job-forge 2.0.0

package/modes/offer.md

@@ -0,0 +1,199 @@
+# Mode: offer — Full Evaluation A-F
+
+When the candidate pastes an offer (text or URL), ALWAYS deliver all 6 blocks:
+
+## Emit-once rule — REQUIRED
+
+Before writing any of the blocks below, **emit the Score JSON block first** (per `_shared.md` → "Score Emission — EMIT-ONCE JSON"). The blocks that follow reference the JSON's score keys and rationales; they do NOT re-enumerate the 10 dimensions.
+
+Workflow:
+
+1. Detect archetype (Step 0 below).
+2. `cv.md` is already in your context via `opencode.json:instructions` — **do NOT Read it again**. If you need detailed proof points beyond cv.md, Read `article-digest.md` (if it exists) — once, not per block.
+3. Decide the 10 scores. **Do not narrate this process in thinking.** Write the scores straight into the JSON block.
+4. Emit the JSON block exactly once.
+5. Then write Blocks A-F referencing the JSON — not parallel to it.
+
+Re-scoring mid-report is banned. If you change your mind on a dimension, update the JSON and regenerate the prose from it — don't keep two copies in sync.
+
+## Step 0 — Archetype Detection
+
+Classify the offer into one of the 6 archetypes (see `_shared.md`). If it's a hybrid, indicate the 2 closest ones. This determines:
+- Which proof points to prioritize in block B
+- How to rewrite the summary in block E
+- Which STAR stories to prepare in block F
+
+## Block A — Role Summary
+
+Build a table with these rows.
+
+- Detected archetype.
+- Domain (platform/agentic/LLMOps/ML/enterprise).
+- Function (build/consult/manage/deploy).
+- Seniority.
+- Remote (full/hybrid/onsite).
+- Team size (if mentioned).
+- TL;DR in 1 sentence.
+
+## Match the CV (Block B)
+
+**cv.md is already in your context** (via opencode.json:instructions) — use it directly, don't Read it again. Build a table mapping each JD requirement to exact lines from the CV.
+
+**Adapt the proof points to the archetype.**
+
+- For FDE: prioritize proof points about fast delivery and client-facing work.
+- For SA: prioritize system design and integrations.
+- For PM: prioritize product discovery and metrics.
+- For LLMOps: prioritize evals, observability, pipelines.
+- For Agentic: prioritize multi-agent, HITL, orchestration.
+- For Transformation: prioritize change management, adoption, scaling.
+
+**Gaps** section with a mitigation strategy for each one. For each gap, answer in order.
+
+1. Classify it as a hard blocker or an optional requirement.
+2. Identify any adjacent experience the candidate can demonstrate.
+3. Identify any portfolio project that covers the gap.
+4. Write a concrete mitigation plan (a cover-letter phrase or a quick project).
+
+## Compute level and strategy (Block C)
+
+1. **Compare detected level** in the JD against the candidate's calibrated level for that archetype.
+2. **"Sell senior without lying" plan**: specific phrases adapted to the archetype, concrete achievements to highlight, how to position founder experience as an advantage.
+3. **"If I get downleveled" plan**: accept if comp is fair, negotiate 6-month review, clear promotion criteria.
+
+## Compute comp and demand (Block D)
+
+**Check first:** If `reports/deep-{company-slug}-*.md` exists, read it — deep research may already have comp data, funding info, and hiring signals. Use it as a starting point instead of duplicating WebSearch effort.
+
+Use WebSearch for:
+- Current salaries for the role (Glassdoor, Levels.fyi, Blind)
+- Company compensation reputation
+- Role demand trend
+
+Table with data and cited sources. If no data is available, say so instead of making things up.
+
+## List the customization plan (Block E)
+
+Build a table with these columns: # | Section | Current State | Proposed Change | Why.
+
+| # | Section | Current State | Proposed Change | Why |
+|---|---------|---------------|-----------------|-----|
+| 1 | Summary | ... | ... | ... |
+| ... | ... | ... | ... | ... |
+
+Top 5 changes to the CV + Top 5 changes to LinkedIn to maximize match.
+
+## List interview-prep stories (Block F)
+
+Map 6-10 STAR+R stories to JD requirements (STAR + **Reflection**).
+
+| # | JD Requirement | STAR+R Story | S | T | A | R | Reflection |
+|---|----------------|--------------|---|---|---|---|------------|
+
+The **Reflection** column captures what was learned or what would be done differently. This signals seniority — junior candidates describe what happened, senior candidates extract lessons.
