Status: draft from the blank-shell scout, not reviewed. The library is the product; the draft is a byproduct. This file says how an engagement runs, what the client gives at each stage, and where the harness does the work versus the human.
| The harness carries | The human carries |
|---|---|
| Framework mapping (register and requirement files) | Scope and materiality cuts |
| Drafting against the register and data table | Conversations with ESG leads, CFOs, operations |
| Interview prep packs (from register gaps) | Interpretation of what the evidence means |
| Documentation (content index, methodology notes) | Sign-off and the client relationship |
QA execution (qa/run_qa.py, logged every run) | Judgment on anything QA flags as NA or FAIL |
| # | Stage | Client gives | Harness does | Output | Gate |
|---|---|---|---|---|---|
| 1 | Intake | LP list and the requests they sent; fund and company list with sector, size, revenue, locations; the entity facts the screen needs; any existing reports | Converts the facts into entity-facts-v1.json, runs engine/run_applicability.py, drafts the scope note from its result | Scope note; applicability result; settings file (report date, period, declared year_status, any earlier partial disclosures) | Owner signs scope |
| 2 | Materiality | The prior materiality assessment (it exists; do not redo it) | Converts it to concept ids in settings.material_topics; flags register concepts it does not cover | Material-topic list; gap list | Owner signs the cut |
| 3 | References | Nothing (Year 1 build) / nothing new (Year 2) | Year 1: confirm framework files are current (re-check-by dates). Year 2: re-verify each file past its date | Library at a known version | Check: no framework past its re-check date |
| 4 | Data | Emissions inventory, revenue, physical-risk model output, transition and initiatives view, governance documents (charters, minutes) | Converts each into data-table rows with source, factor version and grade; recomputes register status | Data table; filled register; coverage matrix and gap list | QA Q01, Q02, Q18 pass |
| 5 | Interviews | Time with ESG lead, CFO, operations per company | engine/build_prep_packs.py writes one pack per company: material gaps first, then gaps a framework that applies to that company needs, then grade C figures to upgrade, then missing applicability facts; each with question, who to ask and why | Interview notes, turned into rows or narrative inputs | Each gap closed or recorded as a disclosed gap |
| 6 | Draft | Qualification sentence; any authorisation to name a company | Writes narrative sections with figure tokens only; runs the engine | LP report skeleton; content index per framework | QA full run |
| 7 | QA | Nothing | Runs all checks; logs every run to qa/qa-log-v1.jsonl | QA result file | All PASS or explained NA |
| 8 | Sign-off | Named reviewers at the GP and every portfolio company | Builds the sign-off list from settings; checks every company contact is on it | Signed report; frozen files | Owner and client sign |
{{fig:C-MET-01}} or {{row:R-001}}). Never type a number into narrative. QA Q05 fails a typed number.[S2-MT-a1]) that resolves to a row in a framework file. QA Q14 checks this.../methodology/narrative-voice-v1.md, ../methodology/anonymisation-convention-v1.md.outputs/run-all.sh (or qa/run_qa.py --case <folder>).qa/qa-log-v1.jsonl with the report's hash, the verdict and each check's status and detail. NA results say why. The log is append-only.-vN. Draft files are edited in place with each edit logged (hash, reason). Once a file has been in a reviewer packet or frozen at sign-off, a change is a new version. Superseded versions go to _archive/; nothing is deleted.These are the scout's estimates for a GP with about ten companies, unverified: from model knowledge; no engagement has been timed against them.
| Stage | Year 1 (hours) | Year 2 (hours) | Why it shrinks, or does not |
|---|---|---|---|
| Intake | 6 | 3 | Applicability rules exist; only changes in LP list or thresholds need work |
| Materiality | 6 | 2 | Re-confirm the prior cut; change only if the portfolio changed |
| References | 20 | 6 | Year 1 builds and verifies framework files; Year 2 re-verifies past their re-check date. Does not reach zero: regulation moves |
| Data | 30 | 18 | Data-table shape and source mapping are known; new and exited companies still cost full effort; prior-year reconciliation is new work |
| Interviews | 20 | 14 | Prep packs come from the gap list; conversations themselves do not get shorter |
| Draft | 16 | 8 | Year 2 starts from last year's narrative and the same tokens |
| QA | 6 | 3 | Checks are scripts; time goes on judging FAILs. Year 2 adds two live prior-year checks |
| Sign-off | 8 | 7 | People time; barely shrinks |
| Total | 112 | 61 | About 45% less, mostly from references, drafting and QA |
Next year: copy last year's case folder; save last year's published data table as prior-published-v1.csv; update settings-v1.json (report date, period, factor version, and prior_report with the published file, its period end, explanations for changes over 10% and any restatement reasons); add this year's rows and keep last year's as comparatives (make_year2_cases.py shows the shape); re-check every framework file past its re-check date; run outputs/run-all.sh; resolve every FAIL before drafting further.
Portfolio edition: Kith Climate cohort case, fictional client.