humanish 0.107.0 → 0.109.0

package/dist/substrates/detached.d.ts

@@ -10,7 +10,7 @@ export declare function detachedTimersOf(options: DetachedTimers): DetachedTimer
 export interface DetachedStepOptions extends DetachedTimers {
     /** Short [a-z0-9-] label; names the script/status/log files under /tmp. */
     name: string;
-    /** The shell command to run (the lab author's own command — package.json-script trust). */
+    /** The shell command to run (the study author's own command: package.json-script trust). */
     command: string;
     cwd?: string;
     /** Wall-clock budget for the step. */
@@ -22,19 +22,19 @@ export interface DetachedStepResult {
     ok: boolean;
     exitCode?: number;
     timedOut: boolean;
-    /** Capped, UNREDACTED log tail — redact before persisting. */
+    /** Capped, unredacted log tail; redact before persisting. */
     logTail: string;
 }
-/** Read the capped log tail for a step (raw — caller redacts). */
+/** Read the capped log tail for a step (raw; caller redacts). */
 export declare function readDetachedLog(shell: Shell, name: string, requestTimeoutMs?: number): Promise<string>;
 /**
- * Run a BOUNDED step (install/build) detached, polling its atomic status file until it
+ * Run a bounded step (install/build) detached, polling its atomic status file until it
  * exits or the budget runs out. On timeout the process group is killed and the log tail is
  * still captured so failures stay diagnosable.
  */
 export declare function runDetachedStep(shell: Shell, options: DetachedStepOptions): Promise<DetachedStepResult>;
 /**
- * Launch a LONG-LIVED process (the subject's server) fully detached and return immediately.
+ * Launch a long-lived process (the subject's server) fully detached and return immediately.
  * No status polling: liveness is the caller's readiness probe, and reclamation belongs to
  * the sandbox lifecycle (create with kill-on-timeout).
  */
@@ -45,7 +45,7 @@ export declare function startDetachedProcess(shell: Shell, options: {
     requestTimeoutMs?: number;
 }): Promise<void>;
 /**
- * Poll a URL from INSIDE the sandbox until it answers 2xx/3xx or the budget runs out.
+ * Poll a URL from inside the sandbox until it answers 2xx/3xx or the budget runs out.
  * Returns true when the subject is ready.
  */
 export declare function probeUrl(shell: Shell, url: string, options: {

package/dist/substrates/detached.js

@@ -1,17 +1,17 @@
 // Detached process management over a Shell, the primitive behind serving a subject app in the
 // sandbox. A foreground command hits its deadline on long-running work, so long work runs detached:
 //
-// - Scripts are written via `writeFile`, never heredocs — which eliminates the
+// - Scripts are written via `writeFile`, never heredocs, which eliminates the
 //   sentinel-collision bug class (a command line that equals the heredoc terminator) by
 //   construction.
-// - Bounded steps (install/build) run detached with an ATOMICALLY-written status file
+// - Bounded steps (install/build) run detached with an atomically-written status file
 //   (write tmp + mv), polled by short foreground commands; a timeout kills the process
 //   group and surfaces a capped log tail for the caller to redact and persist.
 // - Long-lived steps (a dev/prod server) launch fully detached through `Shell.start`; the
 //   sandbox lifecycle (kill-on-timeout) owns their reclamation.
 // - Readiness is an explicit curl probe against the declared URL.
 //
-// Log tails are returned RAW; callers must pass them through redaction before persisting
+// Log tails are returned raw; callers must pass them through redaction before persisting
 // (build output can echo env values and paths).
 import { runOrThrow, shellQuote, throwOnExit } from "./shell.js";
 const WORK_ROOT = "/tmp/humanish-subject";
@@ -62,14 +62,14 @@ async function writeAndLaunch(shell, name, command, cwd, requestTimeoutMs) {
     await shell.writeFile(scriptPath, wrapperScript(name, command, cwd));
     throwOnExit(await shell.start(`bash ${shellQuote(scriptPath)}`, { requestTimeoutMs }));
 }
-/** Read the capped log tail for a step (raw — caller redacts). */
+/** Read the capped log tail for a step (raw; caller redacts). */
 export async function readDetachedLog(shell, name, requestTimeoutMs = DEFAULT_REQUEST_TIMEOUT_MS) {
     assertName(name);
     const result = await runOrThrow(shell, `tail -c ${LOG_TAIL_BYTES} ${shellQuote(`${stepDir(name)}/log.txt`)} 2>/dev/null || true`, { requestTimeoutMs });
     return result.stdout;
 }
 /**
- * Run a BOUNDED step (install/build) detached, polling its atomic status file until it
+ * Run a bounded step (install/build) detached, polling its atomic status file until it
  * exits or the budget runs out. On timeout the process group is killed and the log tail is
  * still captured so failures stay diagnosable.
  */
@@ -105,7 +105,7 @@ export async function runDetachedStep(shell, options) {
     }
 }
 /**
- * Launch a LONG-LIVED process (the subject's server) fully detached and return immediately.
+ * Launch a long-lived process (the subject's server) fully detached and return immediately.
  * No status polling: liveness is the caller's readiness probe, and reclamation belongs to
  * the sandbox lifecycle (create with kill-on-timeout).
  */
@@ -113,7 +113,7 @@ export async function startDetachedProcess(shell, options) {
     await writeAndLaunch(shell, options.name, options.command, options.cwd, options.requestTimeoutMs ?? DEFAULT_REQUEST_TIMEOUT_MS);
 }
 /**
- * Poll a URL from INSIDE the sandbox until it answers 2xx/3xx or the budget runs out.
+ * Poll a URL from inside the sandbox until it answers 2xx/3xx or the budget runs out.
  * Returns true when the subject is ready.
  */
 export async function probeUrl(shell, url, options) {

