Handbook
Task `pw_grading_signal_infer` v1
From bounded DOM/state deltas before submit and after a dummy submit, infer a declarative correct_signal description for grading (CSS classes, selectors, feedback text patterns). LCDL does not execute Playwright.
Task pw_grading_signal_infer v1
Summary
From bounded DOM/state deltas before submit and after a dummy submit, infer a declarative correct_signal description for grading (CSS classes, selectors, feedback text patterns). LCDL does not execute Playwright.
Inputs
| Field | Type | Required | Notes |
|---|---|---|---|
before_answer |
object | yes | Snapshot-ish JSON (may be {}). |
after_dummy_submit |
object | yes | Post-submit snapshot (may be {}). |
option_candidates |
array | yes | Candidate option descriptors (may be []). |
class_deltas |
array | yes | Class/token deltas observed (may be []). |
attribute_deltas |
array | yes | Attribute deltas (may be []). |
feedback_text |
string | yes | Truncated feedback text excerpt (may be ""). |
temperature |
number | no | Default 0.05. |
timeout_sec |
int | no | Default profile.timeout_sec. |
Output
| Field | Type | Notes |
|---|---|---|
correct_signal |
object | Must include string kind (taxonomy label). Typically includes selectors/patterns as applicable. |
confidence |
number | Coerced to [0.0, 1.0]. |
evidence |
array of string | Short factual bullets grounded in deltas/text. |
Policy
- Output exactly one JSON object (no Python source).
- No bypass guidance for login, CAPTCHA, paywalls, anti-bot, or proctored controls.
Changelog
- v1 — Initial grading-signal inference.