+
+**Story Bank:** If `interview-prep/story-bank.md` exists, manage it actively — not just append:
+
+1. **Check for existing stories** that cover the same theme (e.g., "technical leadership", "delivery under pressure"). Use the `**Best for questions about:**` tags to match.
+2. **If a new story covers the same theme as an existing one**, compare them. Keep the one with stronger quantified results and broader applicability. Update the match count on the keeper. Remove the weaker one.
+3. **If a new story covers a NEW theme**, append it.
+4. **Update match counts**: Each story MUST have a `**Matched N evaluations:**` line listing which reports used it (e.g., `Matched 5 evaluations: #012, #045, #078, #102, #115`). Stories with 5+ matches are the most versatile — practice them first.
+5. **Cap at 10-12 stories max.** If adding a new story would exceed 12, retire the story with the fewest matches and narrowest applicability. Move retired stories to a `## Retired` section at the bottom (don't delete — keep them for niche roles).
+6. **Tag archetypes**: Each story MUST have `**Archetypes:**` listing which role archetypes it's strongest for (e.g., `LLMOps, Platform`).
+
+The goal is a curated bank of 10 versatile stories, not an ever-growing log.
+
+**Frame story selection by archetype.**
+
+- FDE: emphasize delivery speed and client-facing work.
+- SA: emphasize architecture decisions.
+- PM: emphasize discovery and trade-offs.
+- LLMOps: emphasize metrics, evals, production hardening.
+- Agentic: emphasize orchestration, error handling, HITL.
+- Transformation: emphasize adoption, organizational change.
+
+Also include:
+- 1 recommended case study (which of their projects to present and how)
+- Red-flag questions and how to answer them (e.g., "Why did you sell your company?", "Do you have direct reports?")
+
+---
+
+## Run post-evaluation steps
+
+**ALWAYS** after generating blocks A-F:
+
+### Save the report .md (step 1)
+
+Save the full evaluation to `reports/{###}-{company-slug}-{YYYY-MM-DD}.md`.
+
+- `{###}` = next sequential number (3 digits, zero-padded)
+- `{company-slug}` = company name in lowercase, no spaces (use hyphens)
+- `{YYYY-MM-DD}` = current date
+
+**Report format:**
+
+```markdown
+# Evaluation: {Company} — {Role}
+
+**Date:** {YYYY-MM-DD}
+**Archetype:** {detected}
+**Score:** {X.X/5}
+**URL:** {original offer URL}
+**PDF:** {path or pending}
+
+---
+
+## Score
+
+{the JSON block emitted per _shared.md, verbatim, inside a fenced ```json block}
+
+---
+
+## A) Role Summary
+(full content of block A — reference scores by key, do not re-enumerate dimensions)
+
+## B) CV Match
+(full content of block B)
+
+## C) Level and Strategy
+(full content of block C)
+
+## D) Comp and Demand
+(full content of block D)
+
+## E) Customization Plan
+(full content of block E)
+
+## F) Interview Prep Plan
+(full content of block F)
+
+## G) Draft Application Answers
+(only if `draft_answers_threshold_met` in the JSON — draft answers for the application form)
+
+---
+
+## Extracted Keywords
+(list of 15-20 keywords from the JD for ATS optimization)
+```
+
+### Global Score
+
+**Use the Canonical Scoring Model from `modes/_shared.md`.** The per-dimension breakdown lives in the JSON block (`## Score` section of the report). Don't repeat it as a prose table — that's the duplication we're eliminating. If you need to reference a specific dimension in the narrative, quote its score and rationale from the JSON inline: *"Seniority fit (3/5 — Senior IC, no formal mgmt) is the main gap."*
+
+### Append to the tracker (step 2)
+
+**ALWAYS** register in `data/applications/` (the day file for the current date, e.g., `data/applications/2026-04-13.md`):
+- Next sequential number
+- Current date
+- Company
+- Role
+- Score: weighted total from the Canonical Scoring Model (1-5)
+- Status: `Evaluated`
+- PDF: ❌ (or ✅ if auto-pipeline generated PDF)
+- Report: relative link to the report .md (e.g., `[001](reports/001-company-2026-01-01.md)`)
+
+**Tracker format:**
+
+```markdown
+| # | Date | Company | Role | Score | Status | PDF | Report |
+```

package/modes/pdf.md

@@ -0,0 +1,121 @@
+# Mode: pdf — ATS-Optimized PDF Generation
+
+## Full Pipeline
+
+1. `cv.md` is already in your context via `opencode.json:instructions` — use it as the source of truth, do NOT Read it again
+2. Ask the user for the JD if not already in context (text or URL)
+3. Extract 15-20 keywords from the JD
+4. Detect JD language → CV language (EN default)
+5. Detect company location → paper format:
+   - US/Canada → `letter`
+   - Rest of the world → `a4`
+6. Detect role archetype → adapt framing
+7. Rewrite Professional Summary injecting JD keywords + exit narrative bridge ("Built and sold a business. Now applying systems thinking to [JD domain].")
