First deployment
Some checks failed
quality-gates / lint-and-test (push) Has been cancelled
quality-gates / security-scan (push) Has been cancelled

This commit is contained in:
2026-02-13 09:14:04 -05:00
parent 0e21e035f5
commit 679561bcdb
128 changed files with 3479 additions and 120 deletions

View File

@@ -0,0 +1,2 @@
schema: spec-driven
created: 2026-02-13

View File

@@ -0,0 +1,100 @@
## Context
This change addresses unresolved regressions introduced by recent UX and content pipeline updates. The affected areas cross frontend interaction patterns (`frontend/index.html`), policy content delivery (`backend/main.py`), translation generation (`backend/news_service.py`), and admin image maintenance operations (`backend/cli.py`). Current behavior still allows low-quality translations and repeated or weakly related refetched images, and policy links still rely on full-page navigation.
Key constraints:
- Maintain existing OpenSpec capability contracts where possible and ship mostly as requirement deltas.
- Preserve permalink identity for machine/admin targeting, while removing redundant user-facing permalink chrome on hero.
- Keep deterministic fallback behavior (AI-themed image fallback and English fallback for failed translations).
## Goals / Non-Goals
**Goals:**
- Deliver modal-based Terms and Attribution access with keyboard-safe behavior.
- Remove visible hero permalink affordance while preserving internal permalink targeting.
- Improve Tamil/Malayalam hero readability on desktop and mobile.
- Switch copy and back-to-top controls to icon-first interaction while preserving accessibility labels.
- Add translation quality validation gates (wrong-language/gibberish rejection and deterministic fallback).
- Extend refetch-images to support permalink targeting and enforce alternative relevant image selection (non-repeat, relevance/safety filters, AI fallback when uncertain).
**Non-Goals:**
- Replacing the existing rendering framework or routing model.
- Introducing new external image providers in this change.
- Reworking full article ingestion architecture beyond targeted translation/image guardrails.
## Decisions
### Decision 1: Policy disclosure moves to modal-first UI with deep-link-safe state
Use in-page modals for Terms and Attribution content while supporting deterministic open/close state from URL params or equivalent state synchronization.
Alternatives considered:
- Keep route pages only: rejected due to UX friction and context switch from main feed.
- Embed truncated inline footer text: rejected due to disclosure readability and legal clarity risks.
### Decision 2: Hero permalink identity remains internal, not a primary visual affordance
Retain permalink generation/parsing helpers for deep links and admin targeting, but remove visible hero permalink action from hero chrome.
Alternatives considered:
- Remove permalink behavior entirely: rejected because admin targeting and deep-link tooling need stable identifiers.
- Keep visible permalink: rejected because current hero action density and readability are degraded.
### Decision 3: Locale-aware hero readability profile for Tamil/Malayalam
Apply stronger readability guardrails for non-English hero text: increased contrast treatment, safer line-wrapping, language-specific typography tuning.
Alternatives considered:
- Single style for all languages: rejected; longer glyph clusters and script density degrade legibility.
- Separate locale-specific templates: rejected as too heavy for this stabilization change.
### Decision 4: Icon-only controls must remain explicitly accessible
Use icon-based copy and back-to-top controls with accessible names, keyboard operation, and stable tap targets.
Alternatives considered:
- Text-only controls: rejected by updated UX requirement.
- Icon-only without explicit labels: rejected due to WCAG and discoverability concerns.
### Decision 5: Translation quality gate before persistence/serving
Add post-generation validation for expected language/script sanity and basic gibberish detection. If validation fails, mark translation unavailable and serve deterministic English fallback.
Alternatives considered:
- Blindly accept model output: rejected due to current wrong-language/gibberish incidents.
- Human moderation for all translations: rejected for runtime latency and operational overhead.
### Decision 6: Refetch image selection becomes candidate-based and non-repeat
For refetch operations, generate contextual query candidates, filter against relevance/safety constraints, reject current/previously used image matches for the same article, and fall back to AI-themed image when confidence is low or candidates are exhausted.
Alternatives considered:
- First-provider-first-image approach: rejected because it often repeats or returns weakly related imagery.
- Hard blocklist only: rejected; insufficient for relevance and dedupe guarantees.
### Decision 7: Admin refetch supports permalink-targeted execution
Extend admin command contract to accept a permalink input, resolve to article identity, and run the same queue/refetch pipeline with targeted scope.
Alternatives considered:
- Keep limit-based batch-only refetch: rejected; operators need deterministic single-article correction path.
- Add separate ad hoc script: rejected due to fragmented operational surface.
## Risks / Trade-offs
- **[Risk] Modal policy flow may regress keyboard/focus behavior** -> Mitigation: explicit focus trap, escape-close, and focus-return requirements.
- **[Risk] Translation validation may over-reject valid short outputs** -> Mitigation: bounded heuristic thresholds plus deterministic English fallback and logging for tuning.
- **[Risk] Stronger image filtering can reduce image diversity** -> Mitigation: multi-candidate ranking and fallback to AI-themed image instead of unrelated imagery.
- **[Risk] Permalink-targeted admin flow may fail on malformed or stale links** -> Mitigation: strict permalink parser and clear operator error summaries.
## Migration Plan
1. Introduce modal policy behavior and hero/readability/icon control updates in frontend.
2. Add translation quality gate logic and fallback outcomes in backend translation path.
3. Extend image refetch pipeline with candidate ranking, dedupe, relevance/safety filters, and fallback.
4. Add permalink-targeted admin refetch argument and resolution path.
5. Validate through existing/manual verification paths and update OpenSpec tasks.
Rollback:
- Revert p16 frontend modal/icon/readability changes.
- Revert translation gate checks to prior translation acceptance path.
- Revert permalink-targeted and dedupe-aware refetch logic to prior batch refetch behavior.
## Open Questions
- Should policy modals update canonical URL state (`?modal=terms`) for shareable deep links, or stay ephemeral-only?
- What minimum confidence/sanity threshold should trigger translation rejection for short headlines?
- Should image dedupe history persist only per article current/previous image, or maintain wider history window?