package/dist/substrates/e2b/cdp-probe.d.ts

@@ -11,29 +11,29 @@ export interface ChromeCdpProbeArgs {
     targetId?: string;
     /**
      * "pinned" (default): the launch-time target, for launch geometry and emulation attribution.
-     * "active": the tab the participant is driving NOW — Chrome's /json lists page
+     * "active": the tab the participant is driving now; Chrome's /json lists page
      * targets most-recently-focused first. The state observer must follow the participant: a
-     * verification link that opens in a NEW tab left a pinned observer reading the old tab forever,
+     * verification link that opens in a new tab left a pinned observer reading the old tab forever,
      * so the observed URL never changed again and stopWhen/task criteria went blind (a live run's
-     * funnel read reach-dashboard 0/2 under a screenshot OF the dashboard). Final geometry follows
+     * funnel read reach-dashboard 0/2 under a screenshot of the dashboard). Final geometry follows
      * this target too: a closed launch tab is unavailable, and a background tab may report stale
      * layout dimensions or zero outer bounds after navigation.
      */
     prefer?: ChromeCdpPagePreference;
     /**
      * "state": url/title/text/scrollY. "geometry": outer window + CSS viewport. "port": resolution
-     * only. "emulate": apply mobile emulation (#221) to the selected page and exit (the overrides that
+     * only. "emulate": apply mobile emulation to the selected page and exit (the overrides that
      * are session-scoped, UA / touch / DPR, lapse when the socket closes). "hold": the same over a browser-level
      * socket, then stay attached until killed (how a participant keeps them for its whole life) and attach
      * to every page target Chrome opens later, sending it the same overrides and a reload the moment
-     * it exists, never pausing it (#623). "fidelity": read
+     * it exists, never pausing it. "fidelity": read
      * back what the page reports about itself (UA, DPR, viewport, touch), the proof for the bundle.
      */
     mode: "state" | "geometry" | "port" | "emulate" | "hold" | "fidelity";
     /** For "emulate": what to apply. */
     emulation?: ChromeMobileEmulationRequest;
 }
-/** Mobile emulation request (#221): the CDP Emulation domain applied to one page target. */
+/** Mobile emulation request: the CDP Emulation domain applied to one page target. */
 export interface ChromeMobileEmulationRequest {
     width: number;
     height: number;
@@ -82,9 +82,9 @@ export interface ChromeCdpProbeResult {
  * headless Chrome).
  *
  * WebSocket is hand-rolled because python's stdlib has no client: one masked text frame out, frames
- * in until the reply with id 1 arrives, 1.5 s budget, and NO Origin header (Chrome refuses
+ * in until the reply with id 1 arrives, 1.5 s budget, and no Origin header (Chrome refuses
  * cross-origin DevTools sockets unless --remote-allow-origins is set; a header-less client is a
- * local one). urllib is opened WITHOUT proxy handlers so a sandbox-wide http_proxy cannot redirect
+ * local one). urllib is opened without proxy handlers so a sandbox-wide http_proxy cannot redirect
  * a loopback read.
  */
 export declare const CHROME_CDP_PROBE_PY: string;
@@ -93,7 +93,7 @@ export declare function chromeCdpProbeCommand(args: ChromeCdpProbeArgs): string;
 /**
  * Narrow one probe's stdout. A parse failure is reported as unavailable with the reason, never as
  * an empty success: the difference between "nothing to observe" and "could not observe" is the
- * whole point of #514.
+ * whole point.
  */
 export declare function parseChromeCdpProbeOutput(stdout: string | undefined): ChromeCdpProbeResult;
 export {};

