Step 5: Final Verification

An independent audit of the salary table from Step 4. Always runs; this is the only step in the pipeline with no skip option, since it produces the pipeline’s final result. The four detailed checks are file-only, like Step 4’s own work, but the resulting final table is the one thing this step does show you, copied from the file into chat, table only with no surrounding commentary. Whether recalibration happened is stated as a plain fact in the completion message afterward, not as commentary attached to the table. Everything (checks, verdict, and table) is written to a single file, sc-step5b-final-verification.md; this step never creates more than one file. Claude 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 5 begins]) --> Ladder[Command drafts career ladder\non current model, wait for\nconfirmation, save to\nsc-step5a-career-ladder.md]
  Ladder --> Confirm{Career ladder\nconfirmed?}
  Confirm -->|no| EditLadder[Adjust ladder per feedback]
  EditLadder --> Confirm
  Confirm -->|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[1. Candidate positioning\nUse the confirmed ladder from\nsc-step5a-career-ladder.md]
  C1 --> C2[2. Framework calibration review\nIs the framework conservative · calibrated\nor inflated?]
  C2 --> C3[3. Country-by-country verification\nSafe positioning · Stretch positioning\nrecruiter comfort · overseas hiring realism]
  C3 --> C4[4. Framework recommendation\nIs Safe / Stretch philosophy still appropriate?]
  C4 --> Recal{Evidence supports\nrecalibration?}
  Recal -->|yes| Revise[5. Recalibrate: revise affected countries\nRecompute Legal Requirement column\nRevised table becomes the final table]
  Recal -->|no| Confirm2[Confirm existing framework is appropriate\nCopy Step 4's table in unchanged\nas the final table]
  Revise --> SaveAll[Write checks, verdict, and\nfinal table into ONE file:\nsc-step5b-final-verification.md]
  Confirm2 --> SaveAll
  SaveAll --> ShowTable[Show final table in chat\ncopied from the file,\ntable only, no commentary]
  ShowTable --> Done([Completion message])

What it reads

  • sc-step4-salary-table.md: the final salary table from Step 4
  • sc-step2-salary-data.md and sc-step3-adjustment-values.md: the underlying salary data and adjustment values
  • profile.md: used to assess candidate positioning and career level
  • sc-step5a-career-ladder.md: the career ladder drafted and confirmed with you before the step runs

All inputs come from workspace files, so the audit is safe to route to the isolated deep-reasoner subagent. The career ladder is confirmed on your current model before any Opus handoff, so the subagent never has to pause for interactive confirmation.

All four checks and the recalibration verdict are saved to file only; not reproduced in chat, since the raw reasoning would overwhelm rather than help. The resulting final table (whether unchanged or revised) is written into that same file, then also shown directly in chat, copied from what was just saved, table only, no surrounding commentary.

The four checks

(File-only, written to sc-step5b-final-verification.md, not shown in chat. The final table further down is the exception.)

1. Candidate positioning

The career ladder for your role is drafted from profile.md and confirmed with you by the main command before the step runs, then saved to sc-step5a-career-ladder.md. It assumes an IT/tech professional but is inferred from your actual role (any IT role, not assumed to be backend) using that role’s real progression and title conventions (software engineering might run Mid → Senior → Staff → Principal, while data, DevOps, QA, security, product, or design each have their own ladder). The step uses that confirmed ladder to determine your likely current level and target role level. Compensation is benchmarked against the target role, not the highest historical responsibility.

2. Framework calibration review

Assesses whether the overall salary framework is conservative, recruiter-safe, appropriately calibrated, slightly inflated, or heavily inflated.

Evaluates legitimate compensation drivers (experience, technical depth, leadership scope, business impact, domain expertise) separately from potential inflation sources (premium employer weighting, multinational bias, niche specialist premium, higher-level title interpretation, AI optimism bias).

3. Country-by-country verification

For each country, evaluates:

  • Safe positioning: percentile band and classification (conservative to top-tier)
  • Stretch positioning: classification (realistic stretch to international remote premium)
  • Recruiter comfort: how likely is this to convert interviews?
  • Sponsorship realism: if Step 4 set a Legal Requirement flag for this country, that concrete result feeds into this judgment rather than being assessed independently of it
  • Overseas hiring realism: adjusted for your profile as an international candidate

Uses approximate percentile bands (50–60%, 60–70%, etc.), no false precision.

4. Framework recommendation

Determines whether the Safe/Stretch philosophy, employer segmentation, and percentile assumptions remain appropriate. If improvements are recommended, explains what assumption caused the issue and what structural change is recommended.

Recalibration

Only if the evidence genuinely supports it:

  • Affected countries are revised upward or downward
  • A revised table is generated using the same format as Step 4, including the Legal Requirement column; it is recomputed against the revised Safe and Stretch figures, not carried forward from Step 4. Recalibration can push a country below a threshold it previously cleared, or above one it previously missed.
  • Priority is given to recruiter comfort, interview conversion, sponsorship realism, and realistic overseas positioning
  • This revised table becomes the final table.

If recalibration is not supported, the existing framework is explicitly confirmed as appropriate; Step 4’s table is copied in unchanged as the final table, no separate revision is drafted.

Either way, the resulting final table (unchanged or revised) is written into sc-step5b-final-verification.md, in the same file as the four checks and the recalibration verdict. sc-step4-salary-table.md is left untouched as the pre-audit record. This step never creates a second file.

Output

  • sc-step5b-final-verification.md: the only file this step writes. Always contains the full audit (all four checks, file-only) plus the recalibration verdict and the resulting final table (unchanged or revised).

The final table is also shown directly in chat, copied from the file, table only, no surrounding commentary. The four checks themselves are not reproduced in chat. Claude closes with a completion message naming how many countries were audited, whether recalibration occurred, and confirming where the full audit is saved.

Stop condition

Once the audit (and any recalibration) is complete and the completion message is shown, Claude stops and waits for the main command to deliver the final results; this is the last step in the pipeline.

Claude Compass — MIT License