dennisthiessenandClaude Opus 5
fc5f3e49ba
F03 — application records
- Quota already removed with F02's tracker; verified uncapped and that
"interview reached" survives a later withdrawal.
- Aker BP: application date confirmed 2026-07-29 by the user, settling the
one-day conflict. Its Adjacent-despite-79 classification is kept, NOT
reclassified: the user does not remember the rationale and judged it
immaterial on a closed role, and silently reclassifying to make the
arithmetic tidy is the exact objection the review raised. Recorded as an
explicit override reason with a rubric version instead.
F04 — rule hierarchy
- Three levels stated in AGENTS.md and CLAUDE.md: immutable evidence rules,
standing user preferences, market/employer presentation.
- Deliverables separated from QA. The .tex stays the artefact of record; you
still compile, into the application's own output/ folder, because the user
cannot approve what he cannot open.
- `resume_reference.md` said "Capgemini/Bundeswehr: include only when directly
relevant", contradicting the standing never-include rule. Split and fixed.
F05 — a green validator now means something
- Per-claim `forbidden` wording is enforced. 14 of 26 claims carried such a
list and NONE of it was checked, which is why "owned the full ML lifecycle"
passed. Matching tolerates inserted adjectives: the real regression read
"Built a decentralized Data Mesh" and one word defeated the literal list.
- Added identity (e-mail/phone), employer-date, placeholder and
document-exclusion checks. Capgemini's omission rule now has a canonical
home in claims.json instead of living only in memory and config.md.
- Registered Groovy and Snowflake as forbidden skills. Both were absent from
claims.json entirely, so "forbidden" existed in prose with nothing enforcing
it.
- PASS no longer reads as a truth certificate: it prints what it did not check
— scope implications, bullet-to-claim support, whether a number is real.
F06 — the scanner stopped fighting the documents
- LaTeX comments are no longer claims; `% never claim Terraform` was a finding
before, which taught the workflow to write around the checker.
- Markup, accent macros and umlauts are normalized, so `\textbf{Snow}flake`
cannot smuggle a forbidden skill past a substring test and
`Universit\"at der Bundeswehr M\"unchen` is recognisably a school.
- Honest denials pass in English and German.
- The triple-list cadence hint knew only English "and", so it silently did
nothing on every German package. Advisory only, and now labelled as such.
Verification
- The six counter-proofs from the report are caught: 9 errors on one synthetic
document, where the old scanner found none.
- All six packages built under current rules scan clean — SBB, NVIDIA, RUAG,
Kdo Cy, Citadel, Aker BP. Four false positives of my own were found and fixed
on the way there; each is now a regression test.
- 13 older packages do report findings. Those look like genuine historical
violations and are deliberately NOT fixed here: F17 governs archives and they
must not be silently rewritten.
- `tests/test_document_scanner.py` added. Full suite 31/31 pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qej5SBvhcbGXHwLqNWiskM
Evidence-first resume kit
This repository turns verified career evidence into role-specific LaTeX resumes and, when useful, cover letters. It is designed to prevent a common failure mode of AI-assisted applications: making a document sound closer to the job description by silently overstating skills, scope, titles, or outcomes.
The workflow separates three questions that should never be confused:
- Evidence Fit: does the candidate actually meet the role, including hard gates?
- Document Quality: does the resume communicate the supported evidence clearly?
- Channel Strength: is the application cold, referred, or supported by a warm contact?
A polished resume cannot repair a failed qualification gate. A strong fit can still be hidden by a weak document. The system records both.
Source hierarchy
source documents
-> structured extractions
-> canonical claims registry
-> experience records and role bundles
-> session plan
-> generated resume / optional cover letter
resume_builder/canonical/claims.json is the highest-authority career source. Generated files under output/ are never treated as evidence. historical_outputs.json records whether an older package is safe to consult, must be revalidated, or must not be reused.
Default application strategy
- Target Staff/Senior Data Engineering first.
- Use ML Platform/MLOps, data platform, analytics, and semiconductor profiles only where direct evidence covers the actual requirements.
- Treat direct LLM engineering, formal model evaluation, Terraform-heavy SRE, and similar missing required experience as gates rather than keyword opportunities.
- Record fit and outcomes without application quotas; a fit classification is a preparation signal.
- Add a warm-channel action for core applications.
The default document is a two-page International Tech resume for US-tech/FAANG-style roles in Europe. A Swiss/DACH policy overlay is available for employers that expect local conventions. Cover letters are conditional, not automatic.
Workflow
Add evidence
- Put employment records, project material, papers, or notes under
knowledge_base/. - Run
setup-extractto create a source-grounded extraction. - Review it, then run
setup-build-kbto update canonical claims, experience files, bundles, and support data.
Apply to a role
- Run
make-resumewith the real JD. - Review the requirement table and fit gate before approving content generation.
- Generate and validate the LaTeX resume.
- Run
make-clonly when the session records a justifiedYESdecision. - Run
critiquefor separate Evidence Fit, Document Quality, and Channel Strength findings. - Run
edit-resumefor approved repairs.
The skills live in .agents/skills/. Each application receives a session file under output/<Role>/ that preserves the JD, fit decision, claim IDs, content plan, validation results, and application status.
Validation
Validate the knowledge system:
python resume_builder/helpers/validate_resume_system.py
Validate a generated document:
python resume_builder/helpers/validate_resume_system.py --document output/Role/resume.tex
Track the uncapped application record (use summary --all for older applications too):
python resume_builder/helpers/cohort_tracker.py summary
job_scout/state/applications.json owns current status and event history. The old cohort JSON and
resume_builder/reference/current_applications.md are generated views. Update status without erasing interview history:
python resume_builder/helpers/cohort_tracker.py update --id <application-id> --date YYYY-MM-DD --event interview_held --note "Interview completed"
python resume_builder/helpers/cohort_tracker.py update --id <application-id> --date YYYY-MM-DD --outcome withdrawn_by_candidate --note "Reason"
python resume_builder/helpers/application_records.py --check
Use application_records.py --render to recover stale generated views. A classification that differs
from its numeric rubric needs --fit-override-reason; it is never silently relabelled.
Character counts are available only as readability diagnostics. There are no fixed character bands, equal-length bullet rules, or page-fill quotas.
Output and prerequisites
The system writes .tex source only. Compile locally with a LaTeX distribution such as MiKTeX, TeX Live, or MacTeX. Generated resumes should also be checked through PDF rendering and text extraction before submission.
See DOCS.md for the architecture and policy reference.
License
MIT; see LICENSE.