Step 6 — Reality check (optional)

An independent audit of the Step 5 scoring output. Claude asks before running this step — it only proceeds if you confirm. If you confirm, Claude then asks whether to use the deep-reasoner subagent (Opus, high effort) for higher reasoning accuracy. If you decline, the step runs with your current model.

Flow 

flowchart TD
  Start([Step 6 begins]) --> RunQ{Run reality check?}
  RunQ -->|no| SCQ([Offer /salary-calculator\nscoped to one country])
  RunQ -->|yes| OpusQ{Use Opus for\nhigher accuracy?}
  OpusQ -->|yes| DeepReasoner[Route to deep-reasoner\nOpus / high effort]
  OpusQ -->|no| CurrentModel[Run with your\ncurrent model]
  DeepReasoner --> C1
  CurrentModel --> C1[Check 1: Criteria consistency\nSame rules applied to all countries?]
  C1 --> C2[Check 2: Confidence calibration\nHigh confidence backed by real evidence?]
  C2 --> C3[Check 3: Exclusion reasoning\nEach exclusion cites specific evidence?]
  C3 --> C4[Check 4: Cross-track contamination\nEach track scored independently?]
  C4 --> C5[Check 5: Missing candidate check\nExpected absences are evidence-based?]
  C5 --> C6[Check 6: Framework calibration\nStrong / Moderate / Weak meaningfully distributed?]
  C6 --> Recal{Evidence supports\nrecalibration?}
  Recal -->|yes| Revise[Revise classifications or\nexclusion status — explain each change]
  Recal -->|no| Confirm[Confirm Step 5 results\nare appropriate — no changes]
  Revise --> Offer([Offer /salary-calculator])
  Confirm --> Offer

Purpose

The reality check challenges the scoring results from Step 5. It does not automatically defend or attack them — it follows the evidence. Recalibration only happens if the evidence genuinely supports it.

What it reads

  • All data and results from Steps 1 through 5

The six checks

1. Criteria consistency check

  • Was your Step 1 salary minimum applied the same way to every Remote-track country?
  • Was the time zone limit applied only to Remote, never to Sponsorship?
  • Were your Step 1 dealbreakers applied consistently to every Sponsorship-track country?

Any inconsistency is stated explicitly.

2. Confidence calibration check

For every country marked High confidence: verifies that the underlying evidence is genuinely specific, sourced, and dated — not just confidently worded. Flags any confidence level that seems inflated relative to the actual evidence quality.

3. Exclusion reasoning check

Re-examines every exclusion from Step 5. Confirms each one cites a specific fact from stored data, not a vague or reputation-based dismissal. Flags any generic dismissal that lacks real evidence.

4. Cross-track contamination check

For any country scored on both tracks: checks whether the Sponsorship reasoning was influenced by the Remote judgment or vice versa. Each track’s score must stand on its own evidence independently.

5. Missing candidate check

Identifies any country that would commonly be expected to appear but is absent. For each one, states whether the absence was a genuine evidence-based elimination or a process gap (never researched, never pasted into Step 4).

6. Framework calibration check

Assesses whether the Strong / Moderate / Weak classification is meaningfully distinguishing between countries, or whether most countries clustered into a single bucket — which would indicate criteria that are too loose or too strict.

Recalibration

Only if the evidence genuinely supports it:

  • Specific country classifications, confidence levels, or exclusion status may be revised
  • Each change is explained with the evidence that caused it

If recalibration is not supported, the Step 5 results are explicitly confirmed as appropriate and left unchanged.

After the reality check

Claude asks whether you want salary data for any of the results and offers to run /salary-calculator scoped to a single named country. The same offer appears if you decline the reality check entirely — you are never left without the handoff option.

Claude Compass — MIT License