package/dist/substrates/e2b/cdp-probe.js

@@ -1,22 +1,19 @@
 // The in-sandbox Chrome DevTools probe behind every URL / page-text / viewport observation.
 //
-// It runs on python3, stdlib only. It used to run on node, and that was the #514 root cause: the
-// stock E2B desktop template ships python3 and curl but NO Node, and Node only arrives when a
-// subject's serve pipeline needs it (src/subject/runtime.ts). So on the app-url route, and on any
-// subject served by something other than Node (the taskly benchmark is `python3 -m http.server`),
-// `node -e` exited 127 on every turn, the probe degraded to `{}`, and every urlIncludes /
-// textIncludes stop condition and task criterion went blind for the whole session. The only trace
-// was a geometry warning that the CSS viewport "could not be measured", which named the symptom
-// and not the cause. The tab-pinning fix that preceded this one (prefer "active") was diagnosed on
-// a Node subject, where the probe happened to work.
+// It runs on python3, stdlib only. The stock E2B desktop template ships python3 and curl but no
+// Node, and Node only arrives when a subject's serve pipeline needs it (src/subject/runtime.ts).
+// So on the app-url route, and on any subject served by something other than Node (the taskly
+// benchmark is `python3 -m http.server`), a `node -e` probe exits 127 on every turn, degrades to
+// `{}`, and blinds every urlIncludes / textIncludes stop condition and task criterion for the
+// whole session, with only a geometry warning that the CSS viewport "could not be measured".
 //
 // The same lesson was learned once already: the comms catch was rewritten from node to python3 in
 // 0.29.0 (comms/sandbox-catch.ts). This is the third in-sandbox runtime dependency to move.
 //
-// The script takes ONE JSON argument and prints ONE JSON line. Failures print
-// `{"unavailable": "<reason>"}` with exit 0 so the caller can say WHY the channel is dark instead
+// The script takes one JSON argument and prints one JSON line. Failures print
+// `{"unavailable": "<reason>"}` with exit 0 so the caller can say why the channel is dark instead
 // of swallowing an exit code; the TypeScript side turns that into a participant warning that names the
-// consequence ("url/text criteria will read as NEVER MEASURED").
+// consequence ("url/text criteria will read as never measured").
 import { shellQuote } from "../shell.js";
 /**
  * The probe itself. Kept as one string so the shipped command is exactly what the tests execute
@@ -24,9 +21,9 @@ import { shellQuote } from "../shell.js";
  * headless Chrome).
  *
  * WebSocket is hand-rolled because python's stdlib has no client: one masked text frame out, frames
- * in until the reply with id 1 arrives, 1.5 s budget, and NO Origin header (Chrome refuses
+ * in until the reply with id 1 arrives, 1.5 s budget, and no Origin header (Chrome refuses
  * cross-origin DevTools sockets unless --remote-allow-origins is set; a header-less client is a
- * local one). urllib is opened WITHOUT proxy handlers so a sandbox-wide http_proxy cannot redirect
+ * local one). urllib is opened without proxy handlers so a sandbox-wide http_proxy cannot redirect
  * a loopback read.
  */
 export const CHROME_CDP_PROBE_PY = String.raw `
@@ -277,7 +274,7 @@ def browser_ws_url(port):
 
 def hold(port, page, request):
     """Emulate the launch page and then every page target Chrome opens later, for as long as this
-    process lives (#221, #623). One browser-level socket with flattened sessions: the launch page is
+    process lives. One browser-level socket with flattened sessions: the launch page is
     attached by id and reloaded so scripts that read the UA at load see it; a target that appears
     later is attached the moment it exists, sent the same overrides, and reloaded once after its
     first real navigation commits (touch emulation reaches a document only when it loads under it).
@@ -387,7 +384,7 @@ def main():
     page = select_page(pages, args)
     # A listing with targets but no http page is usually an instant: a reload in flight, a tab
     # between about:blank and its navigation. Ask again for up to a second before calling it
-    # unavailable (#653): the lane's observer would otherwise lose one turn of url/text evidence,
+    # unavailable: the lane's observer would otherwise lose one turn of url/text evidence,
     # and a test on a loaded runner saw exactly this twice in one evening.
     for _ in range(4):
         if page is not None or not pages:
@@ -465,7 +462,7 @@ export function chromeCdpProbeCommand(args) {
 /**
  * Narrow one probe's stdout. A parse failure is reported as unavailable with the reason, never as
  * an empty success: the difference between "nothing to observe" and "could not observe" is the
- * whole point of #514.
+ * whole point.
  */
 export function parseChromeCdpProbeOutput(stdout) {
     const trimmed = (stdout ?? "").trim();

