Add Craigslist mirror + task verifiers (site by @FlowRays, verifiers by reviewer)#60
Open
DEM1TASSE wants to merge 4 commits into
Open
Add Craigslist mirror + task verifiers (site by @FlowRays, verifiers by reviewer)#60DEM1TASSE wants to merge 4 commits into
DEM1TASSE wants to merge 4 commits into
Conversation
Reviewer deliverable for the Craigslist mirror (site by @FlowRays, PR aiming-lab#5): one deterministic verifier per task under sites/craigslist/verify/, plus verifier_path + judge_rubric in every tasks.jsonl row. No answer key in tasks.jsonl — ground truth lives only inside the verifiers. Deterministic-first: listing-open navigation gate (anti knowledge-shortcut; listings are synthetic) + SQLite DB after-state for the stateful tasks (saved_listings / saved_searches / hidden_listings / messages / new postings / user profile) + answer vs frozen ground truth, LLM only as an anchored check. Validated against the official react agent (agent_demo/agent.py): a no-op run fails all 20 verifiers; on the full 20-task run verifier and LLM judge agree 14/20, and the divergences favor the deterministic verifier (it catches DB after-state and rejects blank/incomplete answers the judge false-positives). Note: verify_0 and verify_3 are limited by a task bug — their save-targets are already in the seed's saved-listings, so the save cannot be verified; the review asks the author to fix those tasks. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Draft
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Brings the Craigslist mirror + its task verifiers into
mainas one PR, following the samefork-of-fork pattern as #57/#58/#59. Companion to the site PR #5.
Authorship is preserved — both contributors stay credited. This branch is @FlowRays's Craigslist
site commits with the reviewer's verifier commit on top; each commit keeps its original author
(site = @FlowRays, verifiers = the reviewer). Please merge with a regular merge (not squash).
Contents
sites/craigslist/site (Flask app, templates, seed, 20 tasks) — @FlowRays (from Add Craigslist mirror #5)sites/craigslist/verify/verify_<0..19>.py+verify_lib.py, andverifier_path+judge_rubricin every
tasks.jsonlrow — reviewer deliverable. Noanswerkey; ground truth lives onlyinside the verifiers.
Status
Because @FlowRays already rebased #5 onto
main(Craigslist at port 40016, Merriam preserved at40015), this branch does not conflict. However, the full review (posted on #5) is REQUEST CHANGES:
a page-by-page walkthrough found several front-end fidelity / feature-completeness / task-feasibility
issues (a half-fixed sqft leak, two save-tasks whose targets are pre-seeded, a save-search that can't
set the requested category, image/content mismatches, an incomplete hide feature, a
non-category-aware post form, inconsistent favorites labels). The verifiers themselves are valid and
validated and ship here so they're ready once the site fixes land. See #5 for the full list.
Verifier design & validation
Deterministic-first: listing-open navigation gate + SQLite DB after-state for the stateful tasks +
answer vs frozen ground truth, LLM only as an anchored check. No-op fails all 20; on the full agent
run verifier and LLM judge agree 14/20, with divergences favoring the deterministic verifier.
One release-mechanics item
.assets-revisioncurrently pins a SHA that does not containcraigslist.tar.gz; repin to the mergedHF #5 asset SHA. A full review is posted as a comment on #5.