+8. Select the top 3-4 most relevant projects for the offer
+9. Reorder experience bullets by relevance to the JD
+10. Build competency grid from JD requirements (6-8 keyword phrases)
+11. Inject keywords naturally into existing achievements (NEVER fabricate)
+12. Generate complete HTML from template + personalized content
+13. Write HTML to `/tmp/cv-candidate-{company}.html`
+14. Run: `node generate-pdf.mjs /tmp/cv-candidate-{company}.html output/cv-candidate-{company}-{YYYY-MM-DD}.pdf --format={letter|a4}`
+15. Report: PDF path, page count, keyword coverage %
+
+## Apply these ATS rules for clean parsing
+
+- Single-column layout (no sidebars, no parallel columns)
+- Standard headers: "Professional Summary", "Work Experience", "Education", "Skills", "Certifications", "Projects"
+- No text in images/SVGs
+- No critical info in PDF headers/footers (ATS ignores them)
+- UTF-8, selectable text (not rasterized)
+- No nested tables
+- JD keywords distributed across: Summary (top 5), first bullet of each role, Skills section
+
+## Apply this PDF design
+
+- **Fonts**: Space Grotesk (headings, 600-700) + DM Sans (body, 400-500)
+- **Fonts self-hosted**: `fonts/`
+- **Header**: name in Space Grotesk 24px bold + gradient line `linear-gradient(to right, hsl(187,74%,32%), hsl(270,70%,45%))` 2px + contact row
+- **Section headers**: Space Grotesk 13px, uppercase, letter-spacing 0.05em, cyan primary color
+- **Body**: DM Sans 11px, line-height 1.5
+- **Company names**: accent purple color `hsl(270,70%,45%)`
+- **Margins**: 0.6in
+- **Background**: pure white
+
+## Sort sections for the "6-second recruiter scan"
+
+1. Header (large name, gradient, contact, portfolio link)
+2. Professional Summary (3-4 lines, keyword-dense)
+3. Core Competencies (6-8 keyword phrases in flex-grid)
+4. Work Experience (reverse chronological)
+5. Projects (top 3-4 most relevant)
+6. Education & Certifications
+7. Skills (languages + technical)
+
+## Apply ethical keyword injection (truth-based)
+
+Examples of legitimate reformulation:
+- JD says "RAG pipelines" and CV says "LLM workflows with retrieval" → change to "RAG pipeline design and LLM orchestration workflows"
+- JD says "MLOps" and CV says "observability, evals, error handling" → change to "MLOps and observability: evals, error handling, cost monitoring"
+- JD says "stakeholder management" and CV says "collaborated with team" → change to "stakeholder management across engineering, operations, and business"
+
+**NEVER add skills the candidate does not have. Only reformulate real experience using the exact vocabulary of the JD.**
+
+## Write in a style that defeats AI-detection (CRITICAL)
+
+ATS platforms (Indeed, LinkedIn, Workday) increasingly flag AI-generated CVs. The PDF generation MUST produce text that reads as human-written. This is not about deception — it's about ensuring the candidate's real experience isn't filtered out by an automated detector before a human ever sees it.
+
+### Apply sentence-structure variations
+- **Vary sentence length deliberately.** Mix short punchy fragments ("Shipped in 3 weeks.") with longer compound sentences. AI text tends toward uniform medium-length sentences.
+- **Start bullets differently.** Don't begin every bullet with a past-tense action verb. Mix structures: "Led...", "The team needed X, so I...", "After discovering Y, rebuilt Z to..."
+- **Use the candidate's actual phrasing from cv.md when possible.** The candidate's own words have idiosyncratic phrasing that AI-generated text lacks. Preserve their voice — reformulate for keywords, but don't rewrite the whole sentence.
