removed old UI

planning/mvp_implementation_reminders.md

@@ -1,92 +0,0 @@
-# MVP Implementation Reminders (From Scholar Probe)
-
-Last validated probe run: `planning/scholar_probe_tmp/notes/probe_report_run_20260216T182334Z.md`.
-
-## Decision: Are we ready?
-
-Yes, for the scoped MVP.
-
-- We have stable selectors and field coverage for first-page profile parsing.
-- We have explicit failure modes for blocked/inaccessible pages.
-- We have enough fixtures to start parser + ingestion tests immediately.
-
-Not guaranteed (and intentionally out of MVP):
-
-- Full historical completeness from profile pagination.
-- Robots currently disallows `citations?*cstart=`; treat deep pagination as out of scope unless policy changes.
-
-## Non-negotiables
-
-- Keep app container-first (`docker compose`) and test while developing.
-- Keep feature scope small; prefer deterministic behavior over clever scraping.
-- Never crash a whole run because one scholar page fails.
-- Preserve strict tenant isolation for all run/output/read-state data.
-
-## Scraping Contract
-
-Target URL:
-
-- `https://scholar.google.com/citations?hl=en&user=<scholar_id>`
-
-Primary parse markers:
-
-- Row: `tr.gsc_a_tr`
-- Title/link: `a.gsc_a_at`
-- Citation count: `a.gsc_a_ac`
-- Year: `span.gsc_a_h` (fallback year regex)
-- Metadata lines: first/second `div.gs_gray` => authors/venue
-- Cluster id from `citation_for_view=<user>:<cluster_id>`
-
-## Required Parse States
-
-Use and persist one of:
-
-- `ok`
-- `no_results`
-- `blocked_or_captcha`
-- `layout_changed`
-- `network_error`
-
-Plus these page-level flags:
-
-- `has_show_more_button`
-- `articles_range`
-
-## Data Quality Expectations
-
-Observed from probe sample:
-
-- `title`: 100%
-- `cluster_id`: 100%
-- `citation_count`: 100%
-- `authors_text`: 100%
-- `year`: 98.2%
-- `venue_text`: 94.7%
-
-Implications:
-
-- `year` and `venue_text` must remain nullable.
-- Dedupe must not depend solely on `year`/`venue_text` presence.
-
-## Run Semantics
-
-- First successful run per scholar = baseline.
-- If `has_show_more_button=true`, label result as partial in run status/UI.
-- Invalid/inaccessible scholar IDs are expected and should become structured run errors, not exceptions.
-
-## Testing Priorities (Do Before Feature Expansion)
-
-- Fixture parser unit tests using probe HTML snapshots.
-- Parse-state classification tests (ok/blocked/layout/no_results).
-- Dedupe integration tests (`cluster_id` first, fingerprint fallback).
-- Tenant isolation tests for run records and read-state.
-- Smoke test for manual run path through Docker.
-
-## Stop Conditions
-
-Pause implementation and re-probe if:
-
-- Selector markers drop out unexpectedly.
-- Login/redirect pages become frequent for valid IDs.
-- Robots policy for profile endpoints changes.
-

planning/scholar_probe_tmp/README.md

@@ -1,13 +0,0 @@
-# Scholar Scrape Probe (Temporary)
-
-Purpose: short-lived workspace for information gathering and planning.
-
-Rules:
-- No production code changes here.
-- Use this for HTML samples, parsing experiments, and extraction notes.
-- Delete this directory after we lock the scrape contract and parser plan.
-
-Structure:
-- `fixtures/` saved HTML samples for parser validation
-- `notes/` extraction decisions, edge cases, and risk log
-- `scripts/` one-off probe scripts used only during planning

planning/scholar_probe_tmp/notes/findings_phase1_input.md

@@ -1,71 +0,0 @@
-# Probe Findings -> Phase 1 Input
-
-Generated from live probe run: `run_20260216T182334Z`.
-
-## What is stable enough to build on
-
-- Public profile endpoint works for anonymous access:
-  - `GET /citations?hl=en&user=<scholar_id>`
-- Core row structure is present and parseable:
-  - row: `tr.gsc_a_tr`
-  - title + detail URL: `a.gsc_a_at`
-  - citation count: `a.gsc_a_ac`
-  - year: `span.gsc_a_h` (fallback: `td.gsc_a_y` text regex)
-  - metadata lines: first/second `div.gs_gray` => authors/venue
-- `cluster_id` can be extracted reliably from `citation_for_view=<user>:<cluster_id>` in title URLs.
-
-## What can break / where to degrade gracefully
-
-- Invalid or inaccessible profile IDs may redirect to Google sign-in page.
-  - Treat as `blocked_or_captcha` / `inaccessible` state, not parser crash.
-- `Show more` is present for tested profiles.
-  - We currently parse first page only.
-  - Because robots currently disallows `citations?*cstart=`, deep pagination should not be assumed.
-- Some rows are missing year or venue text.
-  - Keep nullable fields and avoid failing dedupe for these gaps.
-
-## Observed quality from probe
-
-- Parsed publication rows: 57 across 4 accessible profiles.
-- Field coverage:
-  - title: 100%
-  - cluster_id: 100%
-  - citation_count: 100%
-  - authors_text: 100%
-  - year: 98.2%
-  - venue_text: 94.7%
-
-## Recommended parser contract for implementation
-
-- Input -> `PublicationCandidate`
-  - `title` (required)
-  - `cluster_id` (required when present in URL; expected high coverage)
-  - `year` (nullable)
-  - `citation_count` (nullable/int default fallback 0)
-  - `authors_text` (nullable)
-  - `venue_text` (nullable)
-  - `title_url` (nullable)
-- Page-level parse status enum:
-  - `ok`, `no_results`, `blocked_or_captcha`, `layout_changed`, `network_error`
-- Page-level flags:
-  - `has_show_more_button`
-  - `articles_range` string (e.g. `Articles 1–20`)
-
-## Immediate test plan unlocked by this probe
-
-- Add fixture-driven unit tests using captured HTML from `fixtures/run_20260216T182334Z`.
-- Add assertions for:
-  - row count > 0 on known-good fixtures
-  - profile name extraction
-  - cluster_id extraction from title URLs
-  - nullable handling for year/venue
-  - blocked/inaccessible page classification
-  - show-more partial warning classification
-
-## Phase 1 implementation guardrails
-
-- Keep requests low-rate with jitter and explicit timeout.
-- Persist parser status per scholar run.
-- If `has_show_more_button=True`, mark run as partial and show this in UI.
-- Never fail entire run because one scholar page is blocked or malformed.
-