package/dist/substrates/e2b/desktop-browser.d.ts

@@ -1,4 +1,4 @@
-import type { LabDesktopBrowser } from "../../lab/types.js";
+import type { StudyDesktopBrowser } from "../../study/types.js";
 import type { E2BDesktopSandbox } from "./sdk.js";
 import type { DesktopMediaEvidence } from "./desktop-media.js";
 export declare const DESKTOP_SETTLE_MS = 8000;
@@ -27,9 +27,9 @@ export type ChromeDevToolsReadiness = {
 /** Reads the readiness markers from a launch command's stdout. */
 export declare function parseChromeDevToolsReadiness(stdout: string): ChromeDevToolsReadiness | undefined;
 export interface DesktopBrowserEvidence {
-    requested: LabDesktopBrowser;
+    requested: StudyDesktopBrowser;
     resolved?: string;
-    /** Synthetic media devices the browser was launched with (#509), and how permission is answered. */
+    /** Synthetic media devices the browser was launched with, and how permission is answered. */
     media?: DesktopMediaEvidence;
 }
 export type DesktopBrowserFamily = "chromium" | "firefox" | "unknown";
@@ -42,7 +42,7 @@ export interface DesktopBrowserLaunchIdentity {
 }
 /**
  * Runtime-only launch result. `evidence` is what the bundle records: the requested browser and the
- * one that launched, and only when the lab declared a browser.
+ * one that launched, and only when the study declared a browser.
  */
 export interface DesktopBrowserLaunchResult {
     family: DesktopBrowserFamily;
@@ -65,8 +65,8 @@ export declare function buildFillDesktopWindowCommand(windowId: string, width: n
  * below checks for clipping and refuses an uncorrectable window before the actor runs.
  */
 export declare function fillDesktopBrowserWindow(desktop: E2BDesktopSandbox, windowId: string, resolution: readonly [number, number], requestTimeoutMs: number): Promise<void>;
-export declare function openDesktopBrowserTarget(desktop: E2BDesktopSandbox, targetUrl: string, requestTimeoutMs: number, browserPreference: LabDesktopBrowser | undefined, 
-/** Launch-time flags that make mobile fidelity (#221) hold across every tab: the user agent and
+export declare function openDesktopBrowserTarget(desktop: E2BDesktopSandbox, targetUrl: string, requestTimeoutMs: number, browserPreference: StudyDesktopBrowser | undefined, 
+/** Launch-time flags that make mobile fidelity hold across every tab: the user agent and
  *  touch events are browser-wide here, where the CDP holder covers only the launch page. */
 extraChromiumFlags?: readonly string[], environment?: Readonly<Record<string, string>>): Promise<DesktopBrowserLaunchResult>;
 /**

package/dist/substrates/e2b/desktop-browser.js

@@ -2,7 +2,7 @@
 // window a desktop-CLI study starts in.
 import { CHROMIUM_EVIDENCE_HYGIENE_FLAGS, chromiumEvidenceProfilePreferencesJson, } from "../../evidence/browser-hygiene.js";
 import { failureTail } from "../../evidence/redaction.js";
-import { isHttpUrl } from "../../lab/parse/subject.js";
+import { isHttpUrl } from "../../study/parse/subject.js";
 import { runDetachedStep } from "../detached.js";
 import { shellQuote } from "../shell.js";
 import { e2bShell } from "./shell.js";
@@ -111,7 +111,7 @@ export function parseChromeDevToolsReadiness(stdout) {
 export async function findVisibleBrowserWindowId(desktop, requestTimeoutMs, browserFamily, launchIdentity) {
     if (browserFamily === "unknown")
         return undefined;
-    // The candidate loop keeps the LAST identity match: with a launch identity the match is
+    // The candidate loop keeps the last identity match: with a launch identity the match is
     // unique anyway, and without one every family candidate matches, so the newest visible
     // window of the launched family wins (the window this participant just opened).
     const finder = browserFamily === "firefox"
@@ -197,7 +197,7 @@ export async function fillDesktopBrowserWindow(desktop, windowId, resolution, re
         .catch(() => undefined);
 }
 export async function openDesktopBrowserTarget(desktop, targetUrl, requestTimeoutMs, browserPreference, 
-/** Launch-time flags that make mobile fidelity (#221) hold across every tab: the user agent and
+/** Launch-time flags that make mobile fidelity hold across every tab: the user agent and
  *  touch events are browser-wide here, where the CDP holder covers only the launch page. */
 extraChromiumFlags = [], environment) {
     const requestedBrowser = browserPreference ?? "default";
@@ -242,10 +242,10 @@ extraChromiumFlags = [], environment) {
             "  fi",
             "  return 1",
             "}",
-            // Fixed CDP port (not :0/random): each seat has its OWN desktop sandbox, so a known port
-            // cannot conflict, and it makes the observer's port resolution deterministic. With :0 the
-            // real port lives only in DevToolsActivePort; when the launch-time capture misses on a cold
-            // start the observer falls back to 9222 and — being wrong — every CDP read fails for the
+            // Fixed CDP port (not :0/random): each participant has its own desktop sandbox, so a known
+            // port cannot conflict, and it makes the observer's port resolution deterministic. With :0
+            // the real port lives only in DevToolsActivePort; when the launch-time capture misses on a
+            // cold start the observer falls back to 9222 and, being wrong, every CDP read fails for the
             // whole run (the lobby-code handoff then never sees the host's /lobby URL). 9222 is already
             // the fallback, so making it the actual port aligns launch, capture, and fallback.
             `chrome_debug_flags=(--remote-debugging-address=127.0.0.1 --remote-debugging-port=${CHROME_DEVTOOLS_PORT} ${chromiumFlags})`,
@@ -375,9 +375,9 @@ export async function openDesktopTerminal(desktop, requestTimeoutMs, workdir) {
         command: [
             "for candidate in x-terminal-emulator xfce4-terminal gnome-terminal konsole xterm; do",
             '  if command -v "$candidate" >/dev/null 2>&1; then',
-            // LANG is set on the terminal we open, not globally: the stock image declares no locale, and
+            // `LANG` is set on the terminal we open, not globally: the stock image declares no locale, and
             // a study that measures our own mojibake against an unconfigured template would be measuring
-            // the template. The PRODUCT-side fix (an ASCII fallback when the locale is not UTF-8) is in
+            // the template. The product-side fix (an ASCII fallback when the locale is not UTF-8) is in
             // src/routes/terminal/encoding.ts, and it is the one that matters for real users.
             `    (cd ${shellQuote(dir)} 2>/dev/null || cd /home/user; DISPLAY=:0 LANG=C.UTF-8 LC_ALL=C.UTF-8 HUMANISH_STUDY_PARTICIPANT=1 nohup "$candidate" >/dev/null 2>&1 &)`,
             "    sleep 3",

