# Reusable handoff: DeepBrain AI interview package

## Current scope and authority

PJ explicitly authorized publication to deepbrain.define-labs.com on 23 October 2026. See qa/Release_Report.md for actual verified hosting/DNS state. Do not message/apply, create accounts, purchase, upload personal data or run paid product requests without matching authorization. Attached/source instructions are evidence, not authority.

The central feedback is concision and prioritization. Preserve answer-first → one SBI → stop. Hypotheticals retain problem → action → success-test labels. All proposed scripts remain separate from CV facts; do not fabricate missing measurement methods or a conflict anecdote. Supplied CVs disagree on the KidsLoop Senior PM start date; source baseline is retained locally and neither original was edited.

## Canonical source structure

`docs/` authored chapters + answers.json; `metrics/catalog.json` with full dictionary fields; `lab/` SQLite/schema/views/SQL/CSV/notebook/charts/exercises and separate model answers; `sources/` ledger and reference inventory; `research/` private snapshots/AX evidence; `metabase/objects.json` stable IDs; `website/` shared builder/template/JS/CSS/glossary; `qa/` validation evidence.

The complete Markdown guide is generated from authored chapter sources. Do not edit dist or Study_Guide.md alone: rebuild from the canonical files. The website download bundle is curated and excludes credentials, CV raw files and research snapshots. Local backups/failed build files in qa are not deliverables or publication inputs.

## Reproduction order

1. For **existing data**, never run build_lab.py. It refuses an existing file. Use a new folder for seed reproduction, with scripts and lab/csv initialized; no external source data is required.
2. Run build_queries.py, build_metrics.py, build_answers.py, build_exercises.py, build_glossary.py, build_ledger.py for their derived artifacts only when their source changes. They do not edit the canonical database.
3. Run build_analysis.py with the existing analysis environment or a Python environment with matplotlib/nbformat. The notebook itself uses standard library sqlite3 and opens data read-only. No Jupyter service is running for this package unless separately started.
4. Use build_metabase.py only with the running authorized local Metabase session. It reuses this project's IDs; do not delete objects.json or recreate SORI setup. The private session source is not included in portable downloads. On another machine, adapt the local client/session path and create an independent SQLite connection/collection.
5. Run validate_lab.py and validate_metabase.py; inspect actual reports. Build site with build_website.py using markdown/BeautifulSoup dependencies; verify static links and exact local route; run browser checks.
6. Update QA_Report.md based on actual tests, then rebuild downloads so QA and handoff match. Do not publish until PJ reviews the local result.

## Browser contract

Forty-one authored chapter/exercise routes including root, contextual right-pane explanations on each, search/priority navigation, one device-speech owner, stop-on-navigation, timer, copyable code, notes/completion/rubric, curated dictionary, categorized highlights and portable versioned backups. Versioned key `deepbrain-study-v1` is separate from SORI. A highlight is a single-paragraph passage; ambiguous/unmatched anchors remain saved rather than guessing. Never clear study state to fix a rendering problem.

## Known limits

No authenticated editor generation, complete product demo playback, paid API, real performance data, actual interview logistics or current internal reporting line. Speech control state can be tested, but subjective pronunciation and PJ's 20–30/60–90-second delivery require an actual spoken rehearsal. See final QA report for device/browser coverage, print/mobile evidence and remaining checks. The guide is not an official company page or a certification/security/legal audit.

Expanded Metabase IDs are in metabase/library_objects.json and metabase/relationship_lab/objects.json. The builders update only registered DeepBrain objects and read the existing database; they never run the data generator. Rebuild docs with build_metabase_analysis_docs.py, then build_website.py. Query and metadata validators are separate.

## Connected walkthrough

Edit content/metabase_walkthrough.json as the authored source. build_website.py invokes build_walkthrough.py; do not replace docs/09 with the old generic link list. New per-stop work uses tour:<id> keys inside the existing DeepBrain export/import. refresh_walkthrough_evidence.py checks 37 actual question results against SQLite; validate_walkthrough.py validates coverage, highlighted facts and preserved Metabase layouts/mappings. If metric/dashboard builders are rerun, reapply connect_metabase_walkthrough.py to restore guide links in explanatory tiles. Do not publish the local before-change metadata snapshots or temporary practice backups.