsd?: number;
 }) => Partial<Record<"product" | "media" | "payment" | "provider", CostLine>> | undefined;
-/** The scorer functions a terminal run calls: `RunLabOptions.scorer` without `deriveArtifacts`. */
+/** The scorer functions a terminal run calls: `RunStudyOptions.scorer` without `deriveArtifacts`. */
 export interface TerminalScorer {
     /**
-     * The layer-6 extension seam (issue #154 acceptance #8: "product-adapter hooks without forking
-     * core"). A thin in-repo/out-of-tree adapter registers a product scorer here. The route calls it
-     * (when provided) over the fully-assembled evidence and attaches the returned, ADAPTER-NAMESPACED
+     * The adapter extension seam: product-adapter hooks without forking
+     * core. A thin in-repo/out-of-tree adapter registers a product scorer here. The route calls it
+     * (when provided) over the fully-assembled evidence and attaches the returned, adapter-namespaced
      * `RunAdapterScore` to `bundle.adapterScore` without core knowing any product noun (the score is
      * namespaced + its component breakdown rides in `data`). When no scorer is given, the default
      * mission-based verdict (`review`) is unchanged. The adopter's scorecard plugs in here; product
@@ -90,20 +90,20 @@ export interface TerminalScorer {
      * Companion seam: derive product-feedback candidates from the same assembled evidence. The route
      * appends the returned candidates to `bundle.feedbackCandidates`. The adapter records its
      * product-specific concepts (public CLI command observed, hosted success-or-blocker, feedback id,
-     * media/job ids, no-spend proof, defection/friction risk) under each candidate's ADAPTER-NAMESPACED
-     * `adapter` block — never as core enums (issue #154's "record product-specific concepts as
-     * NON-core nouns" list). The candidates must still satisfy core's feedback-candidate shape (which
+     * media/job ids, no-spend proof, defection/friction risk) under each candidate's adapter-namespaced
+     * `adapter` block and never as core enums.
+     * The candidates must still satisfy core's feedback-candidate shape (which
      * the bundle verifier enforces), so a malformed adapter candidate fails closed.
      */
     deriveFeedback?: (ctx: TerminalProductScoringContext) => RunFeedbackCandidate[] | Promise<RunFeedbackCandidate[]>;
 }
-export interface RunTerminalProductLabOptions {
+export interface RunTerminalProductStudyOptions {
     /** Cancels post-run analysis only. */
     analysisSignal?: AbortSignal;
-    /** Reports the analysis window to onEvent; built by normalizeRunLabOptions. */
-    emit?: (event: LabEvent) => void;
+    /** Reports the analysis window to onEvent; built by normalizeRunStudyOptions. */
+    emit?: (event: StudyEvent) => void;
     cwd: string;
-    config: LabConfig;
+    config: StudyConfig;
     /** Resolved upstream (scenario.mode + CLI override); defaults safe (dry-run). */
     dryRun: boolean;
     open?: boolean;
@@ -117,9 +117,9 @@ export interface RunTerminalProductLabOptions {
     env?: Readonly<Record<string, string | undefined>>;
     scorer?: TerminalScorer;
     /** Test seams: the E2B module, the Observer renderer, the clock and the cost probe. */
-    deps?: LabDeps;
+    deps?: StudyDeps;
     /**
-     * Present only when the scorer was config-declared and loaded by the CLI (#316). Its presence
+     * Present only when the scorer was config-declared and loaded by the CLI. Its presence
      * is the "declared" marker: a config-declared terminal scorer returning status:"fail" flips
      * bundle.review.verdict (like the browser routes), and one that throws becomes a visible review.gaps
      * entry. A library caller passing `scorer` leaves this absent and keeps today's purely
@@ -127,14 +127,12 @@ export interface RunTerminalProductLabOptions {
      */
     scorerProvenance?: RunScorerProvenance;
 }
-export interface TerminalProductLabResult extends AutomaticAnalysisResult {
-    schema: typeof TERMINAL_PRODUCT_LAB_SCHEMA;
+export interface TerminalProductStudyResult extends AutomaticAnalysisResult, StudyResultIdentity<"terminal"> {
     /** True when the bundle verified and (dry-run, or the live session reached a terminal verdict
      *  without a harness error + cleanup was proven). The agent's pass/fail is evidence, not the
-     *  lab's exit code. */
+     *  run's exit code. */
     ok: boolean;
     cwd: string;
-    labId: string;
     /** The registry-resolved actor id that ran (or would run) the session. */
     actor: string;
     /** The studied product name (public-safe). */
@@ -151,11 +149,11 @@ export interface TerminalProductLabResult extends AutomaticAnalysisResult {
     sandbox?: {
         sandboxId: string;
         killed: boolean;
-        /** BY-ID proof (never a re-list): 0 = confirmed reclaimed, 1 = still present (unconfirmed),
+        /** By-id proof (never a re-list): 0 = confirmed reclaimed, 1 = still present (unconfirmed),
          *  -1 = kill(id) itself failed or was unavailable. See TerminalLedgers["cleanup"]. */
         remaining: number;
     };
-    /** Live-only: the spend ledger surfaced on the result — unknowns are null, never guessed.
