From 2f7ddc6dd3775fa8dedf6bfcbc94050662d158d2 Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Thu, 6 Aug 2026 13:26:29 -0700
Subject: [PATCH] [eric] browser-v3: move the measurement harness into the
 repo, out of a temp dir

---
 e2e/browser-v3/.gitignore                |   4 +
 e2e/browser-v3/HANDOFF.md                | 170 +++++++++++++
 e2e/browser-v3/HOLDOUT_FROZEN.md         | 100 ++++++++
 e2e/browser-v3/README.md                 | 106 +++++++++
 e2e/browser-v3/bench.py                  | 252 ++++++++++++++++++++
 e2e/browser-v3/c2_rounds.sh              |  35 +++
 e2e/browser-v3/c2_tally.py               |  67 ++++++
 e2e/browser-v3/c7_run.sh                 |  24 ++
 e2e/browser-v3/c8_run.sh                 |  18 ++
 e2e/browser-v3/coverage.py               | 289 +++++++++++++++++++++++
 e2e/browser-v3/keep_renderer.sh          |  23 ++
 e2e/browser-v3/probe_evidence.py         | 110 +++++++++
 e2e/browser-v3/rawbrowser.py             |  98 ++++++++
 e2e/browser-v3/results/holdout_raw.jsonl |  20 ++
 e2e/browser-v3/results/known_raw.jsonl   | 158 +++++++++++++
 e2e/browser-v3/skillstats.py             |  89 +++++++
 e2e/browser-v3/stack.sh                  | 129 ++++++++++
 e2e/browser-v3/verify_markers.py         |  86 +++++++
 18 files changed, 1778 insertions(+)
 create mode 100644 e2e/browser-v3/.gitignore
 create mode 100644 e2e/browser-v3/HANDOFF.md
 create mode 100644 e2e/browser-v3/HOLDOUT_FROZEN.md
 create mode 100644 e2e/browser-v3/README.md
 create mode 100644 e2e/browser-v3/bench.py
 create mode 100755 e2e/browser-v3/c2_rounds.sh
 create mode 100644 e2e/browser-v3/c2_tally.py
 create mode 100755 e2e/browser-v3/c7_run.sh
 create mode 100755 e2e/browser-v3/c8_run.sh
 create mode 100644 e2e/browser-v3/coverage.py
 create mode 100755 e2e/browser-v3/keep_renderer.sh
 create mode 100644 e2e/browser-v3/probe_evidence.py
 create mode 100644 e2e/browser-v3/rawbrowser.py
 create mode 100644 e2e/browser-v3/results/holdout_raw.jsonl
 create mode 100644 e2e/browser-v3/results/known_raw.jsonl
 create mode 100644 e2e/browser-v3/skillstats.py
 create mode 100755 e2e/browser-v3/stack.sh
 create mode 100644 e2e/browser-v3/verify_markers.py

diff --git a/e2e/browser-v3/.gitignore b/e2e/browser-v3/.gitignore
new file mode 100644
index 00000000..c949a140
--- /dev/null
+++ b/e2e/browser-v3/.gitignore
@@ -0,0 +1,4 @@
+# Run output: logs, per-trial artifacts, and a multi-gigabyte Electron profile with live logins.
+# The raw jsonl rows under results/ ARE committed; they are the evidence. Nothing else here is.
+runs/
+results/*.log
diff --git a/e2e/browser-v3/HANDOFF.md b/e2e/browser-v3/HANDOFF.md
new file mode 100644
index 00000000..0b2ea1df
--- /dev/null
+++ b/e2e/browser-v3/HANDOFF.md
@@ -0,0 +1,170 @@
+# Handoff: browser v3, state as of 2026-08-06
+
+Branch `eric/browser-merged`, 41 commits past `origin/eric/dev`. Browser suite 707 passing, tsc 0,
+linter no new violations, tree clean. The goal and the harness are described in `README.md`.
+
+**Six of nine criteria pass. Three are open, and none of them is open because the fix is unknown.**
+
+## Scorecard
+
+| # | criterion | target | before | now | |
+| --- | --- | --- | --- | --- | --- |
+| 1 | composer reach | >=90% | 57% | **93%** (28/30) | PASS |
+| 2 | verified writes | >=95% | no honest data | LinkedIn proven end to end; N too small | OPEN |
+| 3 | false success | 0 | unknown | **0** in every live round | PASS |
+| 4 | median write | <=12s | ~21s | **1.99s** (p95 12.2s) | PASS |
+| 5 | cold start tier-0/1 | <=3s | ~16s | **0.13s** | PASS |
+| 6 | other_ms | -50% | none | **-70%** (417 -> 126ms) | PASS |
+| 7 | infra flake | <=1% | 60% | 34 runs, 0 failures; sample too small | OPEN |
+| 8 | holdout | >=80%, <=10pt | untested | **87%**, 6pt gap | PASS |
+| 9 | learned path | remove or >=50% | 0/55 | recording bug fixed; live rate unmeasured | OPEN |
+
+Per-site reach at N=5 (criterion 1): x 5/5, linkedin 5/5, reddit 5/5, **instagram 5/5** (was 1/10),
+youtube 4/5, **twitch 4/5** (was 0/3). gmail, tiktok and substack are excluded, see Exclusions.
+
+Timing at n=28 successful runs (criterion 4/6): wall median 1988ms / p95 12180ms; browser tools
+median 1621ms; **`other_ms` median 126ms, down from 417ms**. Wall fell too, so nothing moved from one
+bucket into another.
+
+## The finding that matters most
+
+**Several "product failures" were the measuring instrument.** This is the single most useful thing to
+carry forward, because it recurred six times in one session and each instance cost hours chasing a
+bug that did not exist.
+
+The worst case: criterion 2 failed on LinkedIn for days. **LinkedIn was never broken.** The canary
+proved a write landed by grepping the backend log for the marker string, and the backend does not log
+page text. A grep for every canary marker ever generated, across every backend log on the machine,
+returns zero lines. The audit could only ever answer "could not look", and that was being read as a
+product failure. `probe_evidence.py` settled it: the session API carries the model's *answer* and
+never the tool results, so auditing it is asking the same model whose claim is under audit.
+
+Every dead grep found, and what each did:
+
+| where | needle | occurrences in source |
+| --- | --- | --- |
+| canary `delivered` | `DELIVERY CONFIRMED` | 0 |
+| canary `saw_page` | two `[browser-action] X` strings | 0 |
+| canary receipt | only the fast-lane string; 2 of 3 producers missed | 1 of 3 |
+| skillstats `replay_full` | `replay(ed\|ing) N steps` | 0 |
+| bench `infra_browser` | `card is unavailable` | 0 |
+| `stack.sh status` | `pgrep -fc` (no such flag on macOS) | printed 0 over a live stack |
+
+`verify_markers.py` now checks all 32 harness literals against the source that prints them: 32/32
+present. **Run it before trusting any number.**
+
+## What is open, why, and exactly what to do
+
+### Criterion 2, verified writes
+
+The instrument was rebuilt (`950755af`) and the first result was linkedin **PASS**: posted,
+receipt-verified, deleted, verified gone. That is the first clean end-to-end LinkedIn round this
+project has recorded. One site is not >=95% across sites, so the criterion is not met.
+
+Blocked on nothing technical. Two things to do first:
+
+1. **Verify reddit's Title field name.** `af68dcd0` prefers the textbox whose accessible name
+   contains the field word the task used, and requires *exactly one* match. The unit tests use an
+   assumed listing (`[21]<textbox "Title" />`) that has never been confirmed against the live page;
+   the backend only logs `textboxes=3`. If reddit's title input is named something else, the fix
+   silently does nothing. reddit is 1 of the 3 sites in this denominator.
+2. Then `ROUNDS=7 c2_rounds.sh` and `c2_tally.py`.
+
+### Criterion 7, infrastructure flake
+
+Last clean sample: **34 of 108 runs, 0 infra failures, 0 backend restarts.** Encouraging but partial.
+An earlier 5.4% reading was contaminated: a second OpenSwarm checkout was up on :8324 and the shared
+9router on :20128 was being evicted.
+
+Needs a box with nothing on :8324, then `N=12 c7_run.sh` for the full 108.
+
+### Criterion 9, learned fast path
+
+Baseline is now measured properly: **427 gate decisions, 95 eligible, 0 recorded (0%), 0 replays**,
+with one refusal reason (`host empty or no robust steps`).
+
+The gate was never the problem. Every eligible run died inside `record_skill`: `distill_steps` reads
+`clicked_name`, and `browser_send_script.py` wrote the element name into `result_summary` prose only.
+An unnameable click truncates the distillation, truncation drops the typing steps, and the
+navigation-only remainder is correctly refused. Fixed in `438a96eb`, proven directly:
+
+```
+OLD shape (no clicked_name) -> []
+NEW shape -> ['BrowserNavigate', 'BrowserClickByName', 'BrowserClickByName']
+```
+
+Safety held rather than added: a `BrowserClickIndex` distills to a name-only `BrowserClickByName`, so
+the payload never enters the skill, and the send click's tool name matches no distill branch, so a
+replay cannot re-fire a send.
+
+**Not yet proven live.** A later dry sweep still showed 0/11, but the eligible hosts were
+`accounts.google.com`, `substack.com` and `twitch.tv` (signed-out and no-composer runs), which the
+fix does not touch. Separately, `skillstats.py` reported 8 prefix replays alongside 0 "matched",
+which cannot both be true, so it has a counting bug of its own. **The honest reading is that
+criterion 9 has no verdict yet, not that the fix failed.**
+
+Fix the `skillstats` counter, then read the recording rate off a fresh 108-run sweep. If replay shows
+no benefit on off-table hosts, removal is the honest answer and the criterion explicitly allows it;
+the two `browser-memory` endpoints have no frontend caller, so removal is not user-visible.
+
+## Product bugs found by the fixed instrument
+
+- **reddit: the task named a field and nobody read it** (`af68dcd0`). "create a text post whose title
+  is exactly X" filled the body, Title stayed empty, and reddit's submit stayed DISABLED on every
+  attempt. A field word from the user's own sentence now beats the compose-shape guess. 6 tests,
+  including that no-hint behaviour is byte-identical.
+- **disqus: a stalled top document denied its own child frames** (`7ba99ba4`). The still-loading
+  retry rethrew on a second failure, killing `find_composer` before the child-frame search ran. An
+  ad-heavy page keeps the top document loading while the composer, in an embedded iframe with its own
+  load state, is perfectly readable.
+
+## Known false negatives, filed not fixed
+
+Both under-claim rather than over-claim, so neither violates criterion 3, but both read as drift on
+every live round.
+
+- **A torn-down browser card wedges, and `navigate` lies about it.** Reproduced 4x: the reply is
+  `{"text":"Navigated to https://x.com/home","url":"https://x.com/home"}` in ~80ms while the webview
+  never leaves reddit. Anything trusting that reply reads the wrong page and answers confidently
+  about it. The audit now re-reads `location.href` and requires a host match.
+- **Delete reports failure on deletions that provably worked.** X, 2 for 2: "the deletion was never
+  confirmed", and an independent read shows the post gone both times. LinkedIn showed the same shape.
+
+## Exclusions (predefined, never quietly dropped)
+
+gmail, substack and tiktok are signed out; tiktok is additionally captcha-walled. That is 14 of 45
+rows on the known suite. Each exclusion is judged on the page's own evidence, never on the product's
+claim. Solving a bot-detection challenge is off-limits, so a captcha-walled page is one this system
+is *choosing* not to reach, and scoring it against reach would charge us for a rule we intend to keep.
+
+## Account hygiene
+
+Every marker written to a real account during this work was cleaned up and independently verified
+gone: X profile scan shows 0 markers with the newest genuine post predating the tests, and the
+LinkedIn post permalink returns nav chrome with no content.
+
+Personal handles are **not** committed. Both live sites read their account from the environment
+(`OSW_CANARY_X_HANDLE`, `OSW_CANARY_REDDIT_HANDLE`), and an unset handle makes the round refuse
+loudly rather than audit a malformed URL.
+
+## Environment notes
+
+- The stack is backend :8326, webpack :3026, Electron on its own `--user-data-dir`. It never touches
+  :8324 or :3000, which belong to whatever else is on the machine.
+- The backend was SIGTERMed three times mid-measurement, always cleanly, always with nothing in its
+  log. Cause unknown. `stack.sh` now supervises it and writes `[stack] BACKEND RESTARTED` into the log
+  the harness slices, so `bench.py`'s `infra_backend` bucket can exclude any trial spanning a restart
+  instead of scoring it as a product failure.
+- `stack.sh down` once matched `pkill -f "uvicorn backend.main"`, which is exactly what the other
+  checkout runs, and it executed while that checkout was live. It is now scoped to this stack's own
+  ports and profile and cannot match a process it did not start.
+
+## Next executable steps, in order
+
+1. `verify_markers.py` (32/32 expected).
+2. Confirm reddit's Title field name against the live page.
+3. Fix the `skillstats.py` replay counter.
+4. `N=12 c7_run.sh` on a box with nothing on :8324. Gives criteria 1, 4, 5, 6, 7.
+5. `skillstats.py` on that sweep's log. Gives criterion 9's live recording rate.
+6. `ROUNDS=7 c2_rounds.sh` then `c2_tally.py`. Gives criterion 2.
+7. `N=2 c8_run.sh` to re-check criterion 8 after the disqus fix.
diff --git a/e2e/browser-v3/HOLDOUT_FROZEN.md b/e2e/browser-v3/HOLDOUT_FROZEN.md
new file mode 100644
index 00000000..71946018
--- /dev/null
+++ b/e2e/browser-v3/HOLDOUT_FROZEN.md
@@ -0,0 +1,100 @@
+# v3 holdout set, FROZEN
+
+Frozen 2026-08-02 before any site-specific v3 work. Repo HEAD at freeze: `1508e03c`.
+
+Integrity claim: no commit touching twitch, tiktok, or instagram behaviour exists at or before
+`1508e03c`. Verify with `git log 1508e03c --oneline`. If a later commit tunes any host below, this
+holdout is burned and a new one must be frozen.
+
+Selection rule: the host string appears NOWHERE under `backend/apps/agents/browser/`. Checked
+against the 106 hosts the module references (`seed_playbooks.py` carries most of them). Candidates
+rejected for contamination: news.ycombinator.com, soundcloud.com, quora.com, pinterest.com,
+wikipedia.org, github.com, stackoverflow.com, medium.com.
+
+## The six
+
+| host | composer architecture | task |
+| --- | --- | --- |
+| bsky.app | React SPA, modal composer | create a post |
+| mastodon.social | inline composer on the home timeline | create a post |
+| dev.to | URL-routed form (`/new`), markdown textarea | create a post |
+| lobste.rs | server-rendered form (`/stories/new`) | create a story |
+| pastebin.com | plain textarea, **no login required** | create a paste |
+| meta.discourse.org | Discourse SPA, "New Topic" composer | create a topic |
+
+Spread is deliberate: 2 plain forms, 2 SPA modals, 1 inline, 1 URL-routed. If reach holds on the
+plain forms and collapses on the SPA modals, that localises the generalisation failure instead of
+producing one useless aggregate.
+
+## Rules
+
+1. **Reach only, in dry run.** No holdout submit, ever. These are not the user's accounts.
+2. **No tuning against these hosts** until the first evaluation is recorded in `holdout_run1.json`.
+3. Sign-in state is measured, never assumed. A signed-out host is reported NOT MEASURABLE with the
+   page's own evidence and leaves the denominator, exactly as gmail does on the known suite.
+   pastebin.com is the one host guaranteed measurable regardless of session state, which is why it
+   is in the set.
+4. Every attempt is published, including failures, retries, and exclusions.
+
+## Addendum, frozen 2026-08-02 at HEAD `ab898964`, before evaluating any of it
+
+The first holdout was unmeasurable: 5 of 6 sites are signed out and I am not permitted to sign in,
+so it tested nothing. It did earn its keep once (pastebin exposed the header-login-widget bug), but
+a holdout that cannot be scored is not a generalisation test.
+
+Fix: six more hosts, chosen because they publish a composer to anonymous users, so session state can
+never be the reason a run fails. Same contamination rule, all six verified absent from everything
+under `backend/apps/agents/browser/`. Frozen before the first run, per criterion 8.
+
+| host | composer architecture | task |
+| --- | --- | --- |
+| rentry.co | plain textarea, markdown | create a paste |
+| dpaste.org | plain textarea in a server form | create a paste |
+| controlc.com | plain textarea | create a paste |
+| txti.es | plain textarea, minimal markup | create a page |
+| justpaste.it | rich contenteditable editor | create a note |
+| telegra.ph | rich contenteditable (Telegram's editor) | create a page |
+
+Four plain textareas and two contenteditable rich editors on purpose. The known suite's wins are
+almost all contenteditable (x, linkedin, youtube, twitch), so a holdout of only rich editors would
+flatter us, and one of only textareas would not exercise the path that actually carries production.
+
+Same rules as above: reach only, dry run, never submitted, every attempt published.
+
+## Editor-shape addendum, frozen 2026-08-04 at HEAD `e445ca3e`, before evaluating any of it
+
+Eric's observation, and it is the sharpest critique of this benchmark so far: the suite was picked by
+site POPULARITY, and popularity is not what determines whether we generalize. The **editor library**
+is. The web's writing surfaces cluster into roughly eight shapes and most sites just adopt one, so
+coverage should be counted per shape, not per famous site.