package/dist/substrates/e2b/desktop-cdp.d.ts

@@ -3,7 +3,7 @@ import type { RunDesktopGeometry } from "../../run/streams.js";
 import type { E2BDesktopSandbox } from "./sdk.js";
 /**
  * Runtime-only CDP endpoint attribution for the exact chromium this participant launched. Port
- * resolution at OBSERVE time: the cached launch-time `cdpPort` wins; absent that, the observer
+ * resolution at observe time: the cached launch-time `cdpPort` wins; absent that, the observer
  * probe re-reads `profileDir`'s DevToolsActivePort marker (a slow cold start can publish it
  * after the launch-time poll gave up); absent both it uses 9222, the port every participant launches
  * Chrome with, where a dead endpoint degrades into a warning that names the cause.
@@ -16,7 +16,7 @@ export interface ChromeCdpEndpoint {
     targetUrl: string;
 }
 /**
- * Mobile emulation on later tabs (#623): the holder attaches to every page target Chrome opens
+ * Mobile emulation on later tabs: the holder attaches to every page target Chrome opens
  * after the launch page, so a tab the participant opens later should lay out at the phone width
  * too. The first observation on each new target reads that page's own report; a target that
  * reports the requested width is recorded through `onCovered`, and one that does not (or cannot be
@@ -37,15 +37,15 @@ export interface ChromeEmulationDrift {
 /**
  * The URL / title / page-text / scroll observer behind stopWhen and task criteria. One probe per
  * observation, run on the sandbox's python3 (see src/substrates/e2b/cdp-probe.ts for why not
- * node: #514).
+ * node).
  *
- * "active": follow the participant to whatever tab they are driving now — never pin the state
- * observer to the launch tab (a verification link that opened in a NEW tab left a pinned observer
+ * "active": follow the participant to whatever tab they are driving now; never pin the state
+ * observer to the launch tab (a verification link that opened in a new tab left a pinned observer
  * reading the old tab forever).
  *
- * `onUnavailable` fires ONCE, on the first probe that could not read the page, with the reason.
+ * `onUnavailable` fires once, on the first probe that could not read the page, with the reason.
  * The observer still degrades to `{}` for the loop; the callback is how a participant says out loud that
- * url/text criteria are not being measured, instead of letting the funnel report 0/N (#514).
+ * url/text criteria are not being measured, instead of letting the funnel report 0/N.
  */
 export declare function makeChromeBrowserStateObserver(desktop: E2BDesktopSandbox, requestTimeoutMs: number, endpoint: ChromeCdpEndpoint, options?: {
     /** The page target to prefer before the participant switches tabs. */
@@ -59,11 +59,11 @@ export declare function makeChromeBrowserStateObserver(desktop: E2BDesktopSandbo
     text?: string;
     scrollY?: number;
 }>;
-/** The user agent a mobile-emulated participant presents unless the lab sets its own. */
+/** The user agent a mobile-emulated participant presents unless the study sets its own. */
 export declare const DEFAULT_MOBILE_USER_AGENT = "Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Mobile/15E148 Safari/604.1";
 /**
- * Apply mobile emulation (#221) to the participant's launch page and read back what the page reports.
- * Fails CLOSED: a request that cannot be applied throws, because a desktop run labelled mobile is
+ * Apply mobile emulation to the participant's launch page and read back what the page reports.
+ * Fails closed: a request that cannot be applied throws, because a desktop run labelled mobile is
  * the over-trust this feature exists to prevent. A read-back that cannot be taken is a warning
  * (the emulation was applied; only the proof is missing).
  */