+
+### Select words carefully
+- **Avoid AI-hallmark phrases.** Never use: "leveraged", "utilized", "spearheaded", "orchestrated" (as a metaphor for "managed"), "cutting-edge", "passionate about", "drive innovation", "synergy", "holistic approach", "navigate complex", "foster collaboration". These are the first things detectors look for.
+- **Use plain, specific verbs.** "Built" not "architected". "Ran" not "orchestrated". "Fixed" not "remediated". "Cut costs by" not "optimized cost efficiency". The more specific and concrete, the more human it reads.
+- **Preserve technical jargon as-is.** Real engineers say "k8s" not "Kubernetes orchestration platform". Keep the candidate's working shorthand.
+
+### Apply structural variations
+- **Don't over-polish.** A real CV has minor asymmetries — one job has 4 bullets, another has 3. One bullet is 2 lines, the next is 1. Don't normalize everything to uniform length.
+- **Keep the Professional Summary under 4 sentences.** AI-generated summaries tend to be dense paragraphs that try to cover everything. A human writes a tighter summary and lets the experience section do the work.
+- **Don't repeat the same metric in both the summary and a bullet.** Humans don't do this. Pick the best place for each number.
+
+### Check the draft before generating HTML
+After drafting all CV content, review it once against these checks.
+1. Do 3+ bullets start with the same word? → Rewrite the openings.
+2. Are all bullets the same length (± 25 chars)? → Vary them.
+3. Does any sentence contain 2+ words from the AI-hallmark list in "Choose words carefully"? → Rewrite.
+4. Does the summary read like a paragraph from a cover letter? → Make it more telegraphic.
+
+## Use this HTML template
+
+Use the template in `cv-template.html`. Replace the `{{...}}` placeholders with personalized content:
+
+| Placeholder | Content |
+|-------------|---------|
+| `{{LANG}}` | `en` or `es` |
+| `{{PAGE_WIDTH}}` | `8.5in` (letter) or `210mm` (A4) |
+| `{{NAME}}` | (from profile.yml) |
+| `{{EMAIL}}` | (from profile.yml) |
+| `{{LINKEDIN_URL}}` | (from profile.yml) |
+| `{{LINKEDIN_DISPLAY}}` | (from profile.yml) |
+| `{{PORTFOLIO_URL}}` | (from profile.yml) (or /es depending on language) |
+| `{{PORTFOLIO_DISPLAY}}` | (from profile.yml) (or /es depending on language) |
+| `{{LOCATION}}` | (from profile.yml) |
+| `{{SECTION_SUMMARY}}` | Professional Summary / Resumen Profesional |
+| `{{SUMMARY_TEXT}}` | Personalized summary with keywords |
+| `{{SECTION_COMPETENCIES}}` | Core Competencies / Competencias Core |
+| `{{COMPETENCIES}}` | `<span class="competency-tag">keyword</span>` × 6-8 |
+| `{{SECTION_EXPERIENCE}}` | Work Experience / Experiencia Laboral |
+| `{{EXPERIENCE}}` | HTML for each job with reordered bullets |
+| `{{SECTION_PROJECTS}}` | Projects / Proyectos |
+| `{{PROJECTS}}` | HTML for top 3-4 projects |
+| `{{SECTION_EDUCATION}}` | Education / Formación |
+| `{{EDUCATION}}` | HTML for education |
+| `{{SECTION_CERTIFICATIONS}}` | Certifications / Certificaciones |
+| `{{CERTIFICATIONS}}` | HTML for certifications |
+| `{{SECTION_SKILLS}}` | Skills / Competencias |
+| `{{SKILLS}}` | HTML for skills |
+
+## Update the tracker post-generation
+
+Update the tracker when the offer is already registered: change PDF from ❌ to ✅.

package/modes/pipeline.md

@@ -0,0 +1,83 @@
+# Mode: pipeline — URL Inbox (Second Brain)
+
+Processes accumulated job offer URLs from `data/pipeline.md`. The user adds URLs at any time and then runs `/job-forge pipeline` to process them all.
+
+## Run This Workflow
+
+1. **Read** `data/pipeline.md` → find `- [ ]` items in the "Pending" section
+2. **For each pending URL**:
+   a. Calculate the next sequential `REPORT_NUM` (read `reports/`, take the highest number + 1)
+   b. **Extract JD** using Geometra MCP (geometra_connect + geometra_page_model) → WebFetch → WebSearch
+   c. If the URL is not accessible → mark as `- [!]` with a note and continue
+   d. **Run full auto-pipeline**: A-F Evaluation → Report .md → PDF (if score >= 3.0, per `_shared.md` thresholds) → Draft answers (if score >= 3.5) → Tracker
+   e. **Move from "Pending" to "Processed"**: `- [x] #NNN | URL | Company | Role | Score/5 | PDF ✅/❌`
+3. **Parallel dispatch — max 2 at a time** (Hard Limit #1 in `AGENTS.md`). For N pending URLs, run `ceil(N/2)` rounds of 2 `task` dispatches. Never 3+ in one message.