+
+What the existing suites actually cover, audited:
+
+| shape | covered by | samples |
+| --- | --- | --- |
+| plain `<textarea>` | rentry, dpaste, controlc, txti, pastebin | 5 |
+| raw contenteditable | telegra.ph, justpaste.it | 2 |
+| Draft.js family | x.com | 1 |
+| Slate / Lexical | twitch | 1 |
+| web components / shadow DOM | youtube | 1 |
+| multi-field form | reddit | 1 |
+| **iframe-embedded composer** | gmail (SIGNED OUT) | **0 measurable** |
+| **Quill / TinyMCE / CKEditor** | nothing | **0** |
+| canvas-based (Docs, Figma) | nothing | 0, likely out of scope |
+
+The last two rows are the real hole, and they are not niche: every WordPress admin, every helpdesk
+reply box, and every Disqus thread lives there. Disqus is the highest-value single target because it
+is BOTH an iframe AND a rich editor, and it is embedded on millions of sites.
+
+Frozen set, all verified absent from everything under `backend/apps/agents/browser/`, and all public
+demo pages so no login is required and grading stays honest:
+
+| host | shape | task |
+| --- | --- | --- |
+| disqus.com | iframe + rich editor | leave a comment on the demo thread |
+| quilljs.com | Quill | write in the playground editor |
+| tiny.cloud | TinyMCE | write in the demo editor |
+| ckeditor.com | CKEditor 5 | write in the demo editor |
+| codepen.io | CodeMirror in an iframe | write in the HTML pane |
+
+Reach only, dry run, never submitted. Every attempt published, including failures and exclusions.
diff --git a/e2e/browser-v3/README.md b/e2e/browser-v3/README.md
new file mode 100644
index 00000000..b61d3216
--- /dev/null
+++ b/e2e/browser-v3/README.md
@@ -0,0 +1,106 @@
+# Browser v3: the goal, where we are, and how to pick this up
+
+This directory is the measurement harness for the browser write path, plus the evidence it has
+produced so far. It exists because the browser agent's numbers were repeatedly wrong in ways that
+looked like product bugs, and the only defence is an instrument you can re-run and audit.
+
+**Read `HANDOFF.md` first if you are continuing this work.** It has the current scorecard, the three
+open criteria, and the exact next commands.
+
+## The goal
+
+Nine exit criteria, all of which must hold at once:
+
+| # | criterion | target | baseline |
+| --- | --- | --- | --- |
+| 1 | composer reach across the known suite | >=90% | 57% |
+| 2 | verified writes across sites and repetitions | >=95% | no honest data |
+| 3 | false success claims | **exactly 0** | unknown |
+| 4 | median successful-write wall time | <=12s | ~21s |
+| 5 | cold-start / prestage on a tier-0/1 hit | <=3s | ~16s |
+| 6 | `other_ms` (the part our code owns) | -50% | unmeasured |
+| 7 | infrastructure flake over >=100 site-runs | <=1% | 60% |
+| 8 | frozen-holdout reach, and its gap to the known set | >=80%, <=10pt | untested |
+| 9 | learned fast path: remove it, or >=50% recording with real replays | either | 0 of 55 |
+
+Definitions that decide arguments later:
+
+- **Reach** = navigated to the intended surface AND page-verified the correct editable composer.
+  Filling *a* box is not reach; filling *the* box is.
+- **Verified write** = the exact content was independently confirmed at the correct destination.
+  Typed-only, unsubmitted, wrong-composer and unverified all count as failures.
+- **False success** = reporting success without postcondition evidence. This is the hard gate.
+- Unknowns, timeouts and silent failures are failures, never omissions.
+
+## Layout
+
+| file | what it does |
+| --- | --- |
+| `coverage.py` | **The one grader.** Site tasks, surface rules, exclusion logic. Never reimplement grading elsewhere; two graders drift and the harness starts lying in a new way. |
+| `bench.py` | N trials per site, infra-vs-product bucketing, timing split, one artifact per trial. |
+| `stack.sh` | Boots the isolated stack (backend :8326, webpack :3026, Electron on its own profile). `up dry`, `up live`, `down`, `status`. |
+| `keep_renderer.sh` | Keeps an Electron renderer alive for the length of a sweep. |
+| `c7_run.sh` | Known suite at N=12 = 108 site-runs. Criteria 1, 4, 5, 6, 7 in one pass. |
+| `c8_run.sh` | The frozen holdout (criterion 8). |
+| `c2_rounds.sh` | Live write rounds on the authorised accounts (criterion 2). |
+| `c2_tally.py` | Aggregates the live rounds: verified writes, false successes, stranded markers. |
+| `skillstats.py` | Criterion 9: recording rate, replay rate, and every refusal reason. |
+| `verify_markers.py` | **Run this before trusting any number.** Checks all 32 literals the harness greps for against the source that prints them. |
+| `rawbrowser.py` | Drives a browser card with no model in the loop. The audit channel, and the tool for cleaning up a stranded test post by hand. |
+| `probe_evidence.py` | Dumps where a run's page text actually surfaces. Written to settle the question empirically instead of by assumption. |
+| `HOLDOUT_FROZEN.md` | The holdout sites and the commit each was frozen at, before its first run. |
+| `results/*.jsonl` | Raw per-trial rows. Every attempt, including failures and exclusions. |
+
+## Running it
+
+Everything needs the isolated stack and a quiet box.
+
+```bash
+e2e/browser-v3/stack.sh up dry          # dry: the backend refuses the irreversible click
+N=12 e2e/browser-v3/c7_run.sh           # criteria 1, 4, 5, 6, 7
+N=2  e2e/browser-v3/c8_run.sh           # criterion 8
+e2e/browser-v3/skillstats.py runs/*.log # criterion 9
+e2e/browser-v3/stack.sh down
+```
+
+Live write tests post to real accounts, so they need explicit handles and a live backend:
+
+```bash
+e2e/browser-v3/stack.sh up live
+OSW_CANARY_X_HANDLE=<handle> OSW_CANARY_REDDIT_HANDLE=<handle> \
+  ROUNDS=7 e2e/browser-v3/c2_rounds.sh
+e2e/browser-v3/c2_tally.py runs/c2_r7.txt
+```
+
+`OSW_BENCH_DIR` relocates run output (logs plus a multi-gigabyte browser profile) off the repo disk.
+
+## Rules this harness enforces on itself
+
+These are not style preferences. Each one is a bug that already shipped a wrong number.
+
+1. **Never grep for a string the code does not print.** Five separate dead greps were found in one
+   session, each silently turning a metric into a constant. `verify_markers.py` exists to make that
+   class impossible. Run it first.
+2. **Absence is not evidence unless the read succeeded.** "The marker is not on the page" and "we
+   never got a good look at the page" are different answers. Collapsing them into `False` falsely
+   accused a working LinkedIn send of lying.
+3. **Never grade the guard with the guard.** The audit reads the destination itself and consults no
+   model. A receipt reporting on its own correctness proves nothing.
+4. **Publish every attempt, exclusion and retry.** An excluded row is a claim that the product was
+   not on trial, and that claim has to survive being read out loud.
+5. **A site being signed out is an account state, not a coverage failure.** Exclusions are judged on
+   the page's own evidence (a sign-in URL, a visible password field, a bot-detection challenge),
+   never on the product's say-so.
+6. **One backend per box.** A second OpenSwarm evicts the shared 9router and every timing becomes
+   noise: the same sweep once scored 1/9 and 4/9 with zero code change. `stack.sh status` warns when
+   :8324 is occupied, and `stack.sh down` is scoped so it can never kill another checkout's stack.
+
+## Safety
+
+- Live rounds write only to accounts the operator explicitly names by environment variable, and every
+  post is deleted in the same run and re-verified gone.
+- Markers are random and carry no removal words, which would otherwise trip the removal classifier.
+- Dry mode is enforced by the backend (`OSW_SENDSCRIPT_DRYRUN=1`), never by a flag this harness sets
+  and does not own. An earlier draft set it in the wrong process, which would have posted for real.
+- The canary prints its marker on every row, not just failures. It is the only record of what was put
+  on a real account, so a stranded post stays cleanable by hand.
diff --git a/e2e/browser-v3/bench.py b/e2e/browser-v3/bench.py
new file mode 100644
index 00000000..5f65242f
--- /dev/null
+++ b/e2e/browser-v3/bench.py
@@ -0,0 +1,252 @@
+"""v3 benchmark: N trials per site, structured traces, honest denominators.
+
+Two things this does that the old sweep did not.
+
+1. N trials. One run per site is noise: on 2026-08-02 x went 2-of-3 and instagram 1-of-3, and a
+   single-run number would have called either one whatever the last run happened to be.
+2. It separates INFRASTRUCTURE failure from PRODUCT failure and publishes both. A router death is
+   not a coverage miss, and hiding it inflates the score; calling a real bug "infrastructure" is the
+   mistake that hid the surrogate crash for a whole session. Every attempt lands in the raw file.
+
+Timings come from browser_metrics/tasks.jsonl (the wall/llm/tools/other split), not from log
+scraping, because the wall clock cannot tell "our code got faster" from "the model took fewer turns".
+
+Grading is imported from coverage.py, never reimplemented. Two graders would drift and the harness
+would start lying again in a new way.
+"""
+import json
+import os
+import statistics
+import sys
+import time
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+# coverage.py sits BESIDE this file. Pointing one directory higher was a live trap rather than a
+# typo: PyPI ships a package called `coverage` (the code-coverage tool), so on any machine that has
+# it installed `import coverage` would quietly bind to that instead of the grader, and every verdict
+# would come from a module that has never heard of a composer. Insert this directory FIRST.
+sys.path.insert(0, HERE)
+import coverage as C  # noqa: E402  the ONE grader
+
+assert hasattr(C, "TASKS"), f"imported the wrong 'coverage' module: {C.__file__}"
+
+ROOT = os.path.dirname(os.path.dirname(HERE))
+METRICS = os.path.join(ROOT, "backend/data/browser_metrics/tasks.jsonl")
+
+# Frozen 2026-08-02 at HEAD 1508e03c, before any site-specific v3 work. See HOLDOUT_FROZEN.md.
+# Reach only, dry run, never submitted: these are not the user's accounts.
+HOLDOUT = {
+    "bsky":      'Go to bsky.app and create a post with exactly this text: "coverage probe alpha"',
+    "mastodon":  'Go to mastodon.social and create a post with exactly this text: "coverage probe alpha"',
+    "devto":     'Go to dev.to and start a new post with the body "coverage probe alpha"',
+    "lobsters":  'Go to lobste.rs and start a new story with the text "coverage probe alpha"',
+    "pastebin":  'Go to pastebin.com and create a new paste with the text "coverage probe alpha"',
+    "discourse": 'Go to meta.discourse.org and start a new topic with the body "coverage probe alpha"',
+}
+HOLDOUT_PAGE = {
+    "bsky":      r"bsky\.app",
+    "mastodon":  r"mastodon\.social",
+    "devto":     r"dev\.to/new",
+    "lobsters":  r"lobste\.rs/stories/new",
+    "pastebin":  r"pastebin\.com",
+    "discourse": r"discourse\.org",
+}
+# Addendum, frozen at HEAD ab898964 before its first run. The six above are mostly signed out and
+# therefore unscoreable; these publish a composer to anonymous users, so a failure is always OUR
+# failure and never a session state. 4 plain textareas + 2 contenteditable rich editors, because a
+# holdout of only rich editors would flatter a suite whose wins are nearly all contenteditable.
+HOLDOUT.update({
+    "rentry":    'Go to rentry.co and create a paste with the text "coverage probe alpha"',
+    "dpaste":    'Go to dpaste.org and create a paste with the text "coverage probe alpha"',
+    "controlc":  'Go to controlc.com and create a paste with the text "coverage probe alpha"',
+    "txti":      'Go to txti.es and create a page with the text "coverage probe alpha"',
+    "justpaste": 'Go to justpaste.it and create a note with the text "coverage probe alpha"',
+    "telegraph": 'Go to telegra.ph and create a page with the text "coverage probe alpha"',
+})
+HOLDOUT_PAGE.update({
+    "rentry":    r"rentry\.co",
+    "dpaste":    r"dpaste\.org",
+    "controlc":  r"controlc\.com",
+    "txti":      r"txti\.es",
+    "justpaste": r"justpaste\.it",
+    "telegraph": r"telegra\.ph",
+})
+
+# Infrastructure, not coverage. Kept separate and PUBLISHED, never folded into either column.
+INFRA = (
+    # The backend itself. It has died mid-sweep on a clean SIGTERM with nothing in its log
+    # (2026-08-05), and an unnoticed death turns every following trial into a fake product failure.
+    # `BACKEND RESTARTED` is written by the supervisor in stack.sh; BACKEND_DOWN by the harness when
+    # it cannot reach :8326 at all.
+    ("infra_backend", (C.BACKEND_DOWN, "[stack] BACKEND RESTARTED")),
+    ("infra_router", ("9Router watchdog", "9Router process died", "No AI provider connected")),
+    # The webview is gone or wedged. "card is unavailable" was in this list and appears NOWHERE in
+    # the codebase, so a third of this bucket could never match; these are P_CARD_GONE_MARKERS
+    # (browser_loop.py) plus the renderer's own read-timeout wording, all grep-verified.
+    ("infra_browser", ("Browser command timed out", "too busy to read", "not an electron webview",
+                       "no dashboard is connected", "page unresponsive")),
+    # No renderer attached. Its own bucket because it is neither the router nor the page: Electron
+    # launched before webpack was serving, hit a dead URL and quit, and 44 straight runs then failed
+    # with "dispatch refused: no dashboard". Filed as generic harness trouble that read as a mystery;
+    # named, it says exactly which process to restart.
+    ("infra_no_renderer", ("dispatch refused: no dashboard", "no OpenSwarm window is connected")),
+)
+
+
+# Editor-shape addendum, frozen at HEAD e445ca3e before its first run. Coverage should be counted per
+# EDITOR LIBRARY, not per famous site: the web's writing surfaces cluster into ~8 shapes and most
+# sites adopt one. These five fill the two holes the popularity-picked suites left entirely empty,
+# iframe-embedded composers and the classic CMS editors. All public demos, so no login gates them.
+HOLDOUT.update({
+    "disqus":    'Go to https://blog.disqus.com/ and open the first blog post, then write a '
+                 'comment "coverage probe alpha" in the Disqus comment box at the bottom',
+    "quill":     'Go to quilljs.com/playground and write "coverage probe alpha" in the editor',
+    "tinymce":   'Go to tiny.cloud and write "coverage probe alpha" in the demo editor',
+    "ckeditor":  'Go to ckeditor.com/ckeditor-5/demo and write "coverage probe alpha" in the editor',
+    "codemirror": 'Go to codepen.io/pen and write "coverage probe alpha" in the HTML pane',
+})
+HOLDOUT_PAGE.update({
+    "disqus": r"disqus\.com", "quill": r"quilljs\.com", "tinymce": r"tiny\.cloud",
+    "ckeditor": r"ckeditor\.com", "codemirror": r"codepen\.io",
+})
+
+
+# Buckets that leave the reach denominator. Defined ONCE: the two call sites used to list them by
+# hand and a new bucket added to only one of them silently changes the score.
+EXCLUDED = ("not_measurable", "infra_backend", "infra_router", "infra_browser",
+            "infra_harness", "infra_no_renderer")
+
+
+def classify(v, slice_):
+    """One bucket per trial. Unknowns and timeouts are failures, never silently dropped."""
+    for name, needles in INFRA:
+        if any(n in slice_ for n in needles):
+            return name
+    if v.get("invalid"):
+        return "infra_harness"
+    if v.get("unmeasurable"):
+        return "not_measurable"
+    if v["composer"]:
+        return "ok"
+    d = v["detail"]
+    if "FILL DIED" in d:
+        return "product_fill_died"
+    if "WRONG SURFACE" in d:
+        return "product_wrong_surface"
+    return "product_no_composer"
+
+
+def metrics_between(t0, t1):
+    """The browser sub-agent's own timing rows for this trial's window."""
+    out = []
+    try:
+        with open(METRICS, encoding="utf-8") as f:
+            for line in f:
+                try:
+                    r = json.loads(line)
+                except ValueError:
+                    continue
+                if t0 <= float(r.get("ts", 0)) <= t1 and r.get("llm_ms") is not None:
+                    out.append(r)
+    except OSError:
+        pass
+    return out
+
+
+def trial(site, task, n_idx):
+    t0 = time.time()
+    slice_, wall = C.run(site, task)
+    t1 = time.time()
+    v = C.verdict(site, slice_)
+    v["site"], v["trial"] = site, n_idx
+    bucket = classify(v, slice_)
+    rows = metrics_between(t0 - 2, t1 + 2)
+    best = max(rows, key=lambda r: r.get("total_ms", 0)) if rows else {}
+    rec = {
+        "site": site, "trial": n_idx, "bucket": bucket, "wall_s": wall,
+        "detail": v["detail"], "composer": bool(v["composer"]), "submit": bool(v["submit"]),
+        "total_ms": best.get("total_ms"), "llm_ms": best.get("llm_ms"),
+        "tools_ms": best.get("tools_ms"), "other_ms": best.get("other_ms"),
+        "turns": best.get("turns"),
+    }
+    art = os.path.join(HERE, "results", f"{site}_{n_idx}.log")
+    os.makedirs(os.path.dirname(art), exist_ok=True)
+    with open(art, "w", encoding="utf-8") as f:
+        f.write(slice_)
+    rec["artifact"] = os.path.relpath(art, HERE)
+    return rec
+
+
+def report(recs, label):
+    print(f"\n=== {label}: {len(recs)} attempts ===")
+    by_site = {}
+    for r in recs:
+        by_site.setdefault(r["site"], []).append(r)
+    print(f"  {'site':<11}{'reach':>8}  buckets")
+    for site, rs in by_site.items():
+        meas = [r for r in rs if r["bucket"] not in EXCLUDED]
+        ok = sum(r["bucket"] == "ok" for r in rs)
+        rate = f"{ok}/{len(meas)}" if meas else "0/0"
+        seen = {}
+        for r in rs:
+            seen[r["bucket"]] = seen.get(r["bucket"], 0) + 1
+        print(f"  {site:<11}{rate:>8}  {seen}")
+    meas = [r for r in recs if r["bucket"] not in EXCLUDED]
+    ok = [r for r in recs if r["bucket"] == "ok"]
+    infra = [r for r in recs if r["bucket"].startswith("infra")]
+    print(f"\n  REACH (product-measurable denominator): {len(ok)}/{len(meas)} "
+          f"= {round(100 * len(ok) / len(meas)) if meas else 0}%")
+    print(f"  not measurable (signed out): {sum(r['bucket'] == 'not_measurable' for r in recs)}")
+    print(f"  INFRASTRUCTURE failures: {len(infra)}/{len(recs)} "
+          f"= {round(100 * len(infra) / len(recs), 1) if recs else 0}%  {[r['bucket'] for r in infra]}")
+    lat = [r for r in ok if r.get("total_ms")]
+    if lat:
+        for key in ("total_ms", "llm_ms", "tools_ms", "other_ms"):
+            vals = sorted(r[key] for r in lat if r.get(key) is not None)
+            if vals:
+                p95 = vals[min(len(vals) - 1, int(0.95 * len(vals)))]
+                print(f"  {key:<10} median={round(statistics.median(vals))}ms  p95={p95}ms  n={len(vals)}")
+    else:
+        print("  no successful runs carried a timing row")
+
+
+USAGE = """usage: bench.py [known|holdout] [N] [site ...]