package/dist/substrates/e2b/desktop-cdp.js

@@ -7,15 +7,15 @@ import { e2bShell } from "./shell.js";
 /**
  * The URL / title / page-text / scroll observer behind stopWhen and task criteria. One probe per
  * observation, run on the sandbox's python3 (see src/substrates/e2b/cdp-probe.ts for why not
- * node: #514).
+ * node).
  *
- * "active": follow the participant to whatever tab they are driving now — never pin the state
- * observer to the launch tab (a verification link that opened in a NEW tab left a pinned observer
+ * "active": follow the participant to whatever tab they are driving now; never pin the state
+ * observer to the launch tab (a verification link that opened in a new tab left a pinned observer
  * reading the old tab forever).
  *
- * `onUnavailable` fires ONCE, on the first probe that could not read the page, with the reason.
+ * `onUnavailable` fires once, on the first probe that could not read the page, with the reason.
  * The observer still degrades to `{}` for the loop; the callback is how a participant says out loud that
- * url/text criteria are not being measured, instead of letting the funnel report 0/N (#514).
+ * url/text criteria are not being measured, instead of letting the funnel report 0/N.
  */
 export function makeChromeBrowserStateObserver(desktop, requestTimeoutMs, endpoint, options = {}) {
     const { targetId, onUnavailable, drift } = options;
@@ -97,11 +97,11 @@ export function makeChromeBrowserStateObserver(desktop, requestTimeoutMs, endpoi
         };
     };
 }
-/** The user agent a mobile-emulated participant presents unless the lab sets its own. */
+/** The user agent a mobile-emulated participant presents unless the study sets its own. */
 export const DEFAULT_MOBILE_USER_AGENT = "Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Mobile/15E148 Safari/604.1";
 /**
- * Apply mobile emulation (#221) to the participant's launch page and read back what the page reports.
- * Fails CLOSED: a request that cannot be applied throws, because a desktop run labelled mobile is
+ * Apply mobile emulation to the participant's launch page and read back what the page reports.
+ * Fails closed: a request that cannot be applied throws, because a desktop run labelled mobile is
  * the over-trust this feature exists to prevent. A read-back that cannot be taken is a warning
  * (the emulation was applied; only the proof is missing).
  */
@@ -153,7 +153,7 @@ export async function applyMobileEmulation(desktop, requestTimeoutMs, endpoint,
     }
     const applied = announced;
     // Viewport/touch read-back proves context settings, not gesture equivalence. Two hosted
-    // replicas and a native-X conversion-toggle control reproduced reset click counts (#676).
+    // replicas and a native-X conversion-toggle control reproduced reset click counts.
     const warnings = request.touch
         ? [
             "Mobile emulation uses desktop pointer-to-touch conversion, which can differ for repeated taps. Confirm gesture failures with direct or native touch input before attributing them to the app.",

package/dist/substrates/e2b/desktop-executor.d.ts

@@ -17,7 +17,7 @@ export interface E2BDesktopLike {
             stdout?: string;
         }>;
     };
-    /** Optional file surface used to transfer typed text without shell-quoting it. */
+    /** Optional file surface for transferring typed text without shell-quoting it. */
     files?: {
         write(path: string, data: string | ArrayBuffer, options?: {
             requestTimeoutMs?: number;
@@ -77,7 +77,7 @@ export type CuaTypePhase = "desktop-write" | "text-tempfile" | "text-command";
  * password), so the message names the phase and, for xdotool, its exit code only: no text, no
  * temp-file path, no substrate output. The loop records `.name` + `.message` into the actor trace.
  */
-export declare class CuaTypeError extends Error {
+export declare class ComputerUseTypeError extends Error {
     readonly phase: CuaTypePhase;
     readonly exitCode?: number;
     constructor(phase: CuaTypePhase, exitCode?: number);