+4. **When finished**, display a summary table:
+
+```
+| # | Company | Role | Score | PDF | Recommended action |
+```
+
+## Apply pipeline.md Format
+
+```markdown
+## Pending
+- [ ] https://jobs.example.com/posting/123
+- [ ] https://boards.greenhouse.io/company/jobs/456 | Company Inc | Senior PM
+- [!] https://private.url/job — Error: login required
+
+## Processed
+- [x] #143 | https://jobs.example.com/posting/789 | Acme Corp | AI PM | 4.2/5 | PDF ✅
+- [x] #144 | https://boards.greenhouse.io/xyz/jobs/012 | BigCo | SA | 2.1/5 | PDF ❌
+```
+
+## Detect JD From URL
+
+1. **Geometra MCP (preferred):** `geometra_connect` + `geometra_page_model`. Works with all SPAs, uses fewer tokens than raw DOM snapshots.
+2. **WebFetch (fallback):** For static pages or when Geometra is not available.
+3. **WebSearch (last resort):** Search on secondary portals that index the JD.
+
+**Special cases:**
+- **LinkedIn**: May require login → mark `[!]` and ask the user to paste the text
+- **PDF**: If the URL points to a PDF, read it directly with the Read tool
+- **`local:` prefix**: Read the local file. Example: `local:jds/linkedin-pm-ai.md` → read `jds/linkedin-pm-ai.md`
+
+## Automatic Numbering
+
+1. List all files in `reports/`
+2. Extract the number from the prefix (e.g., `142-medispend...` → 142)
+3. New number = highest found + 1
+
+## Source Synchronization
+
+Before processing any URL, verify sync:
+```bash
+node cv-sync-check.mjs
+```
+If there is a desynchronization, warn the user before continuing.
+
+## Run This Pipeline Runbook When N >= 2
+
+```
+Step 1  — Read data/pipeline.md; collect "- [ ]" URLs into `pending = [url_1, ..., url_N]`
+Step 2  — Pre-flight cleanup (once, before loop):
+            geometra_list_sessions()
+            geometra_disconnect({ closeBrowser: true })
+Step 3  — For round in ceil(N/2):
+            pair = pending[round*2 : round*2 + 2]
+            # ONE message, 1 or 2 task() calls. Never 3.
+            task(process url pair[0])
+            task(process url pair[1])  # only if pair has 2
+            # WAIT for both returns before the next round.
+Step 4  — Between rounds: geometra_list_sessions() + geometra_disconnect({closeBrowser: true})
+Step 5  — Reconcile outcomes (Hard Limit #6):
+            bash: node merge-tracker.mjs      # TSVs → correct day file
+            bash: node verify-pipeline.mjs    # validate URL/status consistency
+Step 6  — Display summary table; flag any verify-pipeline errors.
+```
+
+**Hard rules:**
+- Max 2 `task` dispatches per message (Hard Limit #1).
+- Never re-dispatch a URL whose previous subagent is still in-flight (Hard Limit #5).
+- Orchestrator does not call `geometra_fill_form` / `geometra_page_model` in multi-URL runs (Hard Limit #4) — delegate.
+- **The only edits allowed to `data/pipeline.md` are flipping `[ ]` → `[x]`** (inbox state) (Hard Limit #6). APPLIED / FAILED / SKIP outcomes go via `batch/tracker-additions/*.tsv` into the day file. Do NOT write application status to pipeline.md.