+
+  known     the known suite (default)
+  holdout   the frozen holdout, see HOLDOUT_FROZEN.md
+  N         trials per site (default 5)
+  site ...  restrict to these sites
+
+Needs the isolated stack up: e2e/browser-v3/stack.sh up dry
+"""
+
+if __name__ == "__main__":
+    # An unrecognised first argument used to be treated as a SUITE NAME, so `bench.py --help`
+    # quietly started firing real trials at a backend that was not there. The first thing anyone
+    # new types is --help, so refuse anything that is not a real suite.
+    if len(sys.argv) > 1 and sys.argv[1] not in ("known", "holdout"):
+        print(USAGE)
+        sys.exit(0 if sys.argv[1] in ("-h", "--help", "help") else 2)
+    suite = sys.argv[1] if len(sys.argv) > 1 else "known"
+    n = int(sys.argv[2]) if len(sys.argv) > 2 else 5
+    only = sys.argv[3:]
+    if suite == "holdout":
+        tasks, C.ON_PAGE = HOLDOUT, {**C.ON_PAGE, **HOLDOUT_PAGE}
+        C.PUBLIC = C.PUBLIC + tuple(HOLDOUT)
+    else:
+        tasks = C.TASKS
+    if only:
+        tasks = {k: v for k, v in tasks.items() if k in only}
+    recs = []
+    out = os.path.join(HERE, "results", f"{suite}_raw.jsonl")
+    os.makedirs(os.path.dirname(out), exist_ok=True)
+    with open(out, "a", encoding="utf-8") as fh:
+        for i in range(1, n + 1):
+            for site, task in tasks.items():
+                r = trial(site, task, i)
+                recs.append(r)
+                fh.write(json.dumps(r) + "\n")
+                fh.flush()
+                print(f"  [{i}/{n}] {site:<11}{r['bucket']:<22}{r['wall_s']:>6.1f}s  {r['detail'][:62]}",
+                      flush=True)
+    report(recs, f"{suite} suite, N={n}")
diff --git a/e2e/browser-v3/c2_rounds.sh b/e2e/browser-v3/c2_rounds.sh
new file mode 100755
index 00000000..35ae4f63
--- /dev/null
+++ b/e2e/browser-v3/c2_rounds.sh
@@ -0,0 +1,35 @@
+#!/bin/bash
+# Criterion 2: verified writes, on the three accounts authorised for real sends.
+#
+# Needs a LIVE backend (`stack.sh up live`). Every round posts a unique marker, audits the
+# destination by reading it, deletes, and audits again. Markers print on every row so anything
+# stranded can be cleaned up by hand.
+# Where logs, profiles and run output go. Defaults to runs/ beside this harness; override with
+# OSW_BENCH_DIR to keep multi-gigabyte browser profiles off the repo disk.
+SP="${OSW_BENCH_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/runs}"
+mkdir -p "$SP"
+# Repo root from this script's own location, so the harness works in any checkout.
+TREE="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+cd "$TREE" || exit 1
+OUT="$SP/c2_r7.txt"
+
+# The operator names the accounts, this file never does. These rounds POST to real profiles, so the
+# handles have to be a deliberate act by whoever runs it, not a default baked into an open-source
+# repo. Refuse rather than run: a missing handle would post and then audit a malformed URL, and the
+# round would report "unprovable" as though the product were at fault.
+for v in OSW_CANARY_X_HANDLE OSW_CANARY_REDDIT_HANDLE; do
+  if [ -z "${!v}" ]; then
+    echo "refusing: \$$v is not set. These rounds write to real accounts; name them explicitly."
+    exit 2
+  fi
+done
+
+: > "$OUT"
+for i in $(seq 1 "${ROUNDS:-7}"); do
+  echo "########## ROUND $i $(date +%H:%M:%S)" >> "$OUT"
+  OSW_CANARY_BASE=http://127.0.0.1:8326 OSW_CANARY_LOG="$SP/r7_be.log" \
+    timeout 2400 ./backend/.venv/bin/python scripts/browser_canary.py --live --sites x,linkedin,reddit \
+    >> "$OUT" 2>&1
+  echo "  (backend restarts so far: $(grep -c 'BACKEND RESTARTED' "$SP/r7_be.log" 2>/dev/null))" >> "$OUT"
+done
+echo "########## DONE $(date +%H:%M:%S)" >> "$OUT"
diff --git a/e2e/browser-v3/c2_tally.py b/e2e/browser-v3/c2_tally.py
new file mode 100644
index 00000000..ea751c30
--- /dev/null
+++ b/e2e/browser-v3/c2_tally.py
@@ -0,0 +1,67 @@
+"""Criterion 2 across rounds: verified writes, false successes, and every marker that was left live.
+
+The canary prints a per-round summary; the criterion asks about the whole sample. Parses the round
+output rather than re-running anything, so the numbers can be re-derived from the artifact.
+
+    python3 c2_tally.py c2_r7.txt
+"""
+
+import re
+import sys
+from collections import Counter
+
+ROW = re.compile(r"^\s+\[(PASS|DRIFT)\]\s+(\S+)\s+(canary[0-9a-f]+)?\s*stage=(\S+)\s+(.*)$")
+PROVEN = re.compile(r"verified writes: (\d+)/(\d+) proven")
+LIARS = re.compile(r"FALSE SUCCESS CLAIMS: (\d+)")
+# re.M or `$` only matches the very end of the file and this finds nothing at all,
+# which reads as "nothing was stranded" -- the exact wrong direction for this line.
+STRANDED = re.compile(r"MANUAL CLEANUP NEEDED: (.+)$", re.M)
+
+
+def main() -> int:
+    text = open(sys.argv[1], errors="ignore").read()
+    rounds = text.count("########## ROUND")
+    proven = denom = liars = 0
+    for m in PROVEN.finditer(text):
+        proven += int(m.group(1))
+        denom += int(m.group(2))
+    for m in LIARS.finditer(text):
+        liars += int(m.group(1))
+
+    per_site = {}
+    stages = Counter()
+    for line in text.splitlines():
+        m = ROW.match(line)
+        if not m:
+            continue
+        flag, site, marker, stage, detail = m.groups()
+        d = per_site.setdefault(site, Counter())
+        d[stage] += 1
+        d["rows"] += 1
+        if flag == "PASS":
+            d["pass"] += 1
+        stages[stage] += 1
+
+    print(f"rounds completed: {rounds}")
+    print(f"\nVERIFIED WRITES: {proven}/{denom}"
+          f" = {round(100*proven/denom) if denom else 0}%   (criterion 2 wants >=95%)")
+    print(f"FALSE SUCCESS CLAIMS: {liars}   (criterion 3 hard gate, must be 0)")
+    print("\nper site (a row is one full post+audit+delete+audit round trip):")
+    print(f"  {'site':<10}{'rows':>5}{'pass':>6}  stages")
+    for site, d in sorted(per_site.items()):
+        st = {k: v for k, v in d.items() if k not in ("rows", "pass")}
+        print(f"  {site:<10}{d['rows']:>5}{d['pass']:>6}  {st}")
+
+    stranded = []
+    for m in STRANDED.finditer(text):
+        stranded.extend(x.strip() for x in m.group(1).split(","))
+    print(f"\nmarkers the canary could not clean up: {len(stranded)}")
+    for s in stranded:
+        print(f"  {s}")
+    if stranded:
+        print("  ^ verify by hand with bench/rawbrowser.py before calling this run finished")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/e2e/browser-v3/c7_run.sh b/e2e/browser-v3/c7_run.sh
new file mode 100755
index 00000000..57432a77
--- /dev/null
+++ b/e2e/browser-v3/c7_run.sh
@@ -0,0 +1,24 @@
+#!/bin/bash
+# Criteria 1, 4, 5, 6, 7 and 9 in one pass: the known suite at N=12 = 108 site-runs, which is the
+# >=100 sample criterion 7 asks for, while the same trials carry the reach, timing and skill data.
+#
+# Dry run only. The backend must be up with OSW_SENDSCRIPT_DRYRUN=1 (`stack.sh up dry`), so the
+# irreversible click is refused in the backend, not by a flag this script sets and does not own.
+# Where logs, profiles and run output go. Defaults to runs/ beside this harness; override with
+# OSW_BENCH_DIR to keep multi-gigabyte browser profiles off the repo disk.
+SP="${OSW_BENCH_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/runs}"
+mkdir -p "$SP"
+# The harness itself lives beside this script; SP is only for run OUTPUT.
+HARNESS="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# Repo root from this script's own location, so the harness works in any checkout.
+TREE="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+cd "$TREE" || exit 1
+
+N="${N:-12}"
+OUT="$SP/c7_r6.txt"
+: > "$OUT"
+echo "started $(date +%H:%M:%S), N=$N, log=$SP/r6_be.log" >> "$OUT"
+OSW_BASE=http://127.0.0.1:8326 OSW_LOG="$SP/r6_be.log" \
+  ./backend/.venv/bin/python "$HARNESS/bench.py" known "$N" >> "$OUT" 2>&1
+echo "########## DONE $(date +%H:%M:%S)" >> "$OUT"
+echo "backend restarts during the sweep: $(grep -c 'BACKEND RESTARTED' "$SP/r6_be.log")" >> "$OUT"
diff --git a/e2e/browser-v3/c8_run.sh b/e2e/browser-v3/c8_run.sh
new file mode 100755
index 00000000..6ac26847
--- /dev/null
+++ b/e2e/browser-v3/c8_run.sh
@@ -0,0 +1,18 @@
+#!/bin/bash
+# Criterion 8: the frozen holdout, re-measured after the round's changes. Dry run, same stack.
+# Where logs, profiles and run output go. Defaults to runs/ beside this harness; override with
+# OSW_BENCH_DIR to keep multi-gigabyte browser profiles off the repo disk.
+SP="${OSW_BENCH_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/runs}"
+mkdir -p "$SP"
+# The harness itself lives beside this script; SP is only for run OUTPUT.
+HARNESS="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# Repo root from this script's own location, so the harness works in any checkout.
+TREE="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+cd "$TREE" || exit 1
+N="${N:-2}"
+OUT="$SP/c8_r6.txt"
+: > "$OUT"
+echo "started $(date +%H:%M:%S), N=$N" >> "$OUT"
+OSW_BASE=http://127.0.0.1:8326 OSW_LOG="$SP/r6_be.log" \
+  ./backend/.venv/bin/python "$HARNESS/bench.py" holdout "$N" >> "$OUT" 2>&1
+echo "########## DONE $(date +%H:%M:%S)" >> "$OUT"
diff --git a/e2e/browser-v3/coverage.py b/e2e/browser-v3/coverage.py
new file mode 100644
index 00000000..c4d87615
--- /dev/null
+++ b/e2e/browser-v3/coverage.py
@@ -0,0 +1,289 @@
+"""Write-path coverage across sites with a GENUINE session, measured in dry run.
+
+Reports the two halves separately, because they fail for different reasons and only one of them was
+ever instrumented: did we REACH a composer and commit a fill, and can we RESOLVE that composer's
+submit. The second half is new. Without it every sweep said "ready to send" about sends that would
+have blind-tapped a coordinate (reddit, live).
+
+Never posts: the backend runs with OSW_SENDSCRIPT_DRYRUN=1, which stops at the irreversible click.
+"""
+import json, os, re, sys, time, urllib.request
+
+# Repo root, derived from this file (e2e/browser-v3/x.py -> two levels up). Hardcoding an
+# absolute path made this harness silently useless on any other checkout.
+ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+BASE = os.environ.get("OSW_BASE", "http://127.0.0.1:8326") + "/api/agents"
+LOG  = os.environ.get("OSW_LOG", "")
+MODEL = "opus-4-8"
+
+# This list was captioned "only sites this profile is genuinely signed into" and that was not true:
+# gmail and substack hit a login wall on every single sweep, and their rows were scored as composer
+# misses, which quietly charged our code for an account state. Sign-in is re-derived per run now,
+# from the page's own evidence, and an unmeasurable site leaves the denominator instead of losing.
+TASKS = {
+  "x":         'Go to x.com and post this tweet, exactly: "coverage probe alpha"',
+  "linkedin":  'Go to linkedin.com and create a post with exactly this text: "coverage probe alpha"',
+  "reddit":    'Go to reddit.com/r/test/submit and create a text post with body "coverage probe alpha"',
+  "gmail":     'Go to gmail and start a new email with the body "coverage probe alpha"',
+  "youtube":   'Go to youtube.com, open the first video, and write the comment "coverage probe alpha"',
+  "instagram": 'Go to instagram.com and write a comment "coverage probe alpha" on the first post',
+  "tiktok":    'Go to tiktok.com and write a comment "coverage probe alpha" on the first video',
+  "substack":  'Go to substack.com and start a new note with the text "coverage probe alpha"',
+  "twitch":    'Go to twitch.tv, open the first live channel, and write "coverage probe alpha" in chat',
+}
+
+def req(method, url, body=None):
+    tok = open(os.path.join(ROOT, "backend/data/auth.token")).read().strip()
+    data = json.dumps(body).encode() if body is not None else None
+    r = urllib.request.Request(url, data=data, method=method,
+        headers={"Content-Type": "application/json", "Authorization": "Bearer " + tok})
+    with urllib.request.urlopen(r, timeout=300) as resp:
+        return json.loads(resp.read().decode() or "{}")
+
+def loglines():
+    try: return sum(1 for _ in open(LOG, errors="ignore"))
+    except OSError: return 0
+
+BACKEND_DOWN = "[harness] BACKEND UNREACHABLE"
+
+
+def run(site, prompt, budget=200):
+    mark = loglines()
+    # A dead backend used to raise URLError straight out of here and kill the whole sweep mid-run,
+    # losing every trial after it. It is also not a product failure and must never be graded as one,
+    # so it gets a marker the classifier can bucket as infrastructure.
+    try:
+        dash = req("GET", BASE.replace("/agents","") + "/dashboards/list")
+        ds = dash if isinstance(dash, list) else dash.get("dashboards", [])
+        sid = req("POST", f"{BASE}/launch", {"mode":"agent","model":MODEL,"provider":"anthropic",
+                  "dashboard_id": ds[0]["id"], "name": f"cov-{site}"})["session"]["id"]
+    except Exception as e:
+        time.sleep(20)   # give the supervisor a chance to bring it back before the next trial
+        return f"{BACKEND_DOWN} ({type(e).__name__}: {str(e)[:80]})", 0.0
+    t0 = time.time()
+    try: req("POST", f"{BASE}/sessions/{sid}/message", {"prompt":prompt,"mode":"agent","model":MODEL})
+    except Exception: pass
+    while time.time() - t0 < budget:
+        try:
+            s = req("GET", f"{BASE}/sessions/{sid}")
+            if str(s.get("status") or "") in ("completed","error","stopped"): break
+        except Exception: pass
+        time.sleep(2)
+    time.sleep(1.5)
+    try: sl = "".join(open(LOG, errors="ignore").readlines()[mark:])
+    except OSError: sl = ""
+    # Tear the trial down before the next one starts. Each run leaves a chat session and a live
+    # browser card with its own webview; across 45 runs that is 45 webviews the renderer keeps
+    # compositing, and a starved renderer stops answering, which reads as a product failure. It
+    # already cost one whole sweep (24 of 45 rows came back "no dashboard" and scored a meaningless
+    # 29%). delete_session stops the browser-agent children first, so this reaps the cards too.
+    # Read the log slice BEFORE deleting, or the teardown's own lines land in the graded window.
+    try: req("DELETE", f"{BASE}/sessions/{sid}")
+    except Exception: pass
+    return sl, round(time.time()-t0, 1)
+
+DRY = re.compile(r"DRYRUN: WOULD send \(fill committed, send_button_listed=(\w+), "
+                 r"submit_resolved=(\w+), submit_rank=(\d+), submit=(.+?)\); not clicking")
+DECL = re.compile(r"\[browser-sendscript\] decline: (.+)")
+DISABLED = re.compile(r"submit (.+?) is present but DISABLED")
+FILL = re.compile(r"\[browser-sendscript\] fill target (.+?) \[(-?\d+)\] on (\S*)")
+# The composer was found AND focused and the text still would not go in. The old grader printed
+# "no send-script activity" for this, which reads as "nothing happened", and it hid the twitch and
+# instagram failures for as long as they existed. A dead fill is a code gap, and it gets a name.