View File

@@ -0,0 +1,46 @@
## Why
Recent UX and content-quality updates still leave several user-facing regressions unresolved across policy access, hero readability, share/back-to-top controls, translation quality, and image refetch behavior. These issues directly affect trust and usability, so we need a focused stabilization change that tightens both frontend interactions and backend content-quality guardrails.
## What Changes
- Convert Terms of Use and Attribution access from full-page navigation to in-page modal dialogs while preserving clear disclosure content and keyboard accessibility.
- Remove hero-level permalink affordance from visible hero chrome.
- Improve hero readability for non-English content (Tamil/Malayalam) across desktop and mobile.
- Replace text-based copy/share and back-to-top controls with icon-based controls that remain accessible.
- Add translation quality validation to reduce wrong-language or gibberish outputs before serving/storing translated content.
- Upgrade `admin refetch-images` behavior so refreshed images are alternative (not same as current), subject-relevant, and filtered against unrelated animal/pet outcomes.
- Extend admin image refetch operations to accept a permalink target and fetch a new relevant image for that article.
- Keep deterministic AI-themed fallback behavior when confidence is low or relevance checks fail.
## Capabilities
### New Capabilities
- `policy-disclosure-modals`: In-page modal experience for Terms of Use and Attribution content (focus trap, escape close, deep-link-safe modal state).
- `translation-quality-validation-gates`: Post-translation validation gates (language/script sanity + gibberish rejection + deterministic fallback policy).
- `permalink-targeted-image-refetch`: Admin command support for refetching summary images by permalink target.
- `alternative-image-selection-and-dedupe`: Refetch pipeline guarantees alternative image selection, relevance scoring, repeat-image avoidance, and unsafe/unrelated-image filtering.
### Modified Capabilities
- `footer-policy-links`: Footer policy access behavior changes from route navigation emphasis to modal activation flow.
- `terms-of-use-risk-disclosure`: Terms content remains required, but delivery surface changes from standalone page-centric flow to modal-capable flow.
- `attribution-disclaimer-page`: Attribution content remains required, but delivery surface changes from standalone page-centric flow to modal-capable flow.
- `hero-display`: Hero chrome removes visible permalink affordance and preserves clean primary/secondary actions.
- `hero-summary-entry-and-readability`: Strengthen multilingual hero readability requirements, especially for Tamil/Malayalam headline and summary rendering.
- `responsive-device-agnostic-layout`: Ensure icon-based controls and modal policy surfaces remain usable and unclipped on mobile/tablet/desktop.
- `wcag-2-2-aa-accessibility`: Icon-only controls and policy modals require explicit accessible labels, keyboard navigation, and focus behavior.
- `article-translations-ml-tm`: Add quality validation outcomes for generated Tamil/Malayalam translations before persistence/serving.
- `language-aware-content-delivery`: Define deterministic fallback behavior when requested translation fails quality validation.
- `admin-maintenance-command-suite`: Extend admin command contract to include permalink-targeted image refetch execution path.
- `queued-image-refetch-with-backoff`: Extend queue behavior to support targeted permalink jobs and non-repeat image outcomes.
- `context-aware-image-selection-recovery`: Improve subject extraction and keyword recovery for alternative relevant image retrieval.
- `news-image-relevance-and-fallbacks`: Tighten relevance/safety checks to reject unrelated animal/pet imagery and preserve AI-themed fallback when uncertain.
## Impact
- **Frontend/UI**: `frontend/index.html` policy-link behavior, hero metadata/actions, icon controls, modal state handling, and locale-specific readability styles.
- **Backend API/Routes**: `backend/main.py` policy content delivery strategy and compatibility with modal-first access.
- **Admin CLI**: `backend/cli.py` image refetch command arguments and permalink-targeted workflow.
- **Image pipeline**: `backend/news_service.py` subject extraction, alternative image candidate selection, dedupe/repeat prevention, relevance/safety filtering, and fallback selection.
- **Translation pipeline**: `backend/news_service.py` translation response validation and fallback policy; potential metadata persistence updates in repository/model layers.
- **Operational behavior**: Additional logging/audit fields for translation-quality failures and image-refetch decision path to support debugging and trust.

