humanish 0.79.0 → 0.81.0

package/docs/contracts/schemas.md

@@ -3,7 +3,7 @@
 Date: 2026-06-02 (current-state note updated 2026-07-14)
 
 Status: reference map for the major contracts shipped through source version
-`0.79.0`; it is not an exhaustive inventory of command/result envelopes. Exported types,
+`0.81.0`; it is not an exhaustive inventory of command/result envelopes. Exported types,
 schema constants, parsers, and validators in `src/` are authoritative. Rows
 marked "reserved" name layering intent only — no code emits or validates them
 yet. Do not emit a reserved schema.

package/docs/goals/current.md

@@ -29,11 +29,19 @@ study completed, reproduced, and produced a real accessibility finding via a
 keyboard-first participant
 ([docs/goals/email-gated-signup/receipts/](email-gated-signup/receipts/)).
 
-## Current Program Truth (source `0.79.0`)
+## Current Program Truth (source `0.81.0`)
 
 The package source and repository implementation in this tree agree on these
 points:
 
+**Mobile input correction, 2026-09-05 (#676).** The historical 4/4 TodoMVC phone-lane rename
+failures describe the shipped desktop pointer-to-touch conversion path. In two new hosted
+conformance probes, SDK double click emitted two single clicks; direct touch opened the same
+original editor in both. A separate local native-X control reproduced the difference by toggling
+conversion. Mobile viewport and touch flags do not certify gesture equivalence or establish a
+physical-device app defect. Mobile lanes using touch conversion now carry that advisory in run
+warnings. [Method, traces summary and limits](computer-use-actor/receipts/mobile-input-conformance-2026-09-05.md).
+
 The immutable 2026-06-10 proof-roadmap packet is paired with a
 [current implementation checkpoint](https://github.com/danielgwilson/humanish/blob/main/docs/goals/proof-roadmap/README.md).
 
@@ -463,6 +471,35 @@ Stop and correct course if:
 
 ## Best Next Work
 
+**2026-09-05 (0.81.0).** First use now distinguishes the free evidence preview
+from a live participant study, with concise successful setup output and complete JSON details
+(#660). The website has runnable docs and a generated CLI reference (#661, #668). Retained participant
+reports survive automatic stops. Supported providers with retained history can add one closing
+report when time and known budget remain; clean reports such as “no confusion or hesitation”
+stay clean (#658, #670, #671). Desktop startup cleanup retains acquired handles and records
+confirmed or unknown reclamation. The SDK's detached screenshot-file cleanup rejection is
+handled (#665, #666).
+Terminal output reconciles SDK callbacks with returned aggregates, preserving legitimate
+repeated lines and usage turns while avoiding doubled capture (#672). Opt-in `openai-egress`
+auth keeps the raw OpenAI runtime key outside the sandbox; every sandbox process can still
+spend through the proxy, so this is not a provider spending limit (#663). Stock terminal
+startup installs a checksum-verified Node archive without refreshing unrelated package mirrors
+and gives newly installed npm a default global prefix on the standard PATH (#677, #680). Mobile studies warn that desktop pointer-to-touch conversion can change repeated-tap
+behavior; gesture failures require direct or native touch confirmation before app attribution
+(#678).
+
+**2026-09-04, night (0.80.0).** The observation window reaches every desktop route: the
+sequential and concurrent shared-world seats forward `dwell` the way they forward `stopWhen`
+(#645), with a plumbing test per route and two live receipts, three participants holding together
+on one shared board whose checkpoint digest stood still until they acted, and two seats holding
+in turn on one sandbox (`receipts/dwell-window-2026-09-04.md`). The sequential route refuses a
+sandbox request over the provider's 60-minute cap before any call and names the per-role ceiling
+(#649); the provider's refusal had surfaced as a bundle that failed verification. The planted-defect
+benchmark ran with a second brain, the operator's own Claude Code as the participant: 14 of 15
+reported, nothing invented in three clean runs, 72 of 75 across both brains
+(`bench/RESULTS-2026-09-04-claude.md`). A two-minute window and a camera feed of the adopter's own
+are live receipts too.
+
 **2026-09-04, evening (0.79.0).** Two of the operator's own asks, each with a live receipt. A
 declared observation window (#510): `dwell` on an actor or a lane holds the page once its condition
 matches (or from the start), captures a frame on a cadence, takes no action and requests no model
@@ -478,8 +515,9 @@ gives a hosted Chrome lane a capture device (ffmpeg's test pattern generated in
 `desktopBrowser.media`; a microphone is refused without an image that has an audio stack, before
 any spend. Live, Chrome raised its real dialog with a preview of the feed, the participant chose
 "Allow this time" and read back 640x480 (`receipts/participant-camera-2026-09-04.md`). Under
-mobile emulation the TodoMVC rename now blocks 3 of 3 phone participants; Excalidraw stays clean
-at 12 of 12.
+the then-shipped mobile-emulation input path, TodoMVC rename had stopped 3 of 3 phone
+participants at this checkpoint; Excalidraw read 12 of 12. The 2026-09-05 input-conformance
+correction above qualifies attribution of those mobile failures.
 