+FILLERR = re.compile(r"fill errored \((.+?)\); handing")
+# Signed out, judged on the page's OWN evidence rather than on the product's label: a sign-in URL
+# or a visible password field. Grading this with the product's verdict would be grading the guard
+# with the guard, which is the mistake that made every earlier number worthless.
+WALL = re.compile(r"decline: login/auth wall \('([^']*)'\)(?: triggered by (url|password field))?")
+SIGNIN_URL = re.compile(r"(accounts\.google\.com|/login|/signin|/sign[-_]?in|/i/flow/login|/auth/)", re.I)
+# Matched on the page's own words, not on the product's verdict, same rule as everything else here.
+CAPTCHA = re.compile(
+    r"enter (?:the )?(?:code|characters) you hear|play the audio|press (?:and|&) hold|"
+    r"drag (?:the )?(?:slider|puzzle)|i'?m not a robot|verify you are (?:a )?human|"
+    r"select all (?:images|squares) (?:with|containing)|recaptcha|hcaptcha|turnstile|"
+    r"bot-detection challenge",
+    re.I)
+SAYS_OUT = "decline: signed OUT"
+
+# The page each task must end up on, written out by hand from what these sites actually use. This
+# is the half the harness was missing: it scored "a composer got filled" and never asked whose.
+# The instagram DM that started all this was filled on instagram.com/<someone>/, a PROFILE, while
+# the task said "the first post". The URL alone rejects it, and unlike a name check it still works
+# when the structural finder reports its box as a bare 'contenteditable'.
+ON_PAGE = {
+  "x":         r"x\.com/(home|compose)",
+  "linkedin":  r"linkedin\.com/(feed|posts)",
+  "reddit":    r"reddit\.com/r/[^/]+/submit",
+  "gmail":     r"mail\.google\.com/",
+  "youtube":   r"youtube\.com/watch",
+  "instagram": r"instagram\.com/(p|reel)/",
+  "tiktok":    r"tiktok\.com/@[^/]+/video/",
+  "substack":  r"substack\.com/",
+  "twitch":    r"twitch\.tv/[^/?]+(/|\?|$)",
+}
+# Graded independently of the product's own surface_mismatch on purpose: grading the guard with the
+# guard proves nothing. These are the names a DM box carries on the sites above.
+DM_NAME = re.compile(r"\b(message|messages|dm)\b", re.I)
+PUBLIC = ("x", "linkedin", "reddit", "youtube", "instagram", "tiktok", "substack")
+
+# A dry run can never actually send, so the completion gate correctly calls EVERY run a ghost
+# ("declared done but the send was not confirmed") and the fast path fires one recovery dispatch.
+# That recovery reuses the same card, wherever it has drifted to, and fills whatever it finds. Those
+# fills are an artifact of measuring in dry run and say nothing about coverage. Scoring them made
+# reddit and youtube look systematically broken (3/3 and 2/2 "wrong page") when both had already hit
+# the right composer seconds earlier: reddit on /r/test/submit, youtube on /watch.
+RECOVERY = "one recovery dispatch"
+
+def gradeable(slice_):
+    """The part of the run that happened before dry-run recovery muddied it."""
+    cut = slice_.find(RECOVERY)
+    return slice_[:cut] if cut > 0 else slice_
+
+def last(rx, slice_):
+    """A run can legitimately fill more than once (reddit tries /r/test/submit, then bare /submit),
+    so the LAST attempt inside the gradeable window owns the verdict; taking the first scored a page
+    the run had already left."""
+    ms = list(rx.finditer(gradeable(slice_)))
+    return ms[-1] if ms else None
+
+def surface(site, slice_):
+    """(ok, description) for the box that actually got filled. ok=False means we reached SOMETHING,
+    which the old harness counted as a win, and it was the wrong thing."""
+    f = last(FILL, slice_)
+    if not f:
+        return None, ""
+    name, url = f.group(1), f.group(3)
+    if site in PUBLIC and DM_NAME.search(name):
+        return False, f"DM box {name} on {url[:44]}"
+    want = ON_PAGE.get(site)
+    if want and not re.search(want, url):
+        return False, f"wrong page {url[:52]}"
+    return True, f"{name} on {url[:40]}"
+
+# A run whose LLM lane was flapping underneath it measured the machine, not the write path. Every
+# one of these was live in the 2026-08-01 sweep, which scored 1/9 while nothing about the code had
+# changed: eight other OpenSwarm backends were on the box, each one's dev logic killing the shared
+# 9router and starting its own on :20128. Wall times went 30s -> 202s and prestage got skipped.
+# "prestage] skipped" used to be in this list bare, and that was too wide: prestage skips for CODE
+# reasons too, and a live run on twitch died with "'utf-8' codec can't encode '\ud83e'" (half an
+# emoji) which this harness then filed as an environment problem and told me to re-run. A real bug
+# wearing an INVALID badge is worse than no harness. Only provider-shaped reasons count as sick.
+UNHEALTHY = ("9Router watchdog", "9Router process died", "No AI provider connected",
+             "prestage] skipped (No AI provider", "prestage] skipped (classifier",
+             "dispatch refused: no dashboard")
+
+def row(**kw):
+    base = {"composer": False, "reached": False, "submit": False, "rank": 0,
+            "invalid": False, "unmeasurable": "", "detail": ""}
+    base.update(kw)
+    return base
+
+def unmeasurable(slice_):
+    """Why this run could not measure our code at all, or '' if it could.
+
+    Being signed out is an account state, not a coverage failure, and charging our code for it is
+    how gmail and substack sat in the miss column across five sweeps. Read the page's own evidence
+    (a sign-in URL, a visible password field), and when the only thing available is the agent's own
+    say-so, return it labelled UNVERIFIED so it can never be quietly banked as a clean excuse.
+    """
+    # A bot-detection challenge is not a coverage failure. Solving one is off-limits, so a page
+    # behind a captcha is a page our code is not allowed to reach, and scoring it against reach
+    # charges us for a rule we are choosing to keep. Eric caught this on tiktok: an audio captcha
+    # over a logged-out feed, while the row read "no composer" as though the finder had missed.
+    cap = CAPTCHA.search(slice_)
+    if cap:
+        return f"bot-detection challenge, PROVEN ({cap.group(0)[:44]})"
+    w = WALL.search(slice_)
+    if w:
+        url, trigger = w.group(1), w.group(2) or ""
+        if trigger == "password field":
+            return f"signed out, PROVEN (password field on {url[:38]})"
+        if SIGNIN_URL.search(url):
+            return f"signed out, PROVEN (sign-in URL {url[:42]})"
+        return f"login wall claimed, UNVERIFIED (no evidence for {url[:34]})"
+    if SAYS_OUT in slice_:
+        return "signed out per the agent, UNVERIFIED (no wall evidence)"
+    return ""
+
+def verdict(site, slice_):
+    sick = [s for s in UNHEALTHY if s in slice_]
+    if sick:
+        return row(invalid=True, detail=f"INVALID: stack was unhealthy ({sick[0]})")
+    ok, where = surface(site, slice_)
+    reached = ok is not None
+    # The harness prompts a PARENT agent, which rewrites the task for the browser sub-agent. When
+    # that rewrite happens to quote two different strings the send script honestly declines, and
+    # that is the harness's phrasing failing, not this site's coverage. Scoring it as a miss put
+    # x at "n" on a run where x was fine 4 minutes earlier. Call it INVALID and re-run the site.
+    if not reached and "no unambiguous quoted payload" in slice_:
+        return row(invalid=True, detail="INVALID: parent rephrased the task, payload came through ambiguous")
+    why = unmeasurable(slice_)
+    if why and not reached:
+        return row(unmeasurable=why, detail=f"NOT MEASURABLE: {why}")
+    if ok is False:
+        return row(reached=True, detail=f"WRONG SURFACE: {where}")
+    m = last(DRY, slice_)
+    d = DISABLED.search(slice_)
+    if m or d:
+        submit = bool(m) and m.group(2) == "True"
+        rank = int(m.group(3)) if m else 0
+        detail = (m.group(4)[:34] if m else f"DISABLED {d.group(1)[:24]}")
+        return row(composer=True, reached=reached, submit=submit, rank=rank,
+                   detail=f"{detail} <- {where}")
+    # Reached the right box, focused it, and the text would not go in. Its own column: this is the
+    # one failure mode a better composer finder cannot fix, and it was invisible until now.
+    fe = last(FILLERR, slice_)
+    if fe:
+        return row(reached=True, detail=f"FILL DIED on {where}: {fe.group(1)[:46]}")
+    c = DECL.findall(slice_)
+    return row(reached=reached,
+               detail=("decline: " + c[-1][:46]) if c else "no send-script activity")
+
+def preflight():
+    """Refuse to measure on a box that cannot hold still. One 9router serves :20128 and every dev
+    backend's startup kills the one it does not own, so a second stack anywhere on the machine turns
+    this sweep into a coin flip. Better to print why than to publish a number that means nothing."""
+    import subprocess
+    # Count the real interpreters only. `pgrep -f "uvicorn backend.main"` also matches the
+    # supervising shell (its command line quotes the whole uvicorn line), so with a supervisor in
+    # place this refused to run every single time, on a box holding exactly one backend.
+    ps = subprocess.run(["ps", "-Ao", "command"], capture_output=True, text=True).stdout
+    n = sum(1 for ln in ps.splitlines()
+            if "-m uvicorn backend.main" in ln and "/bin/python" in ln)
+    router = urllib.request.urlopen("http://127.0.0.1:20128/v1/models", timeout=5).status
+    if n > 1 or router != 200:
+        print(f"REFUSING: {n} backends on this box (want 1), 9router HTTP {router} (want 200).")
+        print("Another session's stack is fighting yours for the router. Wait it out.")
+        sys.exit(2)
+
+if __name__ == "__main__":
+    if os.environ.get("OSW_SKIP_PREFLIGHT") != "1":
+        preflight()
+    only = sys.argv[1:] or list(TASKS)
+    rows = []
+    for site in only:
+        if site not in TASKS: continue
+        sl, wall = run(site, TASKS[site])
+        v = verdict(site, sl); v["site"] = site; v["wall"] = wall
+        rows.append(v)
+        print(f"  {site:11s} composer={'Y' if v['composer'] else 'n'} "
+              f"submit={'Y' if v['submit'] else 'n'} rank={v['rank']} {wall:6.1f}s  {v['detail']}", flush=True)
+    invalid = [r for r in rows if r["invalid"]]
+    unmeas = [r for r in rows if r["unmeasurable"] and not r["invalid"]]
+    graded = [r for r in rows if not r["invalid"] and not r["unmeasurable"]]
+    n = len(graded)
+    wrong = sum("WRONG SURFACE" in r["detail"] for r in graded)
+    died = sum("FILL DIED" in r["detail"] for r in graded)
+    print(f"\n=== RIGHT composer {sum(r['composer'] for r in graded)}/{n} measurable | "
+          f"submit resolved {sum(r['submit'] for r in graded)}/{n} | "
+          f"WRONG surface {wrong} | fill DIED {died} ===")
+    # Printed, never folded in. An excluded row is a claim that our code was not on trial, and that
+    # claim has to survive being read out loud, especially the ones resting on the agent's own word.
+    for label, group in (("NOT MEASURABLE", unmeas), ("INVALID, re-run", invalid)):
+        for r in group:
+            print(f"  {label}: {r['site']} ({r['unmeasurable'] or r['detail']})")
+    shaky = [r["site"] for r in unmeas if "UNVERIFIED" in r["unmeasurable"]]
+    if shaky:
+        print(f"  ^ {', '.join(shaky)} rest on the agent's own word. Confirm by hand before quoting "
+              f"any number that leaves them out.")
+    json.dump(rows, open(os.environ.get("OSW_OUT","/tmp/cov.json"), "w"), indent=1)
diff --git a/e2e/browser-v3/keep_renderer.sh b/e2e/browser-v3/keep_renderer.sh
new file mode 100755
index 00000000..e0555828
--- /dev/null
+++ b/e2e/browser-v3/keep_renderer.sh
@@ -0,0 +1,23 @@
+#!/bin/bash
+# Keep an Electron renderer alive for the length of a sweep.
+#
+# A 45-run sweep takes over an hour and the app does not reliably survive it: one run quit partway
+# through and 24 of 45 rows came back "dispatch refused: no dashboard", which scored as a 29% reach
+# that measured nothing but my own dead window. Relaunching by hand between runs is not a fix, it
+# just moves the gap to whenever I am not looking.
+#
+# Restarts on exit, and waits for webpack first: Electron launched against a dead :3026 loads a
+# blank page and quits immediately, which is how the window went missing the first time.
+# Where logs, profiles and run output go. Defaults to runs/ beside this harness; override with
+# OSW_BENCH_DIR to keep multi-gigabyte browser profiles off the repo disk.
+SP="${OSW_BENCH_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/runs}"
+mkdir -p "$SP"
+cd "$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)/electron" || exit 1
+
+while true; do
+  until curl -s -o /dev/null --max-time 4 "http://localhost:${OPENSWARM_DEV_PORT:-3026}"; do sleep 5; done
+  ELECTRON_DEV=1 OPENSWARM_DEV_PORT="${OPENSWARM_DEV_PORT:-3026}" OPENSWARM_PORT="${OPENSWARM_PORT:-8326}" \
+    ./node_modules/.bin/electron . --user-data-dir="$SP/udd" >> "$SP/renderer.log" 2>&1
+  echo "[keep_renderer] electron exited at $(date +%H:%M:%S), restarting" >> "$SP/renderer.log"
+  sleep 5
+done
diff --git a/e2e/browser-v3/probe_evidence.py b/e2e/browser-v3/probe_evidence.py
new file mode 100644
index 00000000..81c67e6d
--- /dev/null
+++ b/e2e/browser-v3/probe_evidence.py
@@ -0,0 +1,110 @@
+"""Where does a browser run's page text actually SURFACE? Answer it before trusting any audit.
+
+The canary proves a write landed by looking for the marker string in evidence the model did not
+author. It has been looking in the backend LOG, and the backend never logs page text: a grep for
+every canary marker ever generated across every backend log on this box returns nothing. So the
+audit could only ever return "could not look", which is exactly what LinkedIn and reddit scored.
+
+This probe runs one read-only task and dumps every place the marker could live (parent session
+messages, child sessions, tool results), so the replacement evidence channel is chosen on evidence
+rather than on where I assume the text goes.
+"""
+
+import json
+import os
+import sys
+import time
+import urllib.request
+
+# Repo root, derived from this file (e2e/browser-v3/x.py -> two levels up). Hardcoding an
+# absolute path made this harness silently useless on any other checkout.
+ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+BASE = os.environ.get("OSW_BASE", "http://127.0.0.1:8326") + "/api/agents"
+MODEL = "opus-4-8"
+
+
+def req(method: str, url: str, body=None):
+    tok = open(os.path.join(ROOT, "backend/data/auth.token")).read().strip()
+    data = json.dumps(body).encode() if body is not None else None
+    r = urllib.request.Request(url, data=data, method=method,
+                               headers={"Content-Type": "application/json",
+                                        "Authorization": "Bearer " + tok})
+    with urllib.request.urlopen(r, timeout=260) as resp:
+        return json.loads(resp.read().decode() or "{}")
+
+
+def walk(node, path="", out=None, needle=""):
+    """Every JSON path whose leaf string contains `needle`."""
+    out = [] if out is None else out
+    if isinstance(node, dict):
+        for k, v in node.items():
+            walk(v, f"{path}.{k}", out, needle)
+    elif isinstance(node, list):
+        for i, v in enumerate(node):
+            walk(v, f"{path}[{i}]", out, needle)
+    elif isinstance(node, str) and needle and needle in node:
+        out.append((path, node[:160]))
+    return out
+
+
+def main() -> int:
+    task = sys.argv[1]
+    needle = sys.argv[2]
+    dash = req("GET", BASE.replace("/agents", "") + "/dashboards/list")
+    ds = dash if isinstance(dash, list) else dash.get("dashboards", [])
+    sid = req("POST", f"{BASE}/launch", {"mode": "agent", "model": MODEL, "provider": "anthropic",
+                                         "dashboard_id": ds[0]["id"], "name": "probe-evidence"})["session"]["id"]
+    print(f"session {sid}\ntask   {task}\nneedle {needle}\n", flush=True)
+    t0 = time.time()
+    try:
+        req("POST", f"{BASE}/sessions/{sid}/message", {"prompt": task, "mode": "agent", "model": MODEL})
+    except Exception:
+        pass
+    status = ""
+    while time.time() - t0 < 240:
+        try:
+            s = req("GET", f"{BASE}/sessions/{sid}")
+            status = str(s.get("status") or "")
+            if status in ("completed", "error", "stopped"):
+                break
+        except Exception:
+            pass
+        time.sleep(2)
+    time.sleep(2)
+
+    s = req("GET", f"{BASE}/sessions/{sid}")
+    print(f"status={status} wall={round(time.time()-t0,1)}s")
+    print(f"parent top-level keys: {sorted(s.keys())}")
+    print(f"parent messages: {len(s.get('messages') or [])}")
+    hits = walk(s, "session", needle=needle)
+    print(f"\n--- marker in PARENT session json: {len(hits)} hits ---")
+    for p, v in hits[:14]:
+        print(f"  {p}\n      {v!r}")
+
+    # Child sessions: the sub-agent path puts the browser work in its own session.
