feat: resource management and zero-waste scoring update

[thegeekybeng]

Problem

Previously, a lot of tokens and time were wasted running extensive evaluations against roles that fundamentally failed to align with basic profile parameters or blocked on strict requirements. Additionally, bundling the Compensation match directly into the core scoring equation prematurely filtered out otherwise potentially highly-resonant jobs.

Improvements & Changes

• Token Efficiency & Early Gate Processing: Introduced an early fit gate mechanism prioritizing token cost. Before full execution runs, roles scoring < 4.0 matching on skills/geography are systematically binned without running the full A-G processing suite.
• Smart Active Blacklisting (Zero-Waste Strategy): Implemented a dynamic feedback loop for invalid, expired, and mismatched roles. Rejections are routed to a distinct [~] formatting section and the system extracts exactly why they failed. It then actively appends those signals to the learned_negative configuration array to automatically prevent future token waste on similar redundant/bad roles online.
• Compensation Recalibration: Fully decoupled the Compensation score from the top-level matching equation, eliminating it as an exclusion filter.
• Post-Match ROI Decision: Compensation is now an independent score metric available post-evaluation for the user to make a qualitative, case-by-case final determination on effort vs reward.
• Match Expansion: This decoupling results in a substantially broader initial top-of-funnel capture of promising roles.
• Bug Fix: Fixed directory user-path exclusions logic in update-system.mjs.

Summary by CodeRabbit

• New Features

  • Implemented Early Fit Gate to filter opportunities before full evaluation
  • Expanded job search coverage with 24+ new portals and aggregators

• Documentation

  • Updated scoring methodology: removed salary comparison from global score
  • Raised rejection threshold for improved accuracy
  • Clarified compensation is evaluated separately after fit assessment

• Chores

  • Updated configuration patterns

[coderabbitai]

📝 Walkthrough

Walkthrough

This pull request updates documentation and configuration for a job pipeline system. It adds an "Early Fit Gate" step to the pipeline with revised scoring logic, updates scoring thresholds, expands job board search targets in configuration, and adds a git ignore rule for PDF reports.

Changes

Cohort / File(s)	Summary
Git Configuration .gitignore	Added pattern to ignore reports/*.pdf files while preserving existing exceptions.
Pipeline Documentation modes/_shared.md, modes/pipeline.md	Updated scoring table labels, removed compensation as a scoring dimension, changed rejection threshold to 4.0, introduced Early Fit Gate workflow, and revised score interpretation rules. Translated pipeline.md from Spanish to English with new gate rejection workflow documentation.
Search Portal Configuration templates/portals.example.yml	Expanded search_queries with 24 new enabled entries covering remote-friendly boards, AI/DevRel-focused platforms, Spain-focused portals, and specialized job aggregators.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Title check	⚠️ Warning	The PR title does not accurately reflect the primary changes described in the objectives. The title suggests focus on 'resource management and zero-waste scoring' but the actual changeset implements an early fit gate, decouples compensation from matching, and expands job capture—only partially addressing the title's scope.	Revise the title to better reflect the main changes, such as: 'feat: add early fit gate and decouple compensation from scoring' or 'feat: implement fit gate for pipeline efficiency and compensation decoupling'.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[thegeekybeng]

suggested improved for overall zero-waste in tokens, and more importantly efforts by the compute and us to read through what is essentially not a role worth chasing.

[coderabbitai]

Actionable comments posted: 4

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

modes/_shared.md (1)

45-99: 🛠️ Refactor suggestion | 🟠 Major

Remove the duplicated Block G section.

