# Browser evidence log

These are transcribed observations from the supplied local browser tool, not screenshots. Each call used a fresh browser with `site_path: output/site`. All fill/click actions for a scenario were sent in the same call. The tool states that its trusted server supplies the original `/api/leads` behavior. The delivered `server.py` was separately verified byte-identical to `site/server.py` (see `review-checks.json`).

## Calculator and layout

| ID | Width | Actions | Visible result |
|---|---:|---|---|
| B01 | 1280 | Open page; default 1 seat | `CAD 268.80` |
| B02 | 390 | Open page; default 1 seat | `CAD 268.80` |
| B03 | 1280 | Fill `#seats` with `2` | `CAD 537.60` |
| B04 | 390 | Fill `#seats` with `3` | `CAD 725.76` |
| B05 | 1280 | Fill `#seats` with `8` | `CAD 1935.36` |

## Valid submissions

| ID | Width | Name / email / seats; then click submit | Visible result |
|---|---:|---|---|
| B06 | 390 | `Mobile Test` / `mobile@example.test` / `3` | `Thanks! Your workshop request is saved.`; `CAD 725.76` |
| B07 | 1280 | `Desktop Test` / `desktop@example.test` / `8` | `Thanks! Your workshop request is saved.`; `CAD 1935.36` |

File inspection after B06 found exactly one JSONL record, matching Mobile Test. Inspection after B07 and the invalid-seat batch found exactly two records, matching Mobile Test and Desktop Test. Final inspection after all invalid-name/email scenarios still found exactly those two records. They are archived verbatim in `browser-submissions.jsonl`, lines 1–2. The review script asserted the full contents before moving this test data out of the runnable site.

## Invalid submissions

For B08–B12, fill name `Invalid Test`, email `invalid@example.test`, fill the indicated seats value, then click `button[type=submit]`.

| ID | Width | Invalid value | Visible result |
|---|---:|---|---|
| B08 | 390 | seats blank | `Choose a whole number of seats from 1 to 8.`; `CAD —` |
| B09 | 1280 | seats `0` | Same seat error; `CAD —` |
| B10 | 390 | seats `-1` | Same seat error; `CAD —` |
| B11 | 1280 | seats `1.5` | Same seat error; `CAD —` |
| B12 | 390 | seats `9` | Same seat error; `CAD —` |
| B13 | 390 | Click submit on untouched form | `Enter your name. Enter a plausible email address (for example, name@example.test).` |
| B14 | 1280 | Name three spaces; email `valid@example.test`; default 1 seat; submit | `Enter your name.` |
| B15 | 390 | Name `Invalid Test`; email `not-an-email`; default 1 seat; submit | `Enter a plausible email address (for example, name@example.test).` |
| B16 | 1280 | Name `Invalid Test`; email `test@localhost`; default 1 seat; submit | Same email error |

For B08–B12, the returned seats control had `valid: false`. For B14 and B16, native validity was true, but the application's stricter validation displayed the error as required by the supplied server's trimmed-name and dotted-domain rules (`site/server.py`, `Handler.do_POST`).

## Common observations and limits

Every one of B01–B16 returned `overflow: false` and `errors: []`. The visible text included the original headings, content, labels and submit control. Controls were successfully filled and clicked in these scenarios at both viewport widths.

No screenshots, real-device checks, zoom checks, keyboard-only navigation tests or screen-reader tests were available/performed. The tool's text observations cannot establish visual polish or actual touch-target ergonomics. No HTTP request trace was captured; successful persistence is supported by the actual JSONL file inspection, not a claimed network trace. Invalid tests exercised client-side rejection; they were not direct hostile API requests. Network failure and server-error UI paths were inspected in code, not browser-tested.