+    kids = []
+    try:
+        allsess = req("GET", f"{BASE}/sessions")
+        rows = allsess if isinstance(allsess, list) else allsess.get("sessions", [])
+        kids = [r for r in rows if str(r.get("parent_session_id") or "") == sid]
+    except Exception as e:
+        print(f"(child enumeration failed: {e})")
+    print(f"\n--- child sessions: {len(kids)} ---")
+    for k in kids:
+        full = req("GET", f"{BASE}/sessions/{k['id']}")
+        kh = walk(full, "child", needle=needle)
+        print(f"  {k['id']} name={k.get('name')!r} msgs={len(full.get('messages') or [])} marker_hits={len(kh)}")
+        for p, v in kh[:8]:
+            print(f"      {p}\n          {v!r}")
+
+    json.dump(s, open(os.environ.get("OSW_DUMP", "/tmp/probe_session.json"), "w"), indent=1)
+    print(f"\nfull parent dump -> {os.environ.get('OSW_DUMP', '/tmp/probe_session.json')}")
+    try:
+        req("DELETE", f"{BASE}/sessions/{sid}")
+    except Exception:
+        pass
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/e2e/browser-v3/rawbrowser.py b/e2e/browser-v3/rawbrowser.py
new file mode 100644
index 00000000..2eb3ce27
--- /dev/null
+++ b/e2e/browser-v3/rawbrowser.py
@@ -0,0 +1,98 @@
+"""Drive a browser card directly, with no model anywhere in the loop.
+
+The audit channel the canary now uses, exposed as a CLI so I can check a real account by hand
+without hand-escaping JS through three layers of shell quoting (which silently sent a mangled
+expression once and made a working navigate look like a domain-gate refusal).
+
+    rawbrowser.py cards
+    rawbrowser.py read  <url> [substring]
+    rawbrowser.py eval  <url> <js>
+"""
+
+import json
+import os
+import sys
+import time
+import urllib.request
+from urllib.parse import urlparse
+
+# Repo root, derived from this file (e2e/browser-v3/x.py -> two levels up). Hardcoding an
+# absolute path made this harness silently useless on any other checkout.
+ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+API = os.environ.get("OSW_BASE", "http://127.0.0.1:8326") + "/api"
+
+
+def req(method, url, body=None):
+    tok = open(os.path.join(ROOT, "backend/data/auth.token")).read().strip()
+    data = json.dumps(body).encode() if body is not None else None
+    r = urllib.request.Request(url, data=data, method=method,
+                               headers={"Content-Type": "application/json",
+                                        "Authorization": "Bearer " + tok})
+    with urllib.request.urlopen(r, timeout=180) as resp:
+        return json.loads(resp.read().decode() or "{}")
+
+
+def cards():
+    ds = req("GET", f"{API}/dashboards/list")
+    ds = ds if isinstance(ds, list) else ds.get("dashboards", [])
+    if not ds:
+        return []
+    full = req("GET", f"{API}/dashboards/{ds[0]['id']}")
+    return list(((full.get("layout") or {}).get("browser_cards") or {}))
+
+
+def cmd(bid, action, params):
+    return req("POST", f"{API}/browser/command",
+               {"action": action, "browser_id": bid, "params": params})
+
+
+BODY = ('(()=>{try{return {body:(document.body&&document.body.innerText)||"",'
+        'u:location.href};}catch(e){return {body:"",u:""};}})()')
+
+
+def main():
+    what = sys.argv[1] if len(sys.argv) > 1 else "cards"
+    cs = cards()
+    if what == "cards":
+        print("\n".join(cs) or "(none)")
+        return 0
+    if not cs:
+        print("no browser card on the dashboard; run a browser task first")
+        return 2
+    url = sys.argv[2]
+    want = (urlparse(url).hostname or "").lower().replace("www.", "")
+    # A card whose session was torn down mid-run WEDGES: it answers navigate with "Navigated to
+    # <url>" in ~80ms and stays where it was. So never trust the reply, check where we landed, and
+    # move to another card if this one did not go.
+    bid = ""
+    for cand in ([os.environ["OSW_CARD"]] if os.environ.get("OSW_CARD") else list(reversed(cs))):
+        cmd(cand, "navigate", {"url": url})
+        time.sleep(float(os.environ.get("OSW_SETTLE", "5")))
+        got = cmd(cand, "evaluate", {"expression": BODY})
+        try:
+            landed = json.loads(got.get("text") or "{}").get("u") or ""
+        except ValueError:
+            landed = ""
+        if (urlparse(landed).hostname or "").lower().replace("www.", "") == want:
+            bid = cand
+            break
+        print(f"[card {cand} wedged, landed on {landed[:60]!r}]", file=sys.stderr)
+    if not bid:
+        print(f"no card could reach {url}", file=sys.stderr)
+        return 3
+    if what == "eval":
+        print(json.dumps(cmd(bid, "evaluate", {"expression": sys.argv[3]}), indent=1))
+        return 0
+    r = cmd(bid, "evaluate", {"expression": BODY})
+    v = json.loads(r.get("text") or "{}")
+    body = str(v.get("body") or "")
+    print(f"[url] {v.get('u')}\n[chars] {len(body)}")
+    if len(sys.argv) > 3:
+        needle = sys.argv[3]
+        print(f"[{needle!r} present] {needle in body}")
+    print(body[:3000])
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/e2e/browser-v3/results/holdout_raw.jsonl b/e2e/browser-v3/results/holdout_raw.jsonl
new file mode 100644
index 00000000..ff18413e
--- /dev/null
+++ b/e2e/browser-v3/results/holdout_raw.jsonl
@@ -0,0 +1,20 @@
+{"site": "rentry", "trial": 1, "bucket": "ok", "wall_s": 38.7, "detail": "'go' <- 'contenteditable' on https://rentry.co/", "composer": true, "submit": true, "total_ms": 5166, "llm_ms": 0, "tools_ms": 2072, "other_ms": 3094, "turns": 1, "artifact": "results/rentry_1.log"}
+{"site": "dpaste", "trial": 1, "bucket": "infra_browser", "wall_s": 71.2, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 35756, "llm_ms": 0, "tools_ms": 31997, "other_ms": 3759, "turns": 1, "artifact": "results/dpaste_1.log"}
+{"site": "controlc", "trial": 1, "bucket": "ok", "wall_s": 34.7, "detail": "'create paste' <- 'textarea' on https://controlc.com/", "composer": true, "submit": true, "total_ms": 6413, "llm_ms": 0, "tools_ms": 3344, "other_ms": 3069, "turns": 1, "artifact": "results/controlc_1.log"}
+{"site": "txti", "trial": 1, "bucket": "product_no_composer", "wall_s": 31.7, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 5828, "llm_ms": 0, "tools_ms": 2924, "other_ms": 2904, "turns": 1, "artifact": "results/txti_1.log"}
+{"site": "justpaste", "trial": 1, "bucket": "ok", "wall_s": 31.9, "detail": "'publish' <- 'input' on https://justpaste.it/", "composer": true, "submit": true, "total_ms": 6524, "llm_ms": 0, "tools_ms": 3456, "other_ms": 3068, "turns": 1, "artifact": "results/justpaste_1.log"}
+{"site": "telegraph", "trial": 1, "bucket": "ok", "wall_s": 29.8, "detail": "'publish' <- 'contenteditable' on https://telegra.ph/", "composer": true, "submit": true, "total_ms": 3812, "llm_ms": 0, "tools_ms": 1022, "other_ms": 2790, "turns": 1, "artifact": "results/telegraph_1.log"}
+{"site": "quill", "trial": 1, "bucket": "ok", "wall_s": 33.3, "detail": "'share your edits' <- 'contenteditable' on https://quilljs.com/playground/snow", "composer": true, "submit": true, "total_ms": 5320, "llm_ms": 0, "tools_ms": 2286, "other_ms": 3034, "turns": 1, "artifact": "results/quill_1.log"}
+{"site": "tinymce", "trial": 1, "bucket": "ok", "wall_s": 34.0, "detail": "'send' <- 'textarea' on https://www.tiny.cloud/", "composer": true, "submit": true, "total_ms": 8948, "llm_ms": 0, "tools_ms": 5838, "other_ms": 3110, "turns": 1, "artifact": "results/tinymce_1.log"}
+{"site": "ckeditor", "trial": 1, "bucket": "ok", "wall_s": 32.8, "detail": "'no submit control in the composer <- 'Rich Text Editor. Editing area: main. Press \u23250 for help.' on https://ckeditor.com/ckeditor-5/demo/fea", "composer": true, "submit": false, "total_ms": 2478, "llm_ms": 0, "tools_ms": 2024, "other_ms": 454, "turns": 1, "artifact": "results/ckeditor_1.log"}
+{"site": "codemirror", "trial": 1, "bucket": "not_measurable", "wall_s": 55.7, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 16007, "llm_ms": 0, "tools_ms": 10349, "other_ms": 5658, "turns": 1, "artifact": "results/codemirror_1.log"}
+{"site": "rentry", "trial": 2, "bucket": "ok", "wall_s": 28.4, "detail": "'go' <- 'contenteditable' on https://rentry.co/", "composer": true, "submit": true, "total_ms": 3808, "llm_ms": 0, "tools_ms": 964, "other_ms": 2844, "turns": 1, "artifact": "results/rentry_2.log"}
+{"site": "dpaste", "trial": 2, "bucket": "infra_browser", "wall_s": 68.0, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 37815, "llm_ms": 0, "tools_ms": 34733, "other_ms": 3082, "turns": 1, "artifact": "results/dpaste_2.log"}
+{"site": "controlc", "trial": 2, "bucket": "infra_router", "wall_s": 28.7, "detail": "INVALID: stack was unhealthy (No AI provider connected)", "composer": false, "submit": false, "total_ms": 7074, "llm_ms": 0, "tools_ms": 3392, "other_ms": 3682, "turns": 1, "artifact": "results/controlc_2.log"}
+{"site": "txti", "trial": 2, "bucket": "product_no_composer", "wall_s": 32.3, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 5158, "llm_ms": 0, "tools_ms": 2298, "other_ms": 2860, "turns": 1, "artifact": "results/txti_2.log"}
+{"site": "justpaste", "trial": 2, "bucket": "ok", "wall_s": 25.5, "detail": "'publish' <- 'input' on https://justpaste.it/", "composer": true, "submit": true, "total_ms": 5041, "llm_ms": 0, "tools_ms": 2148, "other_ms": 2893, "turns": 1, "artifact": "results/justpaste_2.log"}
+{"site": "telegraph", "trial": 2, "bucket": "ok", "wall_s": 30.3, "detail": "'publish' <- 'contenteditable' on https://telegra.ph/", "composer": true, "submit": true, "total_ms": 4126, "llm_ms": 0, "tools_ms": 1313, "other_ms": 2813, "turns": 1, "artifact": "results/telegraph_2.log"}
+{"site": "quill", "trial": 2, "bucket": "infra_router", "wall_s": 24.9, "detail": "INVALID: stack was unhealthy (No AI provider connected)", "composer": false, "submit": false, "total_ms": 4778, "llm_ms": 0, "tools_ms": 1923, "other_ms": 2855, "turns": 1, "artifact": "results/quill_2.log"}
+{"site": "tinymce", "trial": 2, "bucket": "ok", "wall_s": 64.7, "detail": "'send' <- 'textarea' on https://www.tiny.cloud/", "composer": true, "submit": true, "total_ms": 7947, "llm_ms": 0, "tools_ms": 4528, "other_ms": 3419, "turns": 1, "artifact": "results/tinymce_2.log"}
+{"site": "ckeditor", "trial": 2, "bucket": "ok", "wall_s": 20.4, "detail": "'no submit control in the composer <- 'Rich Text Editor. Editing area: main. Press \u23250 for help.' on https://ckeditor.com/ckeditor-5/demo/fea", "composer": true, "submit": false, "total_ms": 1746, "llm_ms": 0, "tools_ms": 1457, "other_ms": 289, "turns": 1, "artifact": "results/ckeditor_2.log"}
+{"site": "codemirror", "trial": 2, "bucket": "ok", "wall_s": 38.0, "detail": "'no submit control in the composer <- 'textarea' on https://codepen.io/pen", "composer": true, "submit": false, "total_ms": 4657, "llm_ms": 0, "tools_ms": 1817, "other_ms": 2840, "turns": 1, "artifact": "results/codemirror_2.log"}
diff --git a/e2e/browser-v3/results/known_raw.jsonl b/e2e/browser-v3/results/known_raw.jsonl
new file mode 100644
index 00000000..1142339b
--- /dev/null
+++ b/e2e/browser-v3/results/known_raw.jsonl
@@ -0,0 +1,158 @@
+{"site": "x", "trial": 1, "bucket": "ok", "wall_s": 11.7, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 726, "llm_ms": 0, "tools_ms": 680, "other_ms": 46, "turns": 1, "artifact": "results/x_1.log"}
+{"site": "linkedin", "trial": 1, "bucket": "ok", "wall_s": 20.5, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 726, "llm_ms": 0, "tools_ms": 680, "other_ms": 46, "turns": 1, "artifact": "results/linkedin_1.log"}
+{"site": "reddit", "trial": 1, "bucket": "ok", "wall_s": 7.7, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 333, "llm_ms": 0, "tools_ms": 313, "other_ms": 20, "turns": 1, "artifact": "results/reddit_1.log"}
+{"site": "gmail", "trial": 1, "bucket": "not_measurable", "wall_s": 11.4, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 125, "llm_ms": 0, "tools_ms": 107, "other_ms": 18, "turns": 1, "artifact": "results/gmail_1.log"}
+{"site": "youtube", "trial": 1, "bucket": "ok", "wall_s": 31.5, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=zdqCyJcT", "composer": true, "submit": true, "total_ms": 10519, "llm_ms": 0, "tools_ms": 7903, "other_ms": 2616, "turns": 1, "artifact": "results/youtube_1.log"}
+{"site": "instagram", "trial": 1, "bucket": "ok", "wall_s": 14.4, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 3005, "llm_ms": 0, "tools_ms": 391, "other_ms": 2614, "turns": 1, "artifact": "results/instagram_1.log"}
+{"site": "tiktok", "trial": 1, "bucket": "not_measurable", "wall_s": 34.8, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/)", "composer": false, "submit": false, "total_ms": 10336, "llm_ms": 0, "tools_ms": 3515, "other_ms": 6821, "turns": 1, "artifact": "results/tiktok_1.log"}
+{"site": "substack", "trial": 1, "bucket": "not_measurable", "wall_s": 13.5, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 202, "llm_ms": 0, "tools_ms": 193, "other_ms": 9, "turns": 1, "artifact": "results/substack_1.log"}
+{"site": "twitch", "trial": 1, "bucket": "ok", "wall_s": 12.2, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/cinna", "composer": true, "submit": true, "total_ms": 560, "llm_ms": 0, "tools_ms": 533, "other_ms": 27, "turns": 1, "artifact": "results/twitch_1.log"}
+{"site": "x", "trial": 2, "bucket": "ok", "wall_s": 10.6, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 490, "llm_ms": 0, "tools_ms": 470, "other_ms": 20, "turns": 1, "artifact": "results/x_2.log"}
+{"site": "linkedin", "trial": 2, "bucket": "ok", "wall_s": 20.6, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 387, "llm_ms": 0, "tools_ms": 352, "other_ms": 35, "turns": 1, "artifact": "results/linkedin_2.log"}
+{"site": "reddit", "trial": 2, "bucket": "ok", "wall_s": 8.0, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 327, "llm_ms": 0, "tools_ms": 319, "other_ms": 8, "turns": 1, "artifact": "results/reddit_2.log"}
+{"site": "gmail", "trial": 2, "bucket": "not_measurable", "wall_s": 11.6, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 384, "llm_ms": 0, "tools_ms": 376, "other_ms": 8, "turns": 1, "artifact": "results/gmail_2.log"}
+{"site": "youtube", "trial": 2, "bucket": "ok", "wall_s": 31.5, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=nlc41gME", "composer": true, "submit": true, "total_ms": 10192, "llm_ms": 0, "tools_ms": 7579, "other_ms": 2613, "turns": 1, "artifact": "results/youtube_2.log"}
+{"site": "instagram", "trial": 2, "bucket": "ok", "wall_s": 10.3, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 2975, "llm_ms": 0, "tools_ms": 360, "other_ms": 2615, "turns": 1, "artifact": "results/instagram_2.log"}
+{"site": "tiktok", "trial": 2, "bucket": "product_no_composer", "wall_s": 47.6, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 12896, "llm_ms": 0, "tools_ms": 7273, "other_ms": 5623, "turns": 1, "artifact": "results/tiktok_2.log"}
+{"site": "substack", "trial": 2, "bucket": "not_measurable", "wall_s": 15.6, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 210, "llm_ms": 0, "tools_ms": 201, "other_ms": 9, "turns": 1, "artifact": "results/substack_2.log"}
+{"site": "twitch", "trial": 2, "bucket": "ok", "wall_s": 11.6, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/cinna", "composer": true, "submit": true, "total_ms": 809, "llm_ms": 0, "tools_ms": 795, "other_ms": 14, "turns": 1, "artifact": "results/twitch_2.log"}
+{"site": "x", "trial": 3, "bucket": "ok", "wall_s": 10.3, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 562, "llm_ms": 0, "tools_ms": 553, "other_ms": 9, "turns": 1, "artifact": "results/x_3.log"}
+{"site": "linkedin", "trial": 3, "bucket": "ok", "wall_s": 18.5, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 361, "llm_ms": 0, "tools_ms": 351, "other_ms": 10, "turns": 1, "artifact": "results/linkedin_3.log"}
+{"site": "reddit", "trial": 3, "bucket": "ok", "wall_s": 7.6, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 361, "llm_ms": 0, "tools_ms": 351, "other_ms": 10, "turns": 1, "artifact": "results/reddit_3.log"}
+{"site": "gmail", "trial": 3, "bucket": "not_measurable", "wall_s": 11.6, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 120, "llm_ms": 0, "tools_ms": 113, "other_ms": 7, "turns": 1, "artifact": "results/gmail_3.log"}
+{"site": "youtube", "trial": 3, "bucket": "infra_router", "wall_s": 28.9, "detail": "INVALID: stack was unhealthy (No AI provider connected)", "composer": false, "submit": false, "total_ms": 11732, "llm_ms": 0, "tools_ms": 9073, "other_ms": 2659, "turns": 1, "artifact": "results/youtube_3.log"}
+{"site": "instagram", "trial": 3, "bucket": "infra_harness", "wall_s": 80.2, "detail": "INVALID: parent rephrased the task, payload came through ambiguous", "composer": false, "submit": false, "total_ms": 3309, "llm_ms": 0, "tools_ms": 920, "other_ms": 2389, "turns": 1, "artifact": "results/instagram_3.log"}