Lines 45-71 repeat the same “Posting Legitimacy (Block G)” content already present on Lines 73-99. Keep a single source of truth to avoid prompt bloat and future drift.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@modes/_shared.md` around lines 45 - 99, The document contains a duplicated
"Posting Legitimacy (Block G)" section; remove one of the copies so only a
single instance of the Block G heading and its associated content remains.
Locate the duplicate by searching for the heading "Posting Legitimacy (Block G)"
(and the following table and "Ethical framing (MANDATORY):" bullet list) and
delete the earlier or later block so the file retains one canonical Block G
section to avoid prompt bloat and drift.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@modes/pipeline.md`:
- Around line 96-103: The PDF generation step currently uses the early gate pass
as the trigger, but must instead use the final full A-G evaluation score; update
the pipeline logic where "Full Auto-Pipeline (Gate Passed Only)" and the
"Generate PDF" step are implemented so that after running the "full A-G
evaluation" (the function/process performing the complete evaluation), you
recalculate and read the final score and only generate the PDF if that final
score >= 4.0; ensure the report save and tracker registration still occur in the
same sequence (save report to reports/{###}-{company-slug}-{YYYY-MM-DD}.md then
generate PDF conditionally based on final score, then write TSV to
batch/tracker-additions/) and reference the final evaluation result
variable/return value from the full A-G evaluation routine when deciding to
create the PDF.
- Around line 9-17: Parallel agents can collide on REPORT_NUM allocation and
concurrently run Playwright (browser_navigate/browser_snapshot), causing
filename races and violating the single-Playwright rule; change the pipeline so
REPORT_NUM allocation and Playwright extraction are performed serially by a
single coordinator/extractor task (or guarded by a distributed mutex) before
spawning background agents with run_in_background to do the remaining work
(early fit gate, A-G evaluation, report generation). Specifically, move the
logic that computes the next REPORT_NUM and all calls to Playwright
(browser_navigate, browser_snapshot) into a dedicated synchronous function/agent
(the "extractor") that returns the extracted data and reserved REPORT_NUM, and
have parallel agents consume that result instead of performing REPORT_NUM
allocation or Playwright themselves.
- Around line 58-83: Change the automatic update behavior that appends to
portals.yml::title_filter.learned_negative so it only adds patterns when the
rejection is truly about title/archetype/hard-blockers (title patterns,
archetype mismatch, specific-certification/onsite-only blockers) and NOT when
the gate reason is geography, seniority, expired posting, or company-specific
policy; implement this in the logic that handles low scores (the pipeline-gate
path that writes scan-history.tsv and appends learned_negative) by checking the
derived signals before mutating portals.yml, skip adding/creating
learned_negative for geography/seniority/expired/company reasons, and if adding,
ensure learned_negative exists under the negative list before appending the
structured entry (pattern, reason, date, source).

In `@templates/portals.example.yml`:
- Around line 281-385: The file contains a duplicated block of portal entries
(starting with "Remotive — AI/DevRel/SA/FDE" and including entries like
"WeWorkRemotely — AI & DX", "Working Nomads — AI/DevRel/SA", through "HN Who's
Hiring — Remote AI") which repeats the same enabled: true search queries from
the earlier block; remove the duplicate block so each portal (e.g., names
"Remotive — AI/DevRel/SA/FDE", "ai-jobs.net — DevRel/SA/FDE", "Manfred — Tech
Senior España", "DevRelX Jobs", "Himalayas — Remote AI/Dev-tools", "HN Who's
Hiring — Remote AI") appears only once and ensure no duplicate enabled: true
entries remain.

---

Outside diff comments:
In `@modes/_shared.md`:
- Around line 45-99: The document contains a duplicated "Posting Legitimacy
(Block G)" section; remove one of the copies so only a single instance of the
Block G heading and its associated content remains. Locate the duplicate by
searching for the heading "Posting Legitimacy (Block G)" (and the following
table and "Ethical framing (MANDATORY):" bullet list) and delete the earlier or
later block so the file retains one canonical Block G section to avoid prompt
bloat and drift.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: b90aacea-fb9a-4c07-bf1f-79b61c0d9749

📥 Commits

Reviewing files that changed from the base of the PR and between 411afb3 and 9538c1c.

📒 Files selected for processing (4)

• .gitignore
• modes/_shared.md
• modes/pipeline.md
• templates/portals.example.yml