+    /** Live-only: the spend ledger surfaced on the result. Unknowns are null, never guessed.
      *  Lets a programmatic caller read spend without parsing the bundle. */
     cost?: {
         knownTotalUsd: number;
@@ -173,7 +171,7 @@ export interface TerminalProductLabResult extends AutomaticAnalysisResult {
     observer?: ObserverResult;
     warnings: string[];
     error?: {
-        code: "HUMANISH_LAB_ANALYSIS_INVALID" | "HUMANISH_LAB_TASKS_UNSUPPORTED" | "HUMANISH_LAB_OPTION_UNSUPPORTED" | "HUMANISH_TERMINAL_LAB_FAILED" | "HUMANISH_TERMINAL_LAB_ACTOR_UNSUPPORTED" | "HUMANISH_TERMINAL_LAB_SUBJECT_INVALID" | "HUMANISH_TERMINAL_LAB_KEYPLACEMENT_INVALID" | "HUMANISH_TERMINAL_LAB_RUNTIME_AUTH_MISSING" | "HUMANISH_TERMINAL_LAB_KEYS_MISSING" | "HUMANISH_TERMINAL_LAB_CAPS_MISSING" | "HUMANISH_TERMINAL_LAB_CAPS_INVALID" | "HUMANISH_TERMINAL_LAB_UNPRICED_CAP" | "HUMANISH_TERMINAL_LAB_CAPS_EXCEEDED" | "HUMANISH_TERMINAL_LAB_CREDENTIAL_DENIED" | "HUMANISH_TERMINAL_LAB_CLEANUP_UNPROVEN" | "HUMANISH_RUN_ID_IN_USE";
+        code: "HUMANISH_STUDY_ANALYSIS_INVALID" | "HUMANISH_STUDY_TASKS_UNSUPPORTED" | "HUMANISH_STUDY_OPTION_UNSUPPORTED" | "HUMANISH_STUDY_V2_UNSUPPORTED" | "HUMANISH_TERMINAL_FAILED" | "HUMANISH_TERMINAL_ACTOR_UNSUPPORTED" | "HUMANISH_TERMINAL_SUBJECT_INVALID" | "HUMANISH_TERMINAL_KEYPLACEMENT_INVALID" | "HUMANISH_TERMINAL_RUNTIME_AUTH_MISSING" | "HUMANISH_TERMINAL_KEYS_MISSING" | "HUMANISH_TERMINAL_CAPS_MISSING" | "HUMANISH_TERMINAL_CAPS_INVALID" | "HUMANISH_TERMINAL_UNPRICED_CAP" | "HUMANISH_TERMINAL_CAPS_EXCEEDED" | "HUMANISH_TERMINAL_CREDENTIAL_DENIED" | "HUMANISH_TERMINAL_CLEANUP_UNPROVEN" | "HUMANISH_RUN_ID_IN_USE";
         message: string;
     };
 }
@@ -189,7 +187,7 @@ export interface CommandLogRecord {
     at: string;
     /** A public-safe label for the command (e.g. "codex-exec"); the full argv is bound by digest only. */
     label: string;
-    /** sha256-12 of the exact command string — pins "same recipe" without persisting it. */
+    /** sha256-12 of the exact command string. It pins "same recipe" without persisting it. */
     commandDigest: string;
     /** The env var names injected command-scoped (values never persisted): the credential evidence. */
     envNames: string[];
@@ -216,7 +214,7 @@ export interface InterventionRecord {
     inputDigest: string;
 }
 /**
- * One cost line of the spend ledger. The null discipline (issue #154, the cost/no-spend asks):
+ * One cost line of the spend ledger. The null discipline:
  * three distinct states are kept apart:
  *   - `usd: 0`     => known to be zero. A measured-and-zero spend (we metered this category and it
  *                     billed nothing). The no-spend proof may legitimately assert this is zero.
@@ -226,7 +224,7 @@ export interface InterventionRecord {
  *   - line absent  => not applicable to this route/run (n/a). The line simply does not appear in
  *                     `lines`. (The current route emits all four lines, so absence is reserved for
  *                     future routes that genuinely have no such category.)
- * `null` vs missing-key is the load-bearing distinction: a missing key means "this category does not
+ * `null` vs missing-key is the distinction that matters: a missing key means "this category does not
  * exist for this run"; a present key with `null` means "this category exists but we did not measure
  * it". A no-spend proof that claimed zero on a `null` line would claim more than it measured.
