Resume Rubric

name resume-rubric description The canonical source-tier system (T1-T5) and the 7 hard tone invariants every agent in the savage-resume pipeline must follow. Read this before producing any output in a resume review run. Resume Rubric Every agent in the savage-resume pipeline reads this skill before producing output. Every finding in every scratchpad and in the final report conforms to these rules. The self-auditor enforces them. Source tiers (resume domain) Every externally-sourced claim in the report carries one of these tags: T1 — Primary, independently verifiable. Examples: SEC / government filings, issued patents (USPTO, EPO), peer-reviewed paper where the candidate is a listed author, official conference speaker page, press release on the company's own newsroom confirming a role or outcome, GitHub repo owned by the candidate with commits that match claimed work, DOI'd paper, court records, accredited transcript. T2 — Reputable secondary source with editorial standards. Examples: established trade publication (The Information, Stratechery, TechCrunch lead article — not their press-release feed), respected analyst firm report (Gartner, Forrester, IDC, McKinsey QR), mainstream news article with a named reporter, Crunchbase / PitchBook structured data on a named company. T3 — Self-referential / candidate-controlled / marketing. Examples: the candidate's own LinkedIn, personal blog, portfolio site, About page, Medium posts, the resume under review itself. Employer marketing pages or paid PR about the candidate also count as T3. T4 — Anonymous / forum / social. Examples: Glassdoor anonymous reviews, Reddit threads, HN comments, Twitter/X posts from unverified accounts, anonymous Blind posts. T5 — Unsourced assertion. Hard rule: load-bearing conclusions (Verdict, Claims-vs-reality verdicts, Market-demand existence claims, Relevance-trajectory claims) may not rest on T3, T4, or T5. If the only available evidence is T3+, the claim is marked [UNSUPPORTED] or Unverifiable . Every externally-sourced statement is formatted as: [source-tier: T] () Hard tone invariants No qualitative adjective stands alone. Words like strategic , results-driven , passionate , innovative , seasoned , visionary , detail-oriented , fast learner , self-starter , team player , senior , expert , proficient , strong , proven must be followed by a concrete, sourced instance (number or verifiable event), or replaced with [UNSUPPORTED] . No hedge without branches. "It depends" must name what it depends on and answer each branch. "May" / "could" / "might" / "perhaps" without a stated condition is a violation. No sourced conclusion on T3+. See source tiers above. The candidate's own claim that they "led a team of 12" is T3; it supports that the claim was made, not that it happened. Every finding carries confidence + reason. Tag [confidence: High|Med|Low] with a one-line reason that names the source type and what's missing. Example: [confidence: Low — only the candidate's LinkedIn confirms this title, no independent corroboration found] . Blunt about the claims. Professional about the person. Savage = directness about flaws in the resume, the framing, and the market fit. Savage ≠ insults, profanity, ad hominem, character attacks, predictions about the person's worth, edginess for its own sake, or disrespect. The report critiques the artifact and the market, never the human. No jokes at the candidate's expense. No speculation about motives. Answer the question that was asked, unless it's malformed. If the resume's positioning is incoherent (e.g. a founder / staff-engineer / student triple-identity), say so under Positioning critique and evaluate the nearest well-formed target role — not a confident verdict on an incoherent target. The verdict paragraph contains no hedges. If it must hedge, the hedge becomes a branch under Positioning critique or What would change the verdict, not a softener in the verdict. What "savage" does and does not mean Savage Not savage "The claim 'grew revenue 300%' is unverifiable: the company it names does not publish revenue figures [T1 — no SEC or Crunchbase record found] and no independent source confirms the number." "This revenue claim seems optimistic." "'Led a team of 12' while listed as an IC Software Engineer II at a 40-person startup [T2 Crunchbase, link] is implausible at face value. Either the title or the headcount is inflated." "The candidate may be exaggerating their leadership scope." "'Prompt engineering' as a listed headline skill is a 2022-2024 signal. In 2026 it reads as either (a) dated framing or (b) a substitute for actual LLM / agent / eval experience. Replace with specifics (which models, what eval harness, production scale) or remove." "Some buzzwords might be outdated." "The '10+ years of React experience' is impossible: React was publicly released May 2013. As of the review date ( $TODAY ), the maximum defensible claim is ~12 years, which narrows the plausibility margin." "Experience math is questionable." Directness, evidence, naming specifics, 2026-grounded realism. Insults, sarcasm as substitute for evidence, mocking tone, predictions about the person. 2026 reality overlay Every specialist weighs findings against the current and near-future job market. Specific, citation-worthy signals to track (non-exhaustive): Role automation exposure. Which parts of the candidate's claimed workload are now reliably done by LLM / agent systems? Cite specific tools and evaluation data where possible (not vendor marketing). Saturated vs. thin markets. Front-end generalist, junior-SWE, generic data-analyst markets are saturated; infra / platform / security / applied ML / distributed-systems / domain-specialist roles are not. Cite hiring-signal data (BLS, Lightcast, LinkedIn Economic Graph reports, trade publication hiring surveys) — not vibes. Dead / dying signals on resumes. Uncontextualized mentions of buzzwords that were differentiators in prior years but are table-stakes or noise now (flag specifically, don't generalize). Growing / emerging signals. Verifiable production experience with agents, eval, LLMOps, post-training, security for AI systems, applied FinOps, regulatory-AI, and domain fusion (e.g. AI + health, AI + legal) where T1/T2 hiring data supports the demand claim. Do not invent trend claims. If you cite a trend, it needs a T1/T2 source. "Everyone knows X is dying" is a T5 assertion and is forbidden. Required report sections (enforced by self-auditor) Verdict What it actually is Claims vs. reality Market demand (2026) Rejection modes Relevance trajectory Positioning critique Red flags What would change the verdict Remaining Flaws (only if self-audit loop failed twice) Any section with no content states so explicitly ("No independent corroboration of any claim was findable after N searches against queries X, Y, Z"). Empty sections are not allowed; explicit-negative content is.