package/modes/project.md

@@ -0,0 +1,30 @@
+# Mode: project — Portfolio Project Evaluation
+
+Scoring across 6 dimensions (1-5):
+
+| Dimension | Weight | 5 = ... | 1 = ... |
+|-----------|--------|---------|---------|
+| Signal for target roles | 25% | Directly demonstrates JD skill | Unrelated |
+| Uniqueness | 20% | Nobody has done this | Everyone has it |
+| Demo-ability | 20% | Live demo in 2 min | Code only, not visual |
+| Metrics potential | 15% | Clear metrics (latency, cost, accuracy) | No possible metrics |
+| Time to MVP | 10% | 1 week | 3+ months |
+| STAR story potential | 10% | Rich story with trade-offs | Just implementation |
+
+## "Interview Pack" Requirements
+
+For each approved project:
+1. **One-pager**: product + architecture + metrics + evaluation plan
+2. **Demo**: live URL or 2-min recorded walkthrough
+3. **Postmortem**: what worked, what didn't, mitigations
+
+## Apply The 80/20 Plan
+
+- Week 1 → MVP with core metric
+- Week 2 → polish + interview pack
+
+## Return One Of These Verdicts
+
+- **BUILD** → plan with weekly milestones
+- **SKIP** → why and what to do instead
+- **PIVOT TO [alternative]** → more impactful variant

package/modes/rejection.md

@@ -0,0 +1,92 @@
+# Mode: rejection — Rejection Analysis & Learning Loop
+
+When the user reports a rejection, capture structured data and surface patterns over time.
+
+**This mode is additive.** It uses the existing notes column in tracker files (no new columns) and appends a section to existing reports (no changes to report format).
+
+## Use These Trigger Conditions
+
+- User says "I got rejected from [company]".
+- User updates a tracker entry to Rejected.
+- User runs `/job-forge rejection` to review patterns.
+
+## Step 1 — Record The Rejection
+
+Update the entry in `data/applications/` (the day file where the entry exists):
+- Status → `Rejected`
+- Notes → append rejection info in compact format: `REJ@{stage}:{reason}`
+
+**Stages** (ask the user which applies):
+| Stage | Code | Meaning |
+|-------|------|---------|
+| Resume screen | `resume` | Rejected before any human contact |
+| Phone screen | `phone` | Rejected after initial call |
+| Technical interview | `tech` | Rejected after technical assessment |
+| Hiring manager | `hm` | Rejected after HM interview |
+| Final round | `final` | Rejected at final stage |
+| Offer pulled | `offer` | Offer was extended then retracted |
+| Unknown | `unk` | User doesn't know |
+
+**Reasons** (ask the user if they have signal, otherwise `unk`):
+- `exp` — not enough experience / seniority
+- `skill` — missing specific technical skill
+- `culture` — culture/team fit
+- `comp` — comp expectations too high
+- `geo` — location/timezone mismatch
+- `internal` — went with internal candidate
+- `closed` — role was closed/frozen
+- `unk` — no feedback given
+
+Example notes: `REJ@tech:skill — failed system design round`
+
+## Step 2 — Append To Report
+
+Append a `## Rejection` section to the report in `reports/` when one exists:
+
+```markdown
+## Rejection
+**Date:** YYYY-MM-DD
+**Stage:** Technical interview
+**Reason:** Missing specific skill (system design at scale)
+**Signal:** What this tells us about future applications for this archetype
+```
+
+## Step 3 — Run Pattern Analysis
+
+Run pattern analysis when the user runs `/job-forge rejection` without a specific company. Analyze ALL rejected entries:
+
+1. **Read** all day files in `data/applications/` — filter to Rejected entries
+2. **Parse** `REJ@{stage}:{reason}` from notes column
+3. **Aggregate** patterns:
+
+```
+## Rejection Patterns — {date}
+
+### By Stage
+- Resume screen: N rejections (N% of all rejections)
+- Technical: N rejections
+- ...
+
+### By Archetype
+- LLMOps roles: N/M rejected (N% rejection rate)
+- Platform roles: N/M rejected
+- ...
+
+### By Score Range
+- Score 4.0+: N/M rejected (should be low — if high, scoring model needs recalibration)
+- Score 3.0-3.9: N/M rejected
+- Score <3.0: N/M rejected (expected to be high)
+
+### Signals
+- [If 3+ resume-screen rejections for same archetype]: "Consider stronger keyword injection for {archetype} roles — see pdf.md anti-AI-detection rules"
+- [If high-score offers getting rejected]: "Scoring model may be inflated — review calibration anchors in _shared.md"
+- [If rejections cluster at tech stage]: "Story bank may need stronger technical stories for {archetype} — review interview-prep/story-bank.md"
+- [If comp rejections]: "Consider adjusting target range in config/profile.yml"
+```
+
+## Skip These Actions (Out Of Scope)
+
+- Does NOT change the scoring model or thresholds automatically
+- Does NOT modify reports retroactively (only appends Rejection section)
+- Does NOT add new columns to tracker files (uses notes column)
+- Does NOT require any changes to merge-tracker, dedup, or verify scripts