+{"site": "tiktok", "trial": 3, "bucket": "not_measurable", "wall_s": 54.3, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 13522, "llm_ms": 0, "tools_ms": 7841, "other_ms": 5681, "turns": 1, "artifact": "results/tiktok_3.log"}
+{"site": "substack", "trial": 3, "bucket": "not_measurable", "wall_s": 16.0, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 305, "llm_ms": 0, "tools_ms": 286, "other_ms": 19, "turns": 1, "artifact": "results/substack_3.log"}
+{"site": "twitch", "trial": 3, "bucket": "ok", "wall_s": 9.9, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/cinna", "composer": true, "submit": true, "total_ms": 399, "llm_ms": 0, "tools_ms": 384, "other_ms": 15, "turns": 1, "artifact": "results/twitch_3.log"}
+{"site": "x", "trial": 4, "bucket": "ok", "wall_s": 9.5, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 331, "llm_ms": 0, "tools_ms": 310, "other_ms": 21, "turns": 1, "artifact": "results/x_4.log"}
+{"site": "linkedin", "trial": 4, "bucket": "ok", "wall_s": 19.8, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 399, "llm_ms": 0, "tools_ms": 384, "other_ms": 15, "turns": 1, "artifact": "results/linkedin_4.log"}
+{"site": "reddit", "trial": 4, "bucket": "ok", "wall_s": 10.8, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 683, "llm_ms": 0, "tools_ms": 664, "other_ms": 19, "turns": 1, "artifact": "results/reddit_4.log"}
+{"site": "gmail", "trial": 4, "bucket": "not_measurable", "wall_s": 14.1, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 152, "llm_ms": 0, "tools_ms": 144, "other_ms": 8, "turns": 1, "artifact": "results/gmail_4.log"}
+{"site": "youtube", "trial": 4, "bucket": "ok", "wall_s": 37.5, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=nXWaTUBH", "composer": true, "submit": true, "total_ms": 10470, "llm_ms": 0, "tools_ms": 7842, "other_ms": 2628, "turns": 1, "artifact": "results/youtube_4.log"}
+{"site": "instagram", "trial": 4, "bucket": "ok", "wall_s": 12.7, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 1401, "llm_ms": 0, "tools_ms": 1388, "other_ms": 13, "turns": 1, "artifact": "results/instagram_4.log"}
+{"site": "tiktok", "trial": 4, "bucket": "product_no_composer", "wall_s": 35.6, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 13145, "llm_ms": 0, "tools_ms": 7519, "other_ms": 5626, "turns": 1, "artifact": "results/tiktok_4.log"}
+{"site": "substack", "trial": 4, "bucket": "not_measurable", "wall_s": 13.7, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 225, "llm_ms": 0, "tools_ms": 214, "other_ms": 11, "turns": 1, "artifact": "results/substack_4.log"}
+{"site": "twitch", "trial": 4, "bucket": "ok", "wall_s": 10.2, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/cinna", "composer": true, "submit": true, "total_ms": 392, "llm_ms": 0, "tools_ms": 376, "other_ms": 16, "turns": 1, "artifact": "results/twitch_4.log"}
+{"site": "x", "trial": 5, "bucket": "ok", "wall_s": 15.8, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 744, "llm_ms": 0, "tools_ms": 728, "other_ms": 16, "turns": 1, "artifact": "results/x_5.log"}
+{"site": "linkedin", "trial": 5, "bucket": "ok", "wall_s": 19.8, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 430, "llm_ms": 0, "tools_ms": 417, "other_ms": 13, "turns": 1, "artifact": "results/linkedin_5.log"}
+{"site": "reddit", "trial": 5, "bucket": "ok", "wall_s": 7.5, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 368, "llm_ms": 0, "tools_ms": 343, "other_ms": 25, "turns": 1, "artifact": "results/reddit_5.log"}
+{"site": "gmail", "trial": 5, "bucket": "not_measurable", "wall_s": 12.0, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 192, "llm_ms": 0, "tools_ms": 182, "other_ms": 10, "turns": 1, "artifact": "results/gmail_5.log"}
+{"site": "youtube", "trial": 5, "bucket": "ok", "wall_s": 38.2, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=3-EkbYNP", "composer": true, "submit": true, "total_ms": 11844, "llm_ms": 0, "tools_ms": 9154, "other_ms": 2690, "turns": 1, "artifact": "results/youtube_5.log"}
+{"site": "instagram", "trial": 5, "bucket": "ok", "wall_s": 12.7, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 1734, "llm_ms": 0, "tools_ms": 1668, "other_ms": 66, "turns": 1, "artifact": "results/instagram_5.log"}
+{"site": "tiktok", "trial": 5, "bucket": "not_measurable", "wall_s": 43.8, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 15671, "llm_ms": 0, "tools_ms": 8749, "other_ms": 6922, "turns": 1, "artifact": "results/tiktok_5.log"}
+{"site": "substack", "trial": 5, "bucket": "not_measurable", "wall_s": 13.6, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 229, "llm_ms": 0, "tools_ms": 220, "other_ms": 9, "turns": 1, "artifact": "results/substack_5.log"}
+{"site": "twitch", "trial": 5, "bucket": "ok", "wall_s": 11.7, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/cinna", "composer": true, "submit": true, "total_ms": 748, "llm_ms": 0, "tools_ms": 706, "other_ms": 42, "turns": 1, "artifact": "results/twitch_5.log"}
+{"site": "x", "trial": 6, "bucket": "ok", "wall_s": 10.9, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 591, "llm_ms": 0, "tools_ms": 567, "other_ms": 24, "turns": 1, "artifact": "results/x_6.log"}
+{"site": "linkedin", "trial": 6, "bucket": "ok", "wall_s": 19.6, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 385, "llm_ms": 0, "tools_ms": 366, "other_ms": 19, "turns": 1, "artifact": "results/linkedin_6.log"}
+{"site": "reddit", "trial": 6, "bucket": "ok", "wall_s": 7.4, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 385, "llm_ms": 0, "tools_ms": 366, "other_ms": 19, "turns": 1, "artifact": "results/reddit_6.log"}
+{"site": "gmail", "trial": 6, "bucket": "not_measurable", "wall_s": 11.7, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 529, "llm_ms": 0, "tools_ms": 520, "other_ms": 9, "turns": 1, "artifact": "results/gmail_6.log"}
+{"site": "youtube", "trial": 6, "bucket": "ok", "wall_s": 29.5, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=skI5KmOI", "composer": true, "submit": true, "total_ms": 10506, "llm_ms": 0, "tools_ms": 7888, "other_ms": 2618, "turns": 1, "artifact": "results/youtube_6.log"}
+{"site": "instagram", "trial": 6, "bucket": "ok", "wall_s": 13.2, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 3083, "llm_ms": 0, "tools_ms": 465, "other_ms": 2618, "turns": 1, "artifact": "results/instagram_6.log"}
+{"site": "tiktok", "trial": 6, "bucket": "not_measurable", "wall_s": 19.5, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@karlodecke)", "composer": false, "submit": false, "total_ms": 485, "llm_ms": 0, "tools_ms": 448, "other_ms": 37, "turns": 1, "artifact": "results/tiktok_6.log"}
+{"site": "substack", "trial": 6, "bucket": "not_measurable", "wall_s": 16.4, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 363, "llm_ms": 0, "tools_ms": 327, "other_ms": 36, "turns": 1, "artifact": "results/substack_6.log"}
+{"site": "twitch", "trial": 6, "bucket": "ok", "wall_s": 12.5, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/cinna", "composer": true, "submit": true, "total_ms": 850, "llm_ms": 0, "tools_ms": 832, "other_ms": 18, "turns": 1, "artifact": "results/twitch_6.log"}
+{"site": "x", "trial": 7, "bucket": "ok", "wall_s": 11.9, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 993, "llm_ms": 0, "tools_ms": 532, "other_ms": 461, "turns": 1, "artifact": "results/x_7.log"}
+{"site": "linkedin", "trial": 7, "bucket": "infra_harness", "wall_s": 18.5, "detail": "INVALID: stack was unhealthy (prestage] skipped (No AI provider)", "composer": false, "submit": false, "total_ms": 432, "llm_ms": 0, "tools_ms": 420, "other_ms": 12, "turns": 1, "artifact": "results/linkedin_7.log"}
+{"site": "reddit", "trial": 7, "bucket": "infra_router", "wall_s": 209.0, "detail": "INVALID: stack was unhealthy (9Router process died)", "composer": false, "submit": false, "total_ms": 411, "llm_ms": 0, "tools_ms": 398, "other_ms": 13, "turns": 1, "artifact": "results/reddit_7.log"}
+{"site": "gmail", "trial": 7, "bucket": "not_measurable", "wall_s": 12.0, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 161, "llm_ms": 0, "tools_ms": 150, "other_ms": 11, "turns": 1, "artifact": "results/gmail_7.log"}
+{"site": "youtube", "trial": 7, "bucket": "product_no_composer", "wall_s": 36.0, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 10626, "llm_ms": 0, "tools_ms": 7995, "other_ms": 2631, "turns": 1, "artifact": "results/youtube_7.log"}
+{"site": "instagram", "trial": 7, "bucket": "infra_router", "wall_s": 201.7, "detail": "INVALID: stack was unhealthy (9Router watchdog)", "composer": false, "submit": false, "total_ms": null, "llm_ms": null, "tools_ms": null, "other_ms": null, "turns": null, "artifact": "results/instagram_7.log"}
+{"site": "x", "trial": 1, "bucket": "ok", "wall_s": 20.6, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 4087, "llm_ms": 0, "tools_ms": 3442, "other_ms": 645, "turns": 1, "artifact": "results/x_1.log"}
+{"site": "linkedin", "trial": 1, "bucket": "ok", "wall_s": 20.3, "detail": "'post' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 1493, "llm_ms": 0, "tools_ms": 1388, "other_ms": 105, "turns": 1, "artifact": "results/linkedin_1.log"}
+{"site": "reddit", "trial": 1, "bucket": "ok", "wall_s": 8.1, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 705, "llm_ms": 0, "tools_ms": 695, "other_ms": 10, "turns": 1, "artifact": "results/reddit_1.log"}
+{"site": "gmail", "trial": 1, "bucket": "not_measurable", "wall_s": 13.7, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 220, "llm_ms": 0, "tools_ms": 206, "other_ms": 14, "turns": 1, "artifact": "results/gmail_1.log"}
+{"site": "youtube", "trial": 1, "bucket": "ok", "wall_s": 40.8, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=b2dhv1Pl", "composer": true, "submit": true, "total_ms": 12529, "llm_ms": 0, "tools_ms": 9817, "other_ms": 2712, "turns": 1, "artifact": "results/youtube_1.log"}
+{"site": "instagram", "trial": 1, "bucket": "ok", "wall_s": 19.0, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 2087, "llm_ms": 0, "tools_ms": 2048, "other_ms": 39, "turns": 1, "artifact": "results/instagram_1.log"}
+{"site": "tiktok", "trial": 1, "bucket": "not_measurable", "wall_s": 37.6, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@sams.extre)", "composer": false, "submit": false, "total_ms": 424, "llm_ms": 0, "tools_ms": 410, "other_ms": 14, "turns": 1, "artifact": "results/tiktok_1.log"}
+{"site": "substack", "trial": 1, "bucket": "not_measurable", "wall_s": 13.6, "detail": "NOT MEASURABLE: signed out, PROVEN (password field on https://substack.com/)", "composer": false, "submit": false, "total_ms": 317, "llm_ms": 0, "tools_ms": 307, "other_ms": 10, "turns": 1, "artifact": "results/substack_1.log"}
+{"site": "twitch", "trial": 1, "bucket": "product_wrong_surface", "wall_s": 18.1, "detail": "WRONG SURFACE: wrong page https://www.twitch.tv/", "composer": false, "submit": false, "total_ms": 6946, "llm_ms": 0, "tools_ms": 4325, "other_ms": 2621, "turns": 1, "artifact": "results/twitch_1.log"}
+{"site": "x", "trial": 2, "bucket": "ok", "wall_s": 11.6, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 3174, "llm_ms": 0, "tools_ms": 1642, "other_ms": 1532, "turns": 1, "artifact": "results/x_2.log"}
+{"site": "linkedin", "trial": 2, "bucket": "ok", "wall_s": 13.5, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 1211, "llm_ms": 0, "tools_ms": 1198, "other_ms": 13, "turns": 1, "artifact": "results/linkedin_2.log"}
+{"site": "reddit", "trial": 2, "bucket": "ok", "wall_s": 7.6, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 815, "llm_ms": 0, "tools_ms": 801, "other_ms": 14, "turns": 1, "artifact": "results/reddit_2.log"}
+{"site": "gmail", "trial": 2, "bucket": "not_measurable", "wall_s": 11.6, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 148, "llm_ms": 0, "tools_ms": 138, "other_ms": 10, "turns": 1, "artifact": "results/gmail_2.log"}
+{"site": "youtube", "trial": 2, "bucket": "ok", "wall_s": 33.6, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=A4YuhvxN", "composer": true, "submit": true, "total_ms": 10560, "llm_ms": 0, "tools_ms": 7943, "other_ms": 2617, "turns": 1, "artifact": "results/youtube_2.log"}
+{"site": "instagram", "trial": 2, "bucket": "product_no_composer", "wall_s": 29.8, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 7222, "llm_ms": 0, "tools_ms": 4603, "other_ms": 2619, "turns": 1, "artifact": "results/instagram_2.log"}
+{"site": "tiktok", "trial": 2, "bucket": "not_measurable", "wall_s": 21.5, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@poongcrew/)", "composer": false, "submit": false, "total_ms": 275, "llm_ms": 0, "tools_ms": 256, "other_ms": 19, "turns": 1, "artifact": "results/tiktok_2.log"}
+{"site": "substack", "trial": 2, "bucket": "not_measurable", "wall_s": 17.6, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 6017, "llm_ms": 0, "tools_ms": 3400, "other_ms": 2617, "turns": 1, "artifact": "results/substack_2.log"}
+{"site": "twitch", "trial": 2, "bucket": "ok", "wall_s": 9.9, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 777, "llm_ms": 0, "tools_ms": 764, "other_ms": 13, "turns": 1, "artifact": "results/twitch_2.log"}
+{"site": "x", "trial": 3, "bucket": "ok", "wall_s": 14.0, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 2361, "llm_ms": 0, "tools_ms": 1831, "other_ms": 530, "turns": 1, "artifact": "results/x_3.log"}
+{"site": "linkedin", "trial": 3, "bucket": "ok", "wall_s": 13.5, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 1458, "llm_ms": 0, "tools_ms": 1435, "other_ms": 23, "turns": 1, "artifact": "results/linkedin_3.log"}
+{"site": "reddit", "trial": 3, "bucket": "ok", "wall_s": 7.5, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 953, "llm_ms": 0, "tools_ms": 928, "other_ms": 25, "turns": 1, "artifact": "results/reddit_3.log"}
+{"site": "gmail", "trial": 3, "bucket": "not_measurable", "wall_s": 11.6, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 166, "llm_ms": 0, "tools_ms": 155, "other_ms": 11, "turns": 1, "artifact": "results/gmail_3.log"}
+{"site": "youtube", "trial": 3, "bucket": "ok", "wall_s": 35.7, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=c0jZ7sPy", "composer": true, "submit": true, "total_ms": 10746, "llm_ms": 0, "tools_ms": 8127, "other_ms": 2619, "turns": 1, "artifact": "results/youtube_3.log"}
+{"site": "instagram", "trial": 3, "bucket": "ok", "wall_s": 12.3, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbO3F_Mgf7kg", "composer": true, "submit": true, "total_ms": 1344, "llm_ms": 0, "tools_ms": 1308, "other_ms": 36, "turns": 1, "artifact": "results/instagram_3.log"}
+{"site": "tiktok", "trial": 3, "bucket": "not_measurable", "wall_s": 17.4, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@callmetash)", "composer": false, "submit": false, "total_ms": 367, "llm_ms": 0, "tools_ms": 352, "other_ms": 15, "turns": 1, "artifact": "results/tiktok_3.log"}
+{"site": "substack", "trial": 3, "bucket": "not_measurable", "wall_s": 17.9, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 5991, "llm_ms": 0, "tools_ms": 3376, "other_ms": 2615, "turns": 1, "artifact": "results/substack_3.log"}
+{"site": "twitch", "trial": 3, "bucket": "ok", "wall_s": 11.4, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 784, "llm_ms": 0, "tools_ms": 772, "other_ms": 12, "turns": 1, "artifact": "results/twitch_3.log"}
+{"site": "x", "trial": 4, "bucket": "ok", "wall_s": 9.7, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 1562, "llm_ms": 0, "tools_ms": 1046, "other_ms": 516, "turns": 1, "artifact": "results/x_4.log"}
+{"site": "linkedin", "trial": 4, "bucket": "ok", "wall_s": 13.6, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 1415, "llm_ms": 0, "tools_ms": 1391, "other_ms": 24, "turns": 1, "artifact": "results/linkedin_4.log"}
+{"site": "reddit", "trial": 4, "bucket": "ok", "wall_s": 7.4, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 973, "llm_ms": 0, "tools_ms": 956, "other_ms": 17, "turns": 1, "artifact": "results/reddit_4.log"}
+{"site": "gmail", "trial": 4, "bucket": "not_measurable", "wall_s": 12.1, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 172, "llm_ms": 0, "tools_ms": 162, "other_ms": 10, "turns": 1, "artifact": "results/gmail_4.log"}
+{"site": "youtube", "trial": 4, "bucket": "ok", "wall_s": 37.0, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=C0zbypHp", "composer": true, "submit": true, "total_ms": 10787, "llm_ms": 0, "tools_ms": 8167, "other_ms": 2620, "turns": 1, "artifact": "results/youtube_4.log"}
+{"site": "instagram", "trial": 4, "bucket": "ok", "wall_s": 13.7, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbO3F_Mgf7kg", "composer": true, "submit": true, "total_ms": 1135, "llm_ms": 0, "tools_ms": 1122, "other_ms": 13, "turns": 1, "artifact": "results/instagram_4.log"}