 **2026-09-04, later (0.78.0).** `@e2b/desktop` moved from 2.2.3 to 2.3.3 (#638): its 2.3.1
 changelog names the socket #581 found today, a background command's event stream the SDK kept
@@ -514,10 +552,10 @@ mobile-emulated browser: `execution.desktop.fidelity.mobileEmulation: true` appl
 CSS viewport, the preset's device pixel ratio, touch events and a mobile user agent before the
 participant's first observation, and the bundle records what the page then reported about itself
 under `desktopGeometry.fidelity` (#221, `docs/goals/computer-use-actor/receipts/mobile-emulation-2026-09-03.md`).
-On the published 0.76.0, with touch emulated, neither phone participant on TodoMVC could rename at
-all (a double-tap selects text; the double-click editor never opens), where the 500 px runs without
-touch had finished and only called the rename touch-hostile: the responsive-viewport study reported
-an opinion, the mobile-emulated study reported a blocker.
+On the published 0.76.0, neither phone participant could rename through the emulated input path,
+where the 500 px runs without touch had finished. Those are historical instrument observations:
+the 2026-09-05 conformance check found SDK double click and direct touch differ on the original
+editor, so the earlier outcome does not establish a touch-device app defect.
 The persona axis was replicated the same evening on drawDB and TodoMVC and given a third app,
 Excalidraw, as the clean control (`persona-axis-phone-2026-09-03.md`); a multi-lane study's
 second and third findings reach `feedback draft` through `--candidate` (#609); a negated report
@@ -554,16 +592,22 @@ and a Claude participant with no memory across turns (#520). `humanish stats` (#
 
 The standing queue, in rough order:
 
-1. #581, the remaining half: a `Sandbox.create` that fails after the API allocated the sandbox
+1. #509, the microphone tier: a custom desktop image with an audio stack
+   (`execution.desktop.template`), then `media.microphone` threaded into the launch the way the
+   camera is; the parse-time refusal stands until then;
+2. #581, the remaining half: a `Sandbox.create` that fails after the API allocated the sandbox
    (the desktop SDK's own startup) leaves a sandbox this run never learned the id of, reclaimed only
    by the provider's timeout; the linger itself is the SDK's held command stream, closed by the
-   2.3.3 bump (0.78.0), and `doctor` names an older SDK;
-2. #513: there is still nowhere to read the docs on humanish.dev (site copy is locked; a docs
+   2.3.3 bump (0.78.0), and `doctor` names an older SDK; the upstream ask is drafted for the
+   operator;
+3. #221's later tier, a real device or simulator behind the emulated lane, and #623's
+   scripted-browser route, which has no model loop and so no dwell or later-tab applier;
+4. #513: there is still nowhere to read the docs on humanish.dev (site copy is locked; a docs
    route needs the operator's sign-off);
-3. the TUI views over `stats` and `export` (#455's metrics screen and `s` key), design-gated;
-4. the launch post: a draft with every number and its receipt path sits outside the public repo,
+5. the TUI views over `stats` and `export` (#455's metrics screen and `s` key), design-gated;
+6. the launch post: a draft with every number and its receipt path sits outside the public repo,
    awaiting the operator; nothing has been announced;
-5. registry promotions (#431) and the shared-world honesty half (#365, #446), unchanged.
+7. registry promotions (#431) and the shared-world honesty half (#365, #446), unchanged.
 
 Earlier state, kept for the record:
 

package/docs/ramp/README.md

@@ -2,7 +2,7 @@
 
 Status: public-safe contributor and agent ramp.
 
-Package/source version in this tree: `0.79.0` (2026-09-04). The Observer is phone-usable as a stated requirement (observer/AGENTS.md); interactive primitives start from Base UI. The Observer renderer is the observer/ workspace artifact only; the legacy string-concat renderer was deleted at cutover (#426), and rollback is a version pin to 0.42.0. The containment boundary introduced in
+Package/source version in this tree: `0.81.0` (2026-09-05). The Observer is phone-usable as a stated requirement (observer/AGENTS.md); interactive primitives start from Base UI. The Observer renderer is the observer/ workspace artifact only; the legacy string-concat renderer was deleted at cutover (#426), and rollback is a version pin to 0.42.0. The containment boundary introduced in
 `0.15.1` remains in force: managed run and output paths bind to validated
 physical filesystem identities, and stored provider IDs are evidence, not
 cleanup authority. The bundled OSS meta-lab is dry-run only until

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "humanish",
-  "version": "0.79.0",
+  "version": "0.81.0",
   "description": "Open-source-safe CLI for persona simulation, observer review, and public-safe feedback drafts.",
   "author": "Daniel G Wilson <daniel@danielgwilson.com>",
   "keywords": [
@@ -73,7 +73,9 @@
     "tui:typecheck": "pnpm --filter humanish-tui typecheck",
     "tui:smoke": "node scripts/tui-smoke.mjs",
     "tui:test": "pnpm --filter humanish-tui test",
-    "release:dogfood": "node scripts/release-dogfood.mjs"
+    "release:dogfood": "node scripts/release-dogfood.mjs",
+    "docs:generate": "tsx scripts/generate-cli-docs.ts",
+    "docs:check": "tsx scripts/generate-cli-docs.ts --check"
   },
   "repository": {
     "type": "git",