View File

@@ -0,0 +1,9 @@
## MODIFIED Requirements
### Requirement: Unified admin command surface
The system SHALL provide a single admin CLI command family exposing maintenance subcommands, including permalink-targeted image refetch support.
#### Scenario: Subcommand discovery
- **WHEN** an operator runs the admin command help output
- **THEN** available subcommands include refetch-images, clean-archive, clear-cache, clear-news, rebuild-site, regenerate-translations, and fetch
- **AND** refetch-images help output documents optional permalink-targeted execution

View File

@@ -0,0 +1,25 @@
## ADDED Requirements
### Requirement: Refetch selects an alternative image for target article
Refetch operations SHALL select an image different from the article's current summary image when a usable alternative exists.
#### Scenario: Current image is excluded from candidate result
- **WHEN** refetch evaluates image candidates for an article
- **THEN** the system rejects candidates matching the current summary image identity
- **AND** stores a different accepted image candidate when available
### Requirement: Candidate filtering enforces topical relevance and safety
Refetch candidate selection SHALL reject clearly unrelated imagery classes while preserving topic relevance.
#### Scenario: Unrelated animal/pet candidate is rejected
- **WHEN** candidate metadata or source signals indicate unrelated animal/pet imagery
- **THEN** the system rejects that candidate for article refetch
- **AND** continues evaluating other candidates or fallback paths
### Requirement: Low-confidence selection falls back to AI-themed default
If no candidate satisfies relevance and safety thresholds, the system SHALL use deterministic AI-themed fallback imagery.
#### Scenario: All candidates rejected
- **WHEN** candidate evaluation exhausts without an accepted image
- **THEN** the system assigns configured AI-themed fallback image
- **AND** records fallback decision path in command output/logs

View File

@@ -0,0 +1,19 @@
## MODIFIED Requirements
### Requirement: System generates Tamil and Malayalam translations at article creation time
The system SHALL generate Tamil (`ta`) and Malayalam (`ml`) translations for each newly created article during ingestion and validate translation quality before persistence.
#### Scenario: Translation generation for new article
- **WHEN** a new source article is accepted for storage
- **THEN** the system requests Tamil and Malayalam translations for headline and summary
- **AND** translation generation occurs in the same ingestion flow for that article
#### Scenario: Translation failure fallback
- **WHEN** translation generation fails for one or both target languages
- **THEN** the system stores the base article in English
- **AND** marks missing translations as unavailable without failing the whole ingestion cycle
#### Scenario: Translation quality validation failure
- **WHEN** generated translation fails language/script sanity or gibberish validation checks
- **THEN** the system does not persist invalid translation content for that article-language pair
- **AND** records validation failure outcome for diagnostics