+{"site": "tiktok", "trial": 4, "bucket": "not_measurable", "wall_s": 19.5, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@joshh_x0nx)", "composer": false, "submit": false, "total_ms": 367, "llm_ms": 0, "tools_ms": 354, "other_ms": 13, "turns": 1, "artifact": "results/tiktok_4.log"}
+{"site": "substack", "trial": 4, "bucket": "infra_harness", "wall_s": 78.1, "detail": "INVALID: parent rephrased the task, payload came through ambiguous", "composer": false, "submit": false, "total_ms": 3916, "llm_ms": 0, "tools_ms": 736, "other_ms": 3180, "turns": 1, "artifact": "results/substack_4.log"}
+{"site": "twitch", "trial": 4, "bucket": "ok", "wall_s": 9.3, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 747, "llm_ms": 0, "tools_ms": 731, "other_ms": 16, "turns": 1, "artifact": "results/twitch_4.log"}
+{"site": "x", "trial": 5, "bucket": "ok", "wall_s": 12.1, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 3034, "llm_ms": 0, "tools_ms": 2434, "other_ms": 600, "turns": 1, "artifact": "results/x_5.log"}
+{"site": "linkedin", "trial": 5, "bucket": "ok", "wall_s": 17.1, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 2063, "llm_ms": 0, "tools_ms": 2006, "other_ms": 57, "turns": 1, "artifact": "results/linkedin_5.log"}
+{"site": "reddit", "trial": 5, "bucket": "ok", "wall_s": 7.5, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 934, "llm_ms": 0, "tools_ms": 922, "other_ms": 12, "turns": 1, "artifact": "results/reddit_5.log"}
+{"site": "gmail", "trial": 5, "bucket": "not_measurable", "wall_s": 11.4, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 171, "llm_ms": 0, "tools_ms": 162, "other_ms": 9, "turns": 1, "artifact": "results/gmail_5.log"}
+{"site": "youtube", "trial": 5, "bucket": "ok", "wall_s": 33.6, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=hopQXbkX", "composer": true, "submit": true, "total_ms": 10570, "llm_ms": 0, "tools_ms": 7954, "other_ms": 2616, "turns": 1, "artifact": "results/youtube_5.log"}
+{"site": "instagram", "trial": 5, "bucket": "ok", "wall_s": 13.8, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbO3F_Mgf7kg", "composer": true, "submit": true, "total_ms": 1106, "llm_ms": 0, "tools_ms": 1094, "other_ms": 12, "turns": 1, "artifact": "results/instagram_5.log"}
+{"site": "tiktok", "trial": 5, "bucket": "not_measurable", "wall_s": 49.9, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 14469, "llm_ms": 0, "tools_ms": 8843, "other_ms": 5626, "turns": 1, "artifact": "results/tiktok_5.log"}
+{"site": "substack", "trial": 5, "bucket": "not_measurable", "wall_s": 25.7, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 4412, "llm_ms": 0, "tools_ms": 1797, "other_ms": 2615, "turns": 1, "artifact": "results/substack_5.log"}
+{"site": "twitch", "trial": 5, "bucket": "ok", "wall_s": 11.5, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 971, "llm_ms": 0, "tools_ms": 954, "other_ms": 17, "turns": 1, "artifact": "results/twitch_5.log"}
+{"site": "x", "trial": 6, "bucket": "ok", "wall_s": 10.5, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 1816, "llm_ms": 0, "tools_ms": 1302, "other_ms": 514, "turns": 1, "artifact": "results/x_6.log"}
+{"site": "linkedin", "trial": 6, "bucket": "ok", "wall_s": 14.3, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 1419, "llm_ms": 0, "tools_ms": 1401, "other_ms": 18, "turns": 1, "artifact": "results/linkedin_6.log"}
+{"site": "reddit", "trial": 6, "bucket": "ok", "wall_s": 10.7, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 900, "llm_ms": 0, "tools_ms": 874, "other_ms": 26, "turns": 1, "artifact": "results/reddit_6.log"}
+{"site": "gmail", "trial": 6, "bucket": "not_measurable", "wall_s": 15.6, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 186, "llm_ms": 0, "tools_ms": 176, "other_ms": 10, "turns": 1, "artifact": "results/gmail_6.log"}
+{"site": "youtube", "trial": 6, "bucket": "ok", "wall_s": 39.4, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=zmS7kzSH", "composer": true, "submit": true, "total_ms": 10487, "llm_ms": 0, "tools_ms": 7869, "other_ms": 2618, "turns": 1, "artifact": "results/youtube_6.log"}
+{"site": "instagram", "trial": 6, "bucket": "product_no_composer", "wall_s": 23.4, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 7748, "llm_ms": 0, "tools_ms": 5128, "other_ms": 2620, "turns": 1, "artifact": "results/instagram_6.log"}
+{"site": "tiktok", "trial": 6, "bucket": "not_measurable", "wall_s": 19.5, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@futurewag7)", "composer": false, "submit": false, "total_ms": 302, "llm_ms": 0, "tools_ms": 292, "other_ms": 10, "turns": 1, "artifact": "results/tiktok_6.log"}
+{"site": "substack", "trial": 6, "bucket": "not_measurable", "wall_s": 17.6, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 6000, "llm_ms": 0, "tools_ms": 3385, "other_ms": 2615, "turns": 1, "artifact": "results/substack_6.log"}
+{"site": "twitch", "trial": 6, "bucket": "ok", "wall_s": 11.5, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 743, "llm_ms": 0, "tools_ms": 733, "other_ms": 10, "turns": 1, "artifact": "results/twitch_6.log"}
+{"site": "x", "trial": 7, "bucket": "ok", "wall_s": 12.4, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 1629, "llm_ms": 0, "tools_ms": 1110, "other_ms": 519, "turns": 1, "artifact": "results/x_7.log"}
+{"site": "linkedin", "trial": 7, "bucket": "ok", "wall_s": 13.7, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 2291, "llm_ms": 0, "tools_ms": 2276, "other_ms": 15, "turns": 1, "artifact": "results/linkedin_7.log"}
+{"site": "reddit", "trial": 7, "bucket": "ok", "wall_s": 7.8, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 987, "llm_ms": 0, "tools_ms": 965, "other_ms": 22, "turns": 1, "artifact": "results/reddit_7.log"}
+{"site": "gmail", "trial": 7, "bucket": "not_measurable", "wall_s": 14.1, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 201, "llm_ms": 0, "tools_ms": 192, "other_ms": 9, "turns": 1, "artifact": "results/gmail_7.log"}
+{"site": "youtube", "trial": 7, "bucket": "ok", "wall_s": 32.0, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=9xtAKakz", "composer": true, "submit": true, "total_ms": 10755, "llm_ms": 0, "tools_ms": 8141, "other_ms": 2614, "turns": 1, "artifact": "results/youtube_7.log"}
+{"site": "instagram", "trial": 7, "bucket": "product_wrong_surface", "wall_s": 17.8, "detail": "WRONG SURFACE: wrong page https://www.instagram.com/reels/DbktzhHuBnj/", "composer": false, "submit": false, "total_ms": 571, "llm_ms": 0, "tools_ms": 560, "other_ms": 11, "turns": 1, "artifact": "results/instagram_7.log"}
+{"site": "tiktok", "trial": 7, "bucket": "not_measurable", "wall_s": 18.2, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@cadelandmi)", "composer": false, "submit": false, "total_ms": 352, "llm_ms": 0, "tools_ms": 341, "other_ms": 11, "turns": 1, "artifact": "results/tiktok_7.log"}
+{"site": "substack", "trial": 7, "bucket": "not_measurable", "wall_s": 17.9, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 6121, "llm_ms": 0, "tools_ms": 3506, "other_ms": 2615, "turns": 1, "artifact": "results/substack_7.log"}
+{"site": "twitch", "trial": 7, "bucket": "ok", "wall_s": 11.2, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 701, "llm_ms": 0, "tools_ms": 688, "other_ms": 13, "turns": 1, "artifact": "results/twitch_7.log"}
+{"site": "x", "trial": 8, "bucket": "ok", "wall_s": 14.0, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 3956, "llm_ms": 0, "tools_ms": 3324, "other_ms": 632, "turns": 1, "artifact": "results/x_8.log"}
+{"site": "linkedin", "trial": 8, "bucket": "ok", "wall_s": 18.4, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 4182, "llm_ms": 0, "tools_ms": 3480, "other_ms": 702, "turns": 1, "artifact": "results/linkedin_8.log"}
+{"site": "reddit", "trial": 8, "bucket": "ok", "wall_s": 7.4, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 887, "llm_ms": 0, "tools_ms": 873, "other_ms": 14, "turns": 1, "artifact": "results/reddit_8.log"}
+{"site": "gmail", "trial": 8, "bucket": "not_measurable", "wall_s": 13.8, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 225, "llm_ms": 0, "tools_ms": 213, "other_ms": 12, "turns": 1, "artifact": "results/gmail_8.log"}
+{"site": "youtube", "trial": 8, "bucket": "ok", "wall_s": 31.6, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=hN2InEt1", "composer": true, "submit": true, "total_ms": 10583, "llm_ms": 0, "tools_ms": 7956, "other_ms": 2627, "turns": 1, "artifact": "results/youtube_8.log"}
+{"site": "instagram", "trial": 8, "bucket": "ok", "wall_s": 9.9, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbkML-VgTm9/", "composer": true, "submit": true, "total_ms": 900, "llm_ms": 0, "tools_ms": 889, "other_ms": 11, "turns": 1, "artifact": "results/instagram_8.log"}
+{"site": "tiktok", "trial": 8, "bucket": "not_measurable", "wall_s": 25.6, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@wkpnkjhzi0)", "composer": false, "submit": false, "total_ms": 857, "llm_ms": 0, "tools_ms": 847, "other_ms": 10, "turns": 1, "artifact": "results/tiktok_8.log"}
+{"site": "substack", "trial": 8, "bucket": "not_measurable", "wall_s": 25.7, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 4410, "llm_ms": 0, "tools_ms": 1794, "other_ms": 2616, "turns": 1, "artifact": "results/substack_8.log"}
+{"site": "twitch", "trial": 8, "bucket": "ok", "wall_s": 14.3, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 716, "llm_ms": 0, "tools_ms": 703, "other_ms": 13, "turns": 1, "artifact": "results/twitch_8.log"}
+{"site": "x", "trial": 9, "bucket": "ok", "wall_s": 13.5, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 1726, "llm_ms": 0, "tools_ms": 1210, "other_ms": 516, "turns": 1, "artifact": "results/x_9.log"}
+{"site": "linkedin", "trial": 9, "bucket": "ok", "wall_s": 13.6, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 2096, "llm_ms": 0, "tools_ms": 2076, "other_ms": 20, "turns": 1, "artifact": "results/linkedin_9.log"}
+{"site": "reddit", "trial": 9, "bucket": "ok", "wall_s": 8.2, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 1061, "llm_ms": 0, "tools_ms": 1046, "other_ms": 15, "turns": 1, "artifact": "results/reddit_9.log"}
+{"site": "gmail", "trial": 9, "bucket": "not_measurable", "wall_s": 14.3, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 207, "llm_ms": 0, "tools_ms": 198, "other_ms": 9, "turns": 1, "artifact": "results/gmail_9.log"}
+{"site": "youtube", "trial": 9, "bucket": "ok", "wall_s": 38.1, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=l2qHL3e7", "composer": true, "submit": true, "total_ms": 10445, "llm_ms": 0, "tools_ms": 7829, "other_ms": 2616, "turns": 1, "artifact": "results/youtube_9.log"}
+{"site": "instagram", "trial": 9, "bucket": "product_wrong_surface", "wall_s": 22.6, "detail": "WRONG SURFACE: wrong page https://www.instagram.com/reels/DbkML-VgTm9/", "composer": false, "submit": false, "total_ms": 1954, "llm_ms": 0, "tools_ms": 1941, "other_ms": 13, "turns": 1, "artifact": "results/instagram_9.log"}
+{"site": "tiktok", "trial": 9, "bucket": "not_measurable", "wall_s": 19.5, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@smartround)", "composer": false, "submit": false, "total_ms": 357, "llm_ms": 0, "tools_ms": 346, "other_ms": 11, "turns": 1, "artifact": "results/tiktok_9.log"}
+{"site": "substack", "trial": 9, "bucket": "not_measurable", "wall_s": 23.6, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 4450, "llm_ms": 0, "tools_ms": 1835, "other_ms": 2615, "turns": 1, "artifact": "results/substack_9.log"}
+{"site": "twitch", "trial": 9, "bucket": "ok", "wall_s": 12.4, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 721, "llm_ms": 0, "tools_ms": 710, "other_ms": 11, "turns": 1, "artifact": "results/twitch_9.log"}
+{"site": "x", "trial": 10, "bucket": "ok", "wall_s": 12.5, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 1753, "llm_ms": 0, "tools_ms": 1220, "other_ms": 533, "turns": 1, "artifact": "results/x_10.log"}
+{"site": "linkedin", "trial": 10, "bucket": "ok", "wall_s": 19.9, "detail": "'comment' <- 'Text editor for creating content' on https://www.linkedin.com/feed/?shareActi", "composer": true, "submit": true, "total_ms": 4378, "llm_ms": 0, "tools_ms": 4355, "other_ms": 23, "turns": 1, "artifact": "results/linkedin_10.log"}
+{"site": "reddit", "trial": 10, "bucket": "ok", "wall_s": 7.5, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 1167, "llm_ms": 0, "tools_ms": 1156, "other_ms": 11, "turns": 1, "artifact": "results/reddit_10.log"}
+{"site": "gmail", "trial": 10, "bucket": "not_measurable", "wall_s": 14.4, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 298, "llm_ms": 0, "tools_ms": 274, "other_ms": 24, "turns": 1, "artifact": "results/gmail_10.log"}
+{"site": "youtube", "trial": 10, "bucket": "ok", "wall_s": 35.6, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=QbYdd0io", "composer": true, "submit": true, "total_ms": 10818, "llm_ms": 0, "tools_ms": 8202, "other_ms": 2616, "turns": 1, "artifact": "results/youtube_10.log"}
+{"site": "instagram", "trial": 10, "bucket": "ok", "wall_s": 12.0, "detail": "'post' <- 'Add a comment\u2026' on https://www.instagram.com/p/DbphiQPlmjFa", "composer": true, "submit": true, "total_ms": 3174, "llm_ms": 0, "tools_ms": 555, "other_ms": 2619, "turns": 1, "artifact": "results/instagram_10.log"}
+{"site": "tiktok", "trial": 10, "bucket": "not_measurable", "wall_s": 53.8, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 13248, "llm_ms": 0, "tools_ms": 7625, "other_ms": 5623, "turns": 1, "artifact": "results/tiktok_10.log"}
+{"site": "substack", "trial": 10, "bucket": "not_measurable", "wall_s": 23.2, "detail": "NOT MEASURABLE: signed out per the agent, UNVERIFIED (no wall evidence)", "composer": false, "submit": false, "total_ms": 6162, "llm_ms": 0, "tools_ms": 3548, "other_ms": 2614, "turns": 1, "artifact": "results/substack_10.log"}
+{"site": "twitch", "trial": 10, "bucket": "ok", "wall_s": 11.8, "detail": "'send chat' <- 'Send a message' on https://www.twitch.tv/caedrel", "composer": true, "submit": true, "total_ms": 6162, "llm_ms": 0, "tools_ms": 3548, "other_ms": 2614, "turns": 1, "artifact": "results/twitch_10.log"}
+{"site": "x", "trial": 11, "bucket": "ok", "wall_s": 60.7, "detail": "'post' <- 'Post text' on https://x.com/compose/post", "composer": true, "submit": true, "total_ms": 22712, "llm_ms": 0, "tools_ms": 20722, "other_ms": 1990, "turns": 1, "artifact": "results/x_11.log"}
+{"site": "linkedin", "trial": 11, "bucket": "infra_router", "wall_s": 61.4, "detail": "INVALID: stack was unhealthy (No AI provider connected)", "composer": false, "submit": false, "total_ms": 4298, "llm_ms": 0, "tools_ms": 4141, "other_ms": 157, "turns": 1, "artifact": "results/linkedin_11.log"}
+{"site": "reddit", "trial": 11, "bucket": "ok", "wall_s": 31.1, "detail": "'the submit control is present but <- 'Post body text field' on https://www.reddit.com/r/test/submit/?ty", "composer": true, "submit": false, "total_ms": 2283, "llm_ms": 0, "tools_ms": 2233, "other_ms": 50, "turns": 1, "artifact": "results/reddit_11.log"}
+{"site": "gmail", "trial": 11, "bucket": "not_measurable", "wall_s": 39.5, "detail": "NOT MEASURABLE: signed out, PROVEN (sign-in URL https://accounts.google.com/v3/signin/acco)", "composer": false, "submit": false, "total_ms": 1468, "llm_ms": 0, "tools_ms": 1445, "other_ms": 23, "turns": 1, "artifact": "results/gmail_11.log"}
+{"site": "youtube", "trial": 11, "bucket": "ok", "wall_s": 60.3, "detail": "'post' <- 'contenteditable' on https://www.youtube.com/watch?v=6dROuF_T", "composer": true, "submit": true, "total_ms": 16285, "llm_ms": 0, "tools_ms": 13547, "other_ms": 2738, "turns": 1, "artifact": "results/youtube_11.log"}
+{"site": "instagram", "trial": 11, "bucket": "product_no_composer", "wall_s": 82.3, "detail": "decline: no composer, opener, or structural editable", "composer": false, "submit": false, "total_ms": 25483, "llm_ms": 0, "tools_ms": 19770, "other_ms": 5713, "turns": 1, "artifact": "results/instagram_11.log"}
+{"site": "tiktok", "trial": 11, "bucket": "not_measurable", "wall_s": 42.5, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://www.tiktok.com/@elliegetaj)", "composer": false, "submit": false, "total_ms": 929, "llm_ms": 0, "tools_ms": 900, "other_ms": 29, "turns": 1, "artifact": "results/tiktok_11.log"}
+{"site": "substack", "trial": 11, "bucket": "not_measurable", "wall_s": 24.9, "detail": "NOT MEASURABLE: login wall claimed, UNVERIFIED (no evidence for https://substack.com/)", "composer": false, "submit": false, "total_ms": 921, "llm_ms": 0, "tools_ms": 800, "other_ms": 121, "turns": 1, "artifact": "results/substack_11.log"}
diff --git a/e2e/browser-v3/skillstats.py b/e2e/browser-v3/skillstats.py
new file mode 100644
index 00000000..9af960aa
--- /dev/null
+++ b/e2e/browser-v3/skillstats.py
@@ -0,0 +1,89 @@
+"""Criterion 9: does the learned fast path record, and does replaying it help?