  */
@@ -238,7 +236,7 @@ export interface CostLine {
     /** How this line's value was established (provenance for the verifier + the human reviewer). */
     source: "provider-token-usage" | "no-spend-signal" | "operator-cap" | "unmeasured"
     /** Tokens were counted but no rate could price them, so `usd` stays null while the note
-     *  carries the measured token totals (#531). Distinct from "unmeasured", which means no
+     *  carries the measured token totals. Distinct from "unmeasured", which means no
      *  signal at all. */
      | "unpriced-token-usage";
     /** A short, public-safe note (never a secret value). */
@@ -260,8 +258,8 @@ export interface TerminalCostLedger {
     fullyMeasured: boolean;
 }
 /**
- * The no-spend proof, derived from the cost ledger (issue #154: "derived from a ledger, not
- * asserted"). It lists the known-zero lines it can vouch for and, separately, the unmeasured
+ * The no-spend proof, derived from the cost ledger.
+ * It lists the known-zero lines it can vouch for and, separately, the unmeasured
  * (null) lines it cannot vouch for. `satisfied` is true only when every known line is zero (a
  * known non-zero line fails it). A proof with unmeasured lines says it could not measure them;
  * it never claims zero on a line the ledger marks null.
@@ -274,14 +272,14 @@ export interface NoSpendProof {
     satisfied: boolean;
     /** Categories the ledger measured and found at (known) zero: the proof can vouch for these. */
     knownZeroLines: CostCategory[];
-    /** Categories the ledger measured with a known NON-zero spend (these break `satisfied`). */
+    /** Categories the ledger measured with a known non-zero spend (these break `satisfied`). */
     knownNonZeroLines: CostCategory[];
     /** Categories the ledger marks `null` (not measured). The proof lists these and does not claim
      *  they are zero; it claims only that this run could not measure them. */
     unmeasuredLines: CostCategory[];
     /** Sum of the known lines (== 0 for a satisfied no-spend run). */
     knownTotalUsd: number;
-    /** Human-readable honesty statement covering both what is proven and what is unmeasured. */
+    /** Human-readable statement covering both what is proven and what is unmeasured. */
     statement: string;
 }
 /** The persisted terminal-product ledgers artifact (substrate lifecycle + command log + interventions + cleanup + cost). */
@@ -302,7 +300,7 @@ export interface TerminalLedgers {
          *  this exact sandbox running/paused (still present); -1 = kill(id) itself failed, threw, or
          *  was unavailable (the server-side kill-on-timeout is the backstop). */
         remaining: number;
-        /** Honest, human-readable statement of which by-id signal produced `remaining`. */
+        /** Human-readable statement of which by-id signal produced `remaining`. */
         reason: string;
     };
     /** The spend ledger. Unknowns are `null`, never guessed; the no-spend proof
@@ -311,13 +309,13 @@ export interface TerminalLedgers {
     /** The no-spend proof derived from `cost`. Never an independent assertion. */
     noSpendProof: NoSpendProof;
 }
-/** What a terminal run takes besides its plan. The plan carries the config, dry run and lab. */
-export type TerminalRunInput = Omit<RunTerminalProductLabOptions, "config" | "dryRun" | "lab">;
-/** A live terminal plan: its caps are the fail-closed ones planTerminalLab required. */
+/** What a terminal run takes besides its plan. The plan carries the config, dry run and study. */
+export type TerminalRunInput = Omit<RunTerminalProductStudyOptions, "config" | "dryRun">;
+/** A live terminal plan: its caps are the fail-closed ones planTerminalStudy required. */
 export type LiveTerminalPlan = Extract<TerminalPlan, {
     readonly dryRun: false;
 }>;
-export type TerminalProductLabErrorCode = NonNullable<TerminalProductLabResult["error"]>["code"];
+export type TerminalProductStudyErrorCode = NonNullable<TerminalProductStudyResult["error"]>["code"];
 /** The runtime key's command-scoped placement, built by a live run's checks of this machine. */
 export type LiveTerminalAuth = Extract<ReturnType<typeof buildRuntimeAuth>, {
     ok: true;
@@ -327,8 +325,8 @@ export interface RunLiveTerminalSessionArgs {
     input: TerminalRunInput;
     cwd: string;
     warnings: string[];
-    failed: (code: TerminalProductLabErrorCode, message: string) => TerminalProductLabResult;
-    /** The lab's run scope; the live session starts its run in it. */
+    failed: (code: TerminalProductStudyErrorCode, message: string) => TerminalProductStudyResult;
+    /** The study's run scope; the live session starts its run in it. */
     scope: RunScope;
     runtimeEnv: LiveTerminalAuth;
 }