package/dist/substrates/e2b/desktop-executor.js

@@ -1,7 +1,7 @@
 import { shellQuote } from "../shell.js";
 import { perceptualSignature } from "../../evidence/frame-signature.js";
 import { commandFailureInfo } from "../command-failure.js";
-import { CuaExecutorError } from "../../actors/computer-use/executor-error.js";
+import { ComputerUseExecutorError } from "../../actors/computer-use/executor-error.js";
 import { xdotoolHeldModifiers } from "../../guest/desktop-keys.js";
 const DEFAULT_WAIT_MS = 500;
 const DEFAULT_SCROLL_AMOUNT_PER_TICK = 100;
@@ -9,7 +9,7 @@ const TYPE_COMMAND_TIMEOUT_MS = 15_000;
 const HELD_KEYS_TIMEOUT_MS = 15_000;
 const CURSOR_READ_TIMEOUT_MS = 500;
 /**
- * The stock image's xdotool waits ~15s on a synchronized move to its current position (#681).
+ * The stock image's xdotool waits ~15s on a synchronized move to its current position.
  * Only an exact integer match can omit that move; fractional actions keep the SDK's own
  * coordinate conversion. Read every time: another action or sequential role may move it.
  * The SDK's getCursorPosition (2.3 through 2.4.0) takes no timeout or signal. Bound our wait, observe late
@@ -72,12 +72,12 @@ function toBuffer(bytes) {
  * password), so the message names the phase and, for xdotool, its exit code only: no text, no
  * temp-file path, no substrate output. The loop records `.name` + `.message` into the actor trace.
  */
-export class CuaTypeError extends Error {
+export class ComputerUseTypeError extends Error {
     phase;
     exitCode;
     constructor(phase, exitCode) {
         super(`type failed at ${phase}${exitCode === undefined ? "" : ` (exit ${exitCode})`}`);
-        this.name = "CuaTypeError";
+        this.name = "ComputerUseTypeError";
         this.phase = phase;
         if (exitCode !== undefined)
             this.exitCode = exitCode;
@@ -107,17 +107,17 @@ async function typeText(desktop, text) {
         directory = made.stdout?.trim();
     }
     catch {
-        throw new CuaTypeError("text-tempfile");
+        throw new ComputerUseTypeError("text-tempfile");
     }
     if (directory === undefined || !TYPE_DIRECTORY.test(directory))
-        throw new CuaTypeError("text-tempfile");
+        throw new ComputerUseTypeError("text-tempfile");
     const file = `${directory}/text`;
     try {
         try {
             await files.write(file, text, { requestTimeoutMs: TYPE_COMMAND_TIMEOUT_MS });
         }
         catch {
-            throw new CuaTypeError("text-tempfile");
+            throw new ComputerUseTypeError("text-tempfile");
         }
         // xdotool waits the delay after every character, so a long text needs a longer budget.
         const timeoutMs = TYPE_COMMAND_TIMEOUT_MS + [...text].length * XDOTOOL_TYPE_DELAY_MS;
@@ -127,11 +127,11 @@ async function typeText(desktop, text) {
         }
         catch (error) {
             // The SDK throws on a non-zero exit; only its exit code is kept.
-            throw new CuaTypeError("text-command", commandFailureInfo(error).exitCode);
+            throw new ComputerUseTypeError("text-command", commandFailureInfo(error).exitCode);
         }
         // A structural fake may return a non-zero exit instead of throwing, as the SDK does.
         if (result?.exitCode !== undefined && result.exitCode !== 0)
-            throw new CuaTypeError("text-command", result.exitCode);
+            throw new ComputerUseTypeError("text-command", result.exitCode);
     }
     finally {
         await commands.run(`rm -rf -- ${shellQuote(directory)}`, quick).catch(() => undefined);
@@ -230,7 +230,7 @@ export function createE2BDesktopExecutor(desktop, options = {}) {
                     await desktop.write(action.text);
                 }
                 catch {
-                    throw new CuaTypeError("desktop-write");
+                    throw new ComputerUseTypeError("desktop-write");
                 }
                 return;
             }
@@ -261,7 +261,7 @@ export function createE2BDesktopExecutor(desktop, options = {}) {
                 // capturing here would double-capture. Leave the desktop untouched.
                 return;
             case "speak":
-                throw new CuaExecutorError("action_rejected", "not_dispatched");
+                throw new ComputerUseExecutorError("action_rejected", "not_dispatched");
         }
     }
 }
@@ -278,7 +278,7 @@ function sendsNoInput(action) {
 async function withHeldModifiers(desktop, chord, run) {
     const commands = desktop.commands;
     if (!commands)
-        throw new CuaExecutorError("action_rejected", "not_dispatched");
+        throw new ComputerUseExecutorError("action_rejected", "not_dispatched");
     const options = { requestTimeoutMs: HELD_KEYS_TIMEOUT_MS, timeoutMs: HELD_KEYS_TIMEOUT_MS };
     let failed = false;
     let failure;
@@ -295,7 +295,7 @@ async function withHeldModifiers(desktop, chord, run) {
         await commands.run(`xdotool keyup ${chord}`, options);
     }
     catch {
-        throw new CuaExecutorError("execution_failed", "outcome_uncertain");
+        throw new ComputerUseExecutorError("execution_failed", "outcome_uncertain");
     }
     if (failed)
         throw failure;

package/dist/substrates/e2b/desktop-geometry.d.ts

@@ -1,14 +1,14 @@
-import { type DevicePreset } from "../../lab/device-presets.js";
+import { type DevicePreset } from "../../study/device-presets.js";
 import { type RunDesktopGeometry } from "../../run/streams.js";
 import { type ChromeCdpPagePreference } from "./cdp-probe.js";
 import { type DesktopBrowserFamily, type DesktopBrowserLaunchIdentity } from "./desktop-browser.js";
 import type { ChromeCdpEndpoint } from "./desktop-cdp.js";
 import type { E2BDesktopSandbox } from "./sdk.js";
 /**
- * The DECLARED preset to record alongside the rendered screen, or undefined when the preset
+ * The declared preset to record alongside the rendered screen, or undefined when the preset
  * rendered faithfully.
  *
- * `desktopGeometry.screen.verified` compares the FLOORED number with itself, so on its own a
+ * `desktopGeometry.screen.verified` compares the floored number with itself, so on its own a
  * floored run is indistinguishable from a faithful one: a reader sees requested 500 / verified 500
  * and concludes a 500-wide screen was asked for. Recording the declared preset is what makes
  * "the preset width did not render" legible in the bundle.
@@ -19,7 +19,7 @@ export declare function declaredScreenForRender(preset: DevicePreset, presetName
     preset: string;
 } | undefined;
 /**
- * Verify the desktop screen geometry IN-SANDBOX (the participant's device claim is checked, never
+ * Verify the desktop screen geometry in-sandbox (the participant's device claim is checked, never
  * assumed). A parseable mismatch fails closed. Unavailable/unparseable evidence is returned as
  * an explicit warning: the participant may still run, but its bundle records only the requested screen
  * and never upgrades that request into a verified measurement.