+
+Reads a backend log and reports the two rates the criterion names, plus the reason every refusal
+gave. Every string below is grep-verified against backend/apps/agents/browser/, because the last
+instrument that measured this layer looked for a line the code never prints.
+
+    python3 skillstats.py <backend.log> [...]
+"""
+
+import re
+import sys
+from collections import Counter
+
+# (label, regex). Anchored on the exact logger.info text in browser_skills.py / browser_agent.py.
+PATTERNS = [
+    ("gate_passed",   re.compile(r"record gate: honest=True informational=False removal=False unconfirmed_send=False")),
+    ("gate_refused",  re.compile(r"record gate: (?!honest=True informational=False removal=False unconfirmed_send=False)")),
+    ("recorded",      re.compile(r"\[browser-skills\] (learned|EDITED) \d+-step skill for (\S+)")),
+    ("re_derived",    re.compile(r"re-derived identical \d+-step skill")),
+    ("not_recorded",  re.compile(r"NOT recorded \(([^)]*)\)")),
+    ("matched",       re.compile(r"\[browser-skills\] skill matched on (\S+)")),
+    ("no_skill",      re.compile(r"no skill for host=")),
+    ("replay_prefix",  re.compile(r"PREFIX replay: (\d+)/(\d+) steps on (\S+)")),
+    ("replay_attempt", re.compile(r"REPLAY attempt: (\d+) steps on (\S+)")),
+    ("replay_ok",      re.compile(r"REPLAY SUCCEEDED in (\d+)ms")),
+    ("replay_failed",  re.compile(r"replay step failed \(")),
+    ("not_replayed",   re.compile(r"skill on (\S+) not replayed: (.+?);")),
+]
+
+
+def main() -> int:
+    text = ""
+    for p in sys.argv[1:]:
+        try:
+            text += open(p, errors="ignore").read()
+        except OSError as e:
+            print(f"skip {p}: {e}")
+    if not text:
+        print("no log content")
+        return 2
+
+    counts = Counter()
+    hosts_recorded = Counter()
+    refusals = Counter()
+    not_replayed = Counter()
+    for line in text.splitlines():
+        for label, rx in PATTERNS:
+            m = rx.search(line)
+            if not m:
+                continue
+            counts[label] += 1
+            if label == "recorded":
+                hosts_recorded[m.group(2)] += 1
+            if label == "not_recorded":
+                refusals[m.group(1)] += 1
+            if label == "not_replayed":
+                not_replayed[m.group(2)[:60]] += 1
+
+    gate_passed = counts["gate_passed"]
+    recorded = counts["recorded"] + counts["re_derived"]
+    print("=== RECORDING ===")
+    print(f"  runs reaching the record gate : {gate_passed + counts['gate_refused']}")
+    print(f"  gate PASSED (eligible)        : {gate_passed}")
+    print(f"  skills recorded               : {recorded}"
+          + (f"  ({100*recorded//gate_passed}% of eligible)" if gate_passed else ""))
+    for why, n in refusals.most_common():
+        print(f"    refused: {why}  x{n}")
+    for h, n in hosts_recorded.most_common(10):
+        print(f"    recorded on {h}  x{n}")
+
+    print("\n=== REPLAY ===")
+    print(f"  skill matched for the host    : {counts['matched']}")
+    print(f"  no skill for the host         : {counts['no_skill']}")
+    print(f"  full replay attempts          : {counts['replay_attempt']}")
+    print(f"  full replays SUCCEEDED        : {counts['replay_ok']}")
+    print(f"  prefix replays performed      : {counts['replay_prefix']}")
+    print(f"  replays refused (unsafe step) : {counts['not_replayed']}")
+    for why, n in not_replayed.most_common(6):
+        print(f"    {why}  x{n}")
+    print(f"  replay step failures          : {counts['replay_failed']}")
+    replayed = counts["replay_attempt"] + counts["replay_prefix"]
+    print(f"\n  RECORDING RATE {recorded}/{gate_passed}"
+          f" = {100*recorded//gate_passed if gate_passed else 0}%   (criterion 9 wants >=50%)")
+    print(f"  REPLAYS PERFORMED {replayed}, of which succeeded {counts['replay_ok']}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/e2e/browser-v3/stack.sh b/e2e/browser-v3/stack.sh
new file mode 100755
index 00000000..959c2d6c
--- /dev/null
+++ b/e2e/browser-v3/stack.sh
@@ -0,0 +1,129 @@
+#!/bin/bash
+# The isolated v3 measurement stack: backend :8326, webpack :3026, Electron on its own profile.
+#
+# One script because rebuilding it by hand three times cost three different half-booted stacks, and a
+# half-booted stack does not fail loudly, it just measures nothing and blames the product.
+#
+#   ./stack.sh up dry    backend refuses the irreversible click (coverage sweeps)
+#   ./stack.sh up live    backend really clicks send (canary write tests)
+#   ./stack.sh down       everything, SIGTERM then SIGKILL, ports verified free
+#   ./stack.sh status     what is up right now
+#
+# Never touches :8324 / :3000. Those belong to whatever else is on this box.
+
+# Where logs, profiles and run output go. Defaults to runs/ beside this harness; override with
+# OSW_BENCH_DIR to keep multi-gigabyte browser profiles off the repo disk.
+SP="${OSW_BENCH_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/runs}"
+mkdir -p "$SP"
+# The harness itself lives beside this script; SP is only for run OUTPUT.
+HARNESS="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+TREE="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+TAG="${TAG:-run}"
+
+status() {
+  # Two bugs lived in the first version of these four lines, and both printed a reassuring 0 over a
+  # fully live stack. macOS pgrep has no -c flag at all, so `pgrep -fc` exits on a usage error and
+  # the count renders empty; and the patterns were the words I type rather than the words ps prints
+  # (uvicorn runs as `python -m uvicorn`). A status check that reads clean while the box is busy is
+  # the worst instrument in this directory, because its whole job is to stop a second stack landing
+  # on the first. Count with grep on real ps output.
+  echo "=== procs ==="
+  local ps_out
+  ps_out="$(ps -Ao pid,command)"
+  for p in "\-m uvicorn backend.main" "bin/webpack" "MacOS/Electron" \
+           "keep_renderer" "bench\.py" "browser_canary"; do
+    printf "  %-34s %s\n" "$p" "$(echo "$ps_out" | grep -cE "$p")"
+  done
+  echo "=== ports ==="
+  for pt in 8324 8326 3000 3026 20128; do
+    printf "  %-6s %s\n" "$pt" "$(lsof -ti tcp:$pt 2>/dev/null | tr '\n' ' ')"
+  done
+  # Anything on :8324 is the OTHER checkout. Say so out loud: while it is up, the single 9router on
+  # :20128 is contended and every timing this stack produces is noise (a sweep once scored 1/9 vs
+  # 4/9 with zero code change, purely on that contention).
+  if [ -n "$(lsof -ti tcp:8324 2>/dev/null)" ]; then
+    echo "  !! another OpenSwarm is on :8324. Do NOT measure, and do NOT kill it."
+  fi
+}
+
+down() {
+  # SCOPED TO THIS STACK ONLY. The first version matched on `uvicorn backend.main`, which is exactly
+  # what the OTHER OpenSwarm checkout on this box runs too: one careless `stack.sh down` would have
+  # killed a colleague's backend on :8324 mid-session. Nothing here may match a process this script
+  # did not start, so identify them by MY ports and MY profile directory, never by a generic name.
+  pkill -f "keep_renderer" 2>/dev/null
+  pkill -f "BACKEND RESTARTED" 2>/dev/null
+  pkill -f "user-data-dir=$SP/udd" 2>/dev/null
+  pkill -f "browser_canary" 2>/dev/null
+  pkill -f "$HARNESS/bench.py" 2>/dev/null
+  sleep 3
+  for pt in 8326 3026; do
+    pids=$(lsof -ti tcp:$pt 2>/dev/null)
+    [ -n "$pids" ] && kill -9 $pids 2>/dev/null
+  done
+  pkill -9 -f "user-data-dir=$SP/udd" 2>/dev/null
+  sleep 1
+  status
+}
+
+up() {
+  local mode="${1:-dry}"
+  local dry=0
+  [ "$mode" = "dry" ] && dry=1
+
+  # A stale stack under a fresh one is the single most expensive failure here: two backends fight
+  # over the one 9router and every number becomes a coin flip. Always start from nothing.
+  down >/dev/null 2>&1
+
+  # Supervised, because the backend has died mid-measurement on a clean SIGTERM with no error in
+  # its log (2026-08-05 07:44, 52 minutes in, no other stack on the box). An unsupervised death does
+  # not announce itself: the harness just starts recording connection errors as product failures.
+  # The restart marker goes into the same log the harness slices, so any trial that spans a restart
+  # can be excluded instead of counted.
+  cd "$TREE" || exit 1
+  nohup bash -c '
+    while true; do
+      OPENSWARM_PORT=8326 OSW_SENDSCRIPT_DRYRUN='"$dry"' OPENSWARM_DEV=1 \
+        ./backend/.venv/bin/python -m uvicorn backend.main:app --port 8326 --host 127.0.0.1 \
+        >> "'"$SP/${TAG}_be.log"'" 2>&1
+      echo "[stack] BACKEND RESTARTED at $(date +%H:%M:%S)" >> "'"$SP/${TAG}_be.log"'"
+      sleep 4
+    done' > /dev/null 2>&1 &
+  disown
+
+  cd "$TREE/frontend" || exit 1
+  OPENSWARM_DEV_PORT=3026 OPENSWARM_PORT=8326 \
+    ./node_modules/.bin/webpack serve --mode development \
+    > "$SP/${TAG}_wp.log" 2>&1 &
+
+  # A real authenticated 200, not just "something accepted a TCP connection". /api/health does not
+  # exist (it 404s), and curl calls a 404 a success, so the old check passed the instant the socket
+  # opened and handed the next step a backend that had not finished booting.
+  echo "waiting for backend :8326 ..."
+  for i in $(seq 1 90); do
+    code=$(curl -s -o /dev/null -w '%{http_code}' --max-time 3 \
+      -H "Authorization: Bearer $(cat "$TREE/backend/data/auth.token" 2>/dev/null)" \
+      "http://127.0.0.1:8326/api/dashboards/list")
+    [ "$code" = "200" ] && break
+    sleep 2
+  done
+  echo "waiting for webpack :3026 ..."
+  for i in $(seq 1 120); do
+    curl -s -o /dev/null --max-time 3 "http://localhost:3026" && break
+    sleep 2
+  done
+
+  OPENSWARM_DEV_PORT=3026 OPENSWARM_PORT=8326 nohup "$HARNESS/keep_renderer.sh" \
+    >> "$SP/renderer.log" 2>&1 &
+  disown
+  sleep 25
+  echo "mode=$mode (OSW_SENDSCRIPT_DRYRUN=$dry) tag=$TAG"
+  status
+}
+
+case "$1" in
+  up) up "$2" ;;
+  down) down ;;
+  status) status ;;
+  *) echo "usage: stack.sh {up dry|up live|down|status}"; exit 1 ;;
+esac
diff --git a/e2e/browser-v3/verify_markers.py b/e2e/browser-v3/verify_markers.py
new file mode 100644
index 00000000..7509bd17
--- /dev/null
+++ b/e2e/browser-v3/verify_markers.py
@@ -0,0 +1,86 @@
+"""Every literal this harness greps for must exist in the code that is supposed to print it.
+
+Written after the fourth instance of the same bug. `DELIVERY CONFIRMED` appeared in the canary and
+nowhere in the backend. `saw_page` listed two `[browser-action] X` strings that are never emitted.
+`replay_full` matched a phrase no logger uses. Each one silently turned a measurement into a
+constant, and each cost hours of chasing a product bug that did not exist.
+
+A grep whose needle is absent from the source cannot fail loudly, so make it fail here instead.
+"""
+
+import os
+import re
+import subprocess
+import sys
+
+TREE = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+HERE = os.path.dirname(os.path.abspath(__file__))
+SP = os.path.dirname(HERE)
+
+# (label, literal, where it must appear). A literal is a fixed substring of a real log line, with
+# the f-string holes cut out, so it can be checked with a plain fixed-string grep.
+MARKERS = [
+    # canary receipt detection
+    ("receipt/sendscript", "done sent_receipt=", "backend/apps"),
+    ("receipt/agent-loop", "two-sided receipt passed", "backend/apps"),
+    ("receipt/autosend", "code-send delivered (receipt verified)", "backend/apps"),
+    # coverage.py grading
+    ("dryrun report", "DRYRUN: WOULD send (fill committed", "backend/apps"),
+    ("decline", "[browser-sendscript] decline: ", "backend/apps"),
+    ("disabled submit", "is present but DISABLED", "backend/apps"),
+    ("fill target", "[browser-sendscript] fill target ", "backend/apps"),
+    ("fill errored", "fill errored (", "backend/apps"),
+    ("login wall", "decline: login/auth wall", "backend/apps"),
+    ("signed out", "decline: signed OUT", "backend/apps"),
+    ("recovery", "one recovery dispatch", "backend/apps"),
+    # coverage.py UNHEALTHY (infrastructure)
+    ("router watchdog", "9Router watchdog", "backend/apps"),
+    ("router died", "9Router process died", "backend/apps"),
+    ("no provider", "No AI provider connected", "backend/apps"),
+    ("no dashboard", "dispatch refused: no dashboard", "backend/apps"),
+    # bench.py infra buckets
+    ("browser timeout", "Browser command timed out", "backend/apps"),
+    ("card gone/webview", "not an electron webview", "backend/apps"),
+    ("card gone/dashboard", "no dashboard is connected", "backend/apps"),
+    ("card gone/unresponsive", "page unresponsive", "backend/apps"),
+    ("busy to read", "too busy to read", "frontend/src"),
+    # skillstats.py
+    ("skill record gate", "record gate: honest=", "backend/apps"),
+    ("skill recorded", "-step skill for ", "backend/apps"),
+    ("skill re-derived", "re-derived identical ", "backend/apps"),
+    ("skill not recorded", "NOT recorded (", "backend/apps"),
+    ("skill matched", "skill matched on ", "backend/apps"),
+    ("no skill", "no skill for host=", "backend/apps"),
+    ("prefix replay", "PREFIX replay: ", "backend/apps"),
+    ("replay attempt", "REPLAY attempt: ", "backend/apps"),
+    ("replay succeeded", "REPLAY SUCCEEDED in ", "backend/apps"),
+    ("replay step failed", "replay step failed (", "backend/apps"),
+    ("not replayed", " not replayed: ", "backend/apps"),
+    # the honesty marker the packaged smokes also grep for
+    ("send-not-verified", "[send clicked, NOT verified]", "backend/apps"),
+]
+
+
+def main() -> int:
+    bad = []
+    for label, needle, where in MARKERS:
+        r = subprocess.run(["grep", "-rF", "--", needle, os.path.join(TREE, where)],
+                           capture_output=True, text=True)
+        n = len([ln for ln in r.stdout.splitlines() if ln.strip()])
+        status = "ok " if n else "DEAD"
+        if not n:
+            bad.append((label, needle))
+        print(f"  {status} {label:<22} x{n:<3} {needle!r}")
+    print()
+    if bad:
+        print(f"{len(bad)} DEAD marker(s): a grep for these can never match, so whatever they")
+        print("measure is a constant. Fix the needle or delete the check.")
+        for label, needle in bad:
+            print(f"  - {label}: {needle!r}")
+        return 1
+    print(f"all {len(MARKERS)} markers exist in the source that prints them")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())