Improve Ask Mantosh citation precision and deep-link evidence navigation

Priority: Medium

Problem

Ask Mantosh links answers to approved public sources, but a visitor may still need to scan a long article, project case study, or experience page to locate the exact evidence supporting a claim. For recruiter-facing use, the path from answer to proof should be as direct as practical.

Goal

Improve citation precision so meaningful Ask Mantosh claims lead visitors to the most relevant public evidence location rather than only to the top of a source page.

Scope

• Review representative recruiter, project, skills, ownership, and experience answers for citation granularity.
• Identify source documents where stable section anchors or other deep links would materially reduce evidence-finding time.
• Define a safe metadata model for section-level source targets without allowing the model to invent anchors or arbitrary URLs.
• Preserve canonical approved-source validation and unsupported-answer suppression.
• Ensure citations remain understandable when a source does not have an appropriate deep link; page-level citations are acceptable when they are genuinely the best evidence target.
• Add deterministic tests for valid/invalid deep-link behavior and source normalization.

Acceptance criteria

• Representative Ask Mantosh claims are audited for evidence-navigation precision.
• High-value sources expose stable deep links where useful.
• Ask Mantosh can return only approved, deterministic section/page targets; generated arbitrary anchors are not trusted.
• Existing citation security and evidence boundaries are preserved.
• Recruiter-style answers can reach supporting evidence with materially less scanning where deep links are available.
• Citation/deep-link behavior is covered by automated tests.

Start by locating the Ask Mantosh citation and source-metadata entry points, along with existing citation validation and test entry points; review representative recruiter, project, skills, ownership, and experience answers. Done means approved deterministic deep-link behavior is defined and tested, while existing source validation and unsupported-answer safeguards remain intact.