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Attribution page discloses AI generation and non-ownership
The system SHALL provide Attribution disclosure content with explicit statements that content is AI-generated and not personally authored by the site owner, including modal-based access from the landing experience.
#### Scenario: Attribution disclosure title and body content
- **WHEN** a user opens Attribution disclosure content from supported entry points
- **THEN** the title clearly indicates attribution/disclaimer purpose
- **AND** the body states content is AI-generated and not generated by the owner as an individual
#### Scenario: Attribution disclosure includes non-involvement statement
- **WHEN** a user reads Attribution details
- **THEN** the content explicitly states owner non-involvement in generated content claims
- **AND** wording appears in primary readable content area

View File

@@ -0,0 +1,17 @@
## MODIFIED Requirements
### Requirement: Context-aware image query generation
Image refetch SHALL construct provider queries from article context by identifying main subject keywords from headline and summary content.
#### Scenario: Context-enriched query
- **WHEN** a queued article is processed for image refetch
- **THEN** the system derives main-subject query terms from article headline and summary
- **AND** includes relevance-supporting cues for improved candidate quality
### Requirement: AI-domain fallback keywords
When context extraction is insufficient or confidence is low, the system SHALL use AI-domain fallback keywords.
#### Scenario: Empty or weak context extraction
- **WHEN** extracted context terms are empty or below quality threshold
- **THEN** the system applies fallback terms such as `ai`, `machine learning`, `deep learning`
- **AND** continues candidate search with deterministic fallback ordering

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Footer exposes policy navigation links
The system SHALL display footer controls for Terms of Use and Attribution on the landing page and open their disclosure content in in-page modal dialogs.
#### Scenario: Footer policy controls visible and focusable
- **WHEN** a user loads the main page
- **THEN** the footer includes controls labeled "Terms of Use" and "Attribution"
- **AND** controls are visually distinguishable and keyboard focusable
#### Scenario: Footer policy controls open modal disclosures
- **WHEN** a user activates either policy control
- **THEN** the system opens the corresponding policy modal
- **AND** activation succeeds without full-page navigation dependency

View File

@@ -0,0 +1,9 @@
## MODIFIED Requirements
### Requirement: Hero block display
The system SHALL display hero article actions without a visible permalink affordance while preserving primary summary and source attribution interactions.
#### Scenario: Hero action chrome excludes permalink control
- **WHEN** the hero block is rendered
- **THEN** no dedicated hero permalink control is displayed
- **AND** summary open and source-link actions remain available

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Hero metadata readability over images
Hero metadata (`LATEST`, relative time, headline, and summary) SHALL remain visually legible across bright and dark images on desktop and mobile, including Tamil and Malayalam rendering paths.
#### Scenario: Bright image background
- **WHEN** the hero image contains bright regions under metadata text
- **THEN** overlay and text styles preserve readable contrast for metadata and headline blocks
- **AND** readability remains stable for English, Tamil, and Malayalam content
#### Scenario: Mobile viewport readability
- **WHEN** the hero renders on a mobile viewport
- **THEN** metadata and title remain readable without overlapping controls or clipping
- **AND** Tamil and Malayalam typography remains legible at mobile breakpoints

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Language fallback to English is deterministic
The system SHALL return English source content when the requested translation is unavailable or fails translation quality validation.
#### Scenario: Missing translation fallback
- **WHEN** a client requests Tamil or Malayalam content for an article lacking that translation
- **THEN** the system returns the English headline and summary for that article
- **AND** response shape remains consistent with language-aware responses
#### Scenario: Invalid translation fallback
- **WHEN** a client requests Tamil or Malayalam content for an article whose translation failed quality validation
- **THEN** the system returns English headline and summary for that article
- **AND** avoids returning invalid translated output

