feat: deterministic LLM-free ATS pipeline (extraction, rewrite, scoring)

fix: fair symmetric stuffing penalty + more extraction retries

fix: reliable live extraction (nemotron) + fair PDF-based before/after

feat: live V1 acceptance — independent audit, 2-parser PDF, summary opt, completeness

feat: V1 NIM fallback + 90% gate + PDF-scored alignment + correction pass

feat: V1 evidence-backed optimization — rewriting, calibration, scoring

fix: evidence-gated ATS pipeline — stop scraped-page contamination & fabrication

feat: LLM keyword extraction via Calibrated Keyword Match Framework

fix: V1 places ALL keywords in exp+project bullets, no Core Competencies

feat: inject 2 keyword items per project list-end before Core Competencies

fix: Core Competencies trusted-only + duplicate word-boundary check

fix: natural keyword placement + fix comma-split + relax V1 frequency filter

fix: always allow PDF download for BELOW_TARGET_REPAIRABLE

fix: V1 keyword quality gate, layout fix, progress anim; add V1+V2 e2e test

feat: V1 noise fix, real progress animation via SSE

fix: drop naked keyword bullets + always re-extract on job switch

perf+fix: parallelize V2 curate+rank; fix extension overlay covering page

feat(r36): keyword_placer wired into V2 pipeline — bullet-level placement guidance

feat: extract up to 100 keywords + match real-checker breadth (Jobalytics gap)

fix: order curated keywords by importance so the 55-cap keeps top keywords, not first 55

fix: broaden curated keyword extraction to match real checkers (Jobalytics 52% gap)

feat: coverage backstop — guaranteed >=90% ATS in Maximum ATS Mode (R35)

feat(phase-15): LLM-curated scoring denominator (generalizes) + weave-to-90 + fixes (R34)

feat: make V2 the default generation mode across all surfaces

feat(phase-14): reliable honest 90%+ ATS — skill-filter + weave-to-target + parseability (R33)

feat(phase-13): atomic ATS scoring + coverage-aware weave + feed isolation (R32)

feat(phase-12e): V2 JD sanitizer — strip scraped job-board page chrome (R31)

feat(phase-12d): narrow fit-gate — no fabricating specialized domains

fix(phase-12c): clean V2 ATS scoring denominator — count genuine keywords only

feat(phase-12b): multi-agent keyword ranking in V2 (R29)

feat(phase-12): V2 reworked into true multi-agent natural integration

feat(phase-11): kill JD-scraping leak in keyword pool — V2-only

fix: designation → Associate Product Manager, filename → Saiteja_Tirunagari_<Company>_Resume.pdf

feat(phase-10): V2 natural sentence resume mode — Kimi LLM generates sentences, same V1 waterfall

fix: server-side JD extraction takes the job description only (all platforms)

feat: telegram egress diag endpoint + proxy support for Bot API

fix: Telegram bot force IPv4 to api.telegram.org + send retries

feat: Telegram bot touchpoint (job link or JD -> tailored resume PDF)

fix: Tectonic sanitizer regex must require uppercase after fa

fix: sanitize fontawesome OTF for Tectonic + dedupe keywords (no duplicates)

feat(09-03): structured keyword placement for hardcoded resume [R22]

feat(09-02): uncapped JD keyword extraction, honesty gate kept [R21]

feat(09-01): hardcoded default resume for HF + extension [R20]

fix: extension JD extraction (LinkedIn collections) + 0% coverage on custom-macro LaTeX

fix: extension LaTeX compile timeout 120s -> 420s

feat: persistent login + resume-from-left-off + start-over button

feat: Supabase login gate + persistent resume + generated LaTeX storage

feat: Scrapling anti-block fetch layer + Direct Company ATS source

fix: pinned musl Tectonic + guaranteed PDF fallback + panel/popup sync (v1.5.2)

fix: Tectonic compile + JD junk-guard from live test (v1.5.1)