Pre-Screen Gate Logic in Career-Ops Pipeline Processing: Economy Tier Deep Dive

In the santifer/career-ops repository, the economy tier completely disables the pre-screen gate and proceeds directly to full A-F evaluation after liveness sweeps, while standard and premium tiers employ an economy-equivalent model to filter mismatches before expensive processing.

The pre-screen gate logic in career-ops pipeline processing serves as a critical cost-optimization checkpoint that determines whether job postings undergo preliminary filtering or immediate evaluation. In the santifer/career-ops codebase, this behavior varies strictly by spend tier configuration, with the economy tier implementing a unique bypass mechanism that eliminates redundant latency while maintaining evaluation throughput.

How the Pipeline Gating System Works

The pipeline mode defined in modes/pipeline.md orchestrates a two-step validation sequence for every URL stored in data/pipeline.md:

1. Liveness sweep – Validates URL accessibility and removes dead postings (lines 9-15).
2. Pre-screen gate – Conditionally executes a North Star archetype check using the economy-equivalent model defined in modes/_shared.md.

This architecture ensures that only viable, relevant postings consume expensive inference resources, though the second step is tier-dependent.

Economy Tier Bypass Behavior

For economy tier configurations, the pre-screen gate is explicitly disabled. According to modes/pipeline.md (lines 19-25):

"economy tier: No gate. The tier is already the cheapest available. Every surviving pending URL goes straight to the full evaluation."

After the liveness sweep eliminates dead links, economy tier pipelines proceed immediately to comprehensive A-F evaluation using the cheapest available model. This design eliminates redundant latency because applying an additional filter would not reduce costs—the economy tier already employs the most cost-efficient model configuration.

Standard and Premium Tier Gate Logic

In contrast, standard and premium tiers activate the pre-screen gate to prevent wasting expensive inference credits on obvious mismatches. The system invokes the economy-equivalent model (mapped in the Spend-Tier table of modes/_shared.md, lines 37-48) to perform rapid North Star archetype validation.

When a job description fails to align with the user's defined archetypes, the pipeline:

• Marks the entry with a #-- prefix in data/pipeline.md
• Appends a discard record to data/discard.log
• Skips the expensive full evaluation

The skip marker appears as:

- [x] #-- | https://jobs.example.com/posting/4 | skipped (pre-screen mismatch: not a North Star archetype)

Configuration and Model Routing

The active spend tier is read from config/profile.yml via the spend_tier key. If unspecified, the system defaults to standard.

The mapping between tier names and actual model implementations resides in modes/_shared.md within the "Spend Tier (Model Routing)" table. This abstraction allows the pre-screen gate to reference economy-equivalent models regardless of which specific provider or model version is currently assigned to the economy tier.

The same gate logic documented in modes/pipeline.md is mirrored in modes/batch.md for batch processing workflows, ensuring consistent behavior across execution modes.

Practical Implementation Examples

Economy Tier Execution

Configure config/profile.yml with the economy tier to bypass pre-screening:

# config/profile.yml

spend_tier: economy

# Execute pipeline

node pipeline.mjs

Result: Following the liveness sweep, each pending URL proceeds directly to full A-F evaluation. The data/pipeline.md file contains no #-- pre-screen markers, and data/discard.log remains empty because the gate is inactive.

Standard Tier Execution

Enable the pre-screen gate using the standard tier:

# config/profile.yml

spend_tier: standard

# Execute pipeline

node pipeline.mjs

Result: The pipeline invokes the economy-equivalent model for preliminary North Star archetype screening. Mismatched URLs receive skip markers in the pipeline file and audit entries in the discard log, while matches proceed to full evaluation.

Summary

• Economy tier bypass: The pre-screen gate is completely disabled in modes/pipeline.md (lines 19-25), allowing immediate progression to full evaluation after liveness verification.
• Standard/premium filtering: These tiers utilize an economy-equivalent model defined in modes/_shared.md (lines 37-48) to filter North Star archetype mismatches before expensive processing.
• Configuration source: The spend_tier value in config/profile.yml controls gate behavior, defaulting to standard when omitted.
• Audit trail: Only standard and premium tiers populate data/discard.log with pre-screen rejection records.
• Batch parity: modes/batch.md implements identical tier-based gate logic for non-pipeline executions.

Frequently Asked Questions

Why does the economy tier skip the pre-screen gate?

The economy tier already utilizes the cheapest available model for full A-F evaluation. Adding a pre-screen gate would introduce latency without reducing costs, as there is no cheaper model tier to use for preliminary filtering. This optimization is explicitly documented in modes/pipeline.md lines 19-25.

What happens to URLs that fail the pre-screen gate in standard tier?

Rejected URLs receive a #-- marker in data/pipeline.md indicating a pre-screen mismatch with the North Star archetype, and the system appends an audit entry to data/discard.log. These postings bypass expensive evaluation but remain cataloged for compliance review.

Where is the economy-equivalent model defined for pre-screening?

The mapping resides in the Spend-Tier table within modes/_shared.md (lines 37-48). This table routes standard and premium tier pre-screen checks to whichever model is currently assigned to the economy tier, ensuring cost-efficient preliminary filtering without hardcoding specific model names.

Does the batch mode use the same pre-screen logic as the pipeline?

Yes. According to the source implementation, modes/batch.md mirrors the identical tier-based gate logic found in modes/pipeline.md, ensuring consistent pre-screen behavior whether processing individual URLs or batch collections.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}