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Fallback behavior remains context-aware first
The system SHALL evaluate context-aware candidates before fallback and require refetched summary images to be relevant, non-redundant alternatives.
#### Scenario: Context-aware attempt precedes fallback
- **WHEN** summary image selection runs for a news item
- **THEN** the system first attempts provider queries from extracted context-aware keywords
- **AND** only falls back to generic AI image if candidate evaluation fails
#### Scenario: Refetch rejects unrelated and duplicate outcomes
- **WHEN** candidate images are evaluated during refetch
- **THEN** the system rejects candidates matching current image identity for the same article
- **AND** rejects clearly unrelated animal/pet imagery classes before final selection

View File

@@ -0,0 +1,14 @@
## ADDED Requirements
### Requirement: Admin image refetch supports permalink targeting
The admin image refetch command SHALL support a permalink-targeted execution mode for deterministic single-article remediation.
#### Scenario: Refetch by permalink
- **WHEN** an operator runs refetch-images with a valid permalink target
- **THEN** the system resolves permalink identity to the matching article
- **AND** executes image refetch for that target without requiring batch mode
#### Scenario: Invalid permalink target fails clearly
- **WHEN** an operator provides a malformed or unresolved permalink target
- **THEN** the command exits with actionable error output
- **AND** no article image is modified

View File

@@ -0,0 +1,27 @@
## ADDED Requirements
### Requirement: Policy disclosures are available as in-page modals
The system SHALL present Terms of Use and Attribution content in in-page modal dialogs without requiring full-page navigation from the landing experience.
#### Scenario: Open Terms modal from footer
- **WHEN** a user activates the "Terms of Use" footer control
- **THEN** a Terms modal opens in place on the current page
- **AND** the underlying page context remains intact
#### Scenario: Open Attribution modal from footer
- **WHEN** a user activates the "Attribution" footer control
- **THEN** an Attribution modal opens in place on the current page
- **AND** disclosure content is readable in the modal body
### Requirement: Policy modals are keyboard-safe and dismissible
Policy disclosure modals SHALL provide deterministic keyboard and pointer dismissal behavior.
#### Scenario: Escape closes active policy modal
- **WHEN** a policy modal is open and the user presses `Escape`
- **THEN** the modal closes
- **AND** focus returns to the triggering control
#### Scenario: Modal focus remains trapped while open
- **WHEN** a keyboard-only user tabs while a policy modal is open
- **THEN** focus cycles within modal interactive controls
- **AND** focus does not escape to background content

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Latest-30 queue construction
The refetch-images command SHALL enqueue up to the latest 30 news items for processing in batch mode and support targeted single-article mode when permalink targeting is provided.
#### Scenario: Queue population
- **WHEN** refetch-images is started without a permalink target
- **THEN** the command loads recent news items
- **AND** enqueues at most 30 items ordered from newest to oldest
#### Scenario: Targeted permalink mode
- **WHEN** refetch-images is started with a valid permalink target
- **THEN** the command enqueues only the resolved article
- **AND** bypasses latest-30 queue expansion

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Core layout is device-agnostic and responsive
The system SHALL render key surfaces (header, hero, feed, modal, footer, policy modals, and floating controls) responsively across mobile, tablet, and desktop viewports.
#### Scenario: Mobile layout behavior
- **WHEN** a user opens the site on a mobile viewport
- **THEN** content remains readable without horizontal overflow
- **AND** interactive controls including icon-only copy/back-to-top and policy modals remain reachable and usable
#### Scenario: Desktop and tablet adaptation
- **WHEN** a user opens the site on tablet or desktop viewports
- **THEN** layout reflows according to breakpoint design rules
- **AND** no key content or controls are clipped

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Terms page states unverified-content risk
The system SHALL provide Terms of Use disclosure content that states information is unverified and use is at the user's own risk, including modal-based access from the landing experience.
#### Scenario: Terms disclosure risk statement visible
- **WHEN** a user opens Terms of Use disclosure content from its supported entry points
- **THEN** the content includes clear at-own-risk usage language
- **AND** the content states information is not independently verified
#### Scenario: Terms disclosure references source uncertainty
- **WHEN** a user reads Terms details
- **THEN** the content explains information is surfaced from external/AI-generated sources
- **AND** users are informed responsibility remains with their own decisions