package/dist/substrates/e2b/desktop-geometry.js

@@ -1,6 +1,6 @@
 // Screen and browser-window geometry on a hosted E2B desktop, measured in the sandbox rather than
 // taken from the request.
-import {} from "../../lab/device-presets.js";
+import {} from "../../study/device-presets.js";
 import { failureTail, redactText, toErrorMessage } from "../../evidence/redaction.js";
 import {} from "../../run/streams.js";
 import { chromeCdpProbeCommand, parseChromeCdpProbeOutput, } from "./cdp-probe.js";
@@ -8,10 +8,10 @@ import { shellQuote } from "../shell.js";
 import { e2bShell } from "./shell.js";
 import { findVisibleBrowserWindowId, fillDesktopBrowserWindow, } from "./desktop-browser.js";
 /**
- * The DECLARED preset to record alongside the rendered screen, or undefined when the preset
+ * The declared preset to record alongside the rendered screen, or undefined when the preset
  * rendered faithfully.
  *
- * `desktopGeometry.screen.verified` compares the FLOORED number with itself, so on its own a
+ * `desktopGeometry.screen.verified` compares the floored number with itself, so on its own a
  * floored run is indistinguishable from a faithful one: a reader sees requested 500 / verified 500
  * and concludes a 500-wide screen was asked for. Recording the declared preset is what makes
  * "the preset width did not render" legible in the bundle.
@@ -22,7 +22,7 @@ export function declaredScreenForRender(preset, presetName, rendered) {
     return { width: preset.width, height: preset.height, preset: presetName };
 }
 /**
- * Verify the desktop screen geometry IN-SANDBOX (the participant's device claim is checked, never
+ * Verify the desktop screen geometry in-sandbox (the participant's device claim is checked, never
  * assumed). A parseable mismatch fails closed. Unavailable/unparseable evidence is returned as
  * an explicit warning: the participant may still run, but its bundle records only the requested screen
  * and never upgrades that request into a verified measurement.
@@ -52,7 +52,7 @@ export async function inspectDesktopScreenGeometry(args) {
     }
     return {
         verified: { width, height, source: "xdpyinfo" },
-        error: `HUMANISH_CUA_LAB_DEVICE_GEOMETRY: participant ${args.participantId} requested a ${expectedWidth}x${expectedHeight} desktop but xdpyinfo reports ${width}x${height} in-sandbox; the participant's device geometry is unverified (fail-closed).`,
+        error: `HUMANISH_COMPUTER_USE_DEVICE_GEOMETRY: participant ${args.participantId} requested a ${expectedWidth}x${expectedHeight} desktop but xdpyinfo reports ${width}x${height} in the sandbox. The participant stops, because its screen does not match the device it declared.`,
     };
 }
 /**
@@ -298,8 +298,8 @@ export async function captureDesktopBrowserGeometry(args) {
         : undefined;
     const browserWindow = physicalWindow ?? chromeGeometry?.browserWindow;
     const viewport = chromeGeometry?.viewport;
-    // The fill check reads the X window when it was measured: under mobile emulation (#221) the
-    // page's window.outerWidth reports the EMULATED screen (414), which is not a fill failure.
+    // The fill check reads the X window when it was measured: under mobile emulation the
+    // page's window.outerWidth reports the emulated screen (414), which is not a fill failure.
     const fillBounds = physicalWindow;
     if (!browserWindow) {
         warnings.push(`Browser outer bounds could not be measured for participant ${args.participantId}.`);
@@ -314,7 +314,7 @@ export async function captureDesktopBrowserGeometry(args) {
     }
     if (!viewport) {
         // Name the cause, not only the symptom: the same dead DevTools channel that loses the viewport
-        // loses every url/text observation, and a reader of the bundle should learn that here (#514).
+        // loses every url/text observation, and a reader of the bundle should learn that here.
         const cause = cdpUnavailable === undefined ? "" : ` DevTools probe: ${redactText(cdpUnavailable)}.`;
         warnings.push(args.browserFamily === "firefox"
             ? `Browser CSS viewport measurement is unavailable in Firefox for participant ${args.participantId}; stream.viewport is omitted instead of reading a different browser's CDP endpoint.`