View File

@@ -0,0 +1,22 @@
## ADDED Requirements
### Requirement: Translation output passes quality validation before use
The system SHALL validate generated Tamil and Malayalam translation output for language/script sanity and gibberish risk before persistence or delivery.
#### Scenario: Valid translation accepted
- **WHEN** generated translation passes configured language and quality checks
- **THEN** the system stores and serves the translated content
- **AND** records a successful validation outcome
#### Scenario: Invalid translation rejected
- **WHEN** generated translation fails language/script or gibberish checks
- **THEN** the system rejects translated output for that article-language pair
- **AND** records a validation failure reason for operations visibility
### Requirement: Validation failure fallback is deterministic
When translation quality validation fails, the system SHALL provide deterministic English fallback behavior.
#### Scenario: Failed translation falls back to English
- **WHEN** a client requests language content for an article whose translation failed validation
- **THEN** the system returns English source headline and summary
- **AND** response shape remains consistent with language-aware responses

View File

@@ -0,0 +1,14 @@
## MODIFIED Requirements
### Requirement: Core user flows comply with WCAG 2.2 AA baseline
The system SHALL meet WCAG 2.2 AA accessibility requirements for primary interactions and content presentation, including icon-only controls and policy-disclosure modals.
#### Scenario: Keyboard-only interaction flow
- **WHEN** a keyboard-only user navigates policy modals and icon-only controls
- **THEN** all primary interactive elements remain reachable and operable
- **AND** visible focus indication is present at each step
#### Scenario: Contrast and non-text alternatives
- **WHEN** users consume text and non-text UI content
- **THEN** color contrast meets AA thresholds for relevant text and controls
- **AND** icon-only controls expose meaningful accessible labels

View File

@@ -0,0 +1,43 @@
## 1. Policy Disclosure Modal Conversion
- [x] 1.1 Implement Terms of Use modal surface and wire footer control to open it in-page.
- [x] 1.2 Implement Attribution modal surface and wire footer control to open it in-page.
- [x] 1.3 Add modal focus trap, Escape-close, and focus-return behavior for policy dialogs.
- [x] 1.4 Ensure policy modal state handling is deep-link-safe and does not break existing navigation.
## 2. Hero and Interaction Chrome Cleanup
- [x] 2.1 Remove visible hero permalink affordance while preserving hero summary and source actions.
- [x] 2.2 Replace modal copy-link text control with icon-only button and accessible labels.
- [x] 2.3 Replace floating back-to-top text control with icon-only button and accessible labels.
## 3. Multilingual Hero Readability Hardening
- [x] 3.1 Improve Tamil/Malayalam hero headline and summary readability across desktop and mobile.
- [x] 3.2 Tune hero overlay/contrast behavior to preserve legibility over bright image regions.
- [x] 3.3 Verify no clipping/overlap regressions in hero metadata and controls on narrow viewports.
## 4. Translation Quality Validation Gates
- [x] 4.1 Add translation quality validation checks for language/script sanity and gibberish rejection.
- [x] 4.2 Persist/log translation validation outcomes for observability and debugging.
- [x] 4.3 Enforce deterministic English fallback when requested translation fails validation.
- [x] 4.4 Prevent invalid translation variants from being served as authoritative localized content.
## 5. Refetch Image Quality and Targeting Enhancements
- [x] 5.1 Extend admin refetch-images to accept permalink-targeted execution.
- [x] 5.2 Resolve permalink targets to article identity with actionable operator error output.
- [x] 5.3 Implement candidate-based refetch selection that prefers subject-relevant alternatives.
- [x] 5.4 Reject duplicate current-image outcomes for same-article refetch operations.
- [x] 5.5 Reject clearly unrelated animal/pet imagery during candidate filtering.
- [x] 5.6 Preserve deterministic AI-themed fallback when no acceptable candidate exists.
- [x] 5.7 Emit progress/decision-path output for targeted and batch refetch execution.
## 6. Verification
- [x] 6.1 Verify Terms and Attribution open as accessible modals from footer controls.
- [x] 6.2 Verify hero permalink affordance is removed and core hero actions still function.
- [x] 6.3 Verify Tamil/Malayalam hero readability on desktop and mobile.
- [x] 6.4 Verify translation-quality gate fallback behavior for wrong-language/gibberish outputs.
- [x] 6.5 Verify permalink-targeted refetch returns alternative relevant image and avoids pet-style mismatches.