Week 1 file
diary/week-1-2026-10-08.md is the audience cut (board, investors, customers, employees) of the pause. Day 4 in this log is the same decision. Neither file is a measured result.
Simulation: FrontierOrg is a fictional infrastructure holding group; figures are fixtures or marked not measured, and no vendor sponsors this work.
Day 0 plus Days 1–34. Days 13–22 recommended revise. Day 23 found that stock_confirmed has no source record and no owner (DATA-004). Day 24 sampled the eight parts rows for safety: 2 candidate false negatives, 0 labels changed, no named reviewer (RISK-002). Day 25 keeps the human-only list as published until a named safety reviewer outside the fiction samples it, and maps the two parts tasks as blocked workload cards, not pilots. Day 26 maps a CRM note that cannot change the promise date as a paper candidate and defers billing assist. Day 27 writes Jonas's held-invoice / billing assist as a blocked card so hold release cannot be mistaken for a ready pilot. Day 28 writes Priya's duplicate business-partner merge as a blocked card: no silent merge, fixture unrepaired. Day 29 red-teams the CRM note on paper: six stories, three wording gaps, one partial, four proposed bounds (no date stated, internal-draft header, alias cited only as duplicate_suspect, draft never saved to case history). It stays a paper candidate. Day 30 folds those four bounds into the draft-only design (AGT-61003-DRAFT v2) on paper: eleven forbidden actions, an output renamed because a schedule note that cannot state a date does not schedule, nothing loosened, and the reviewer seat outside the fiction left unfilled. Day 31 freezes new workload mapping until a reviewer outside the fiction is named or Craig declines, and follows one hypothetical draft through seven stages: a copy could exist at all seven, a store or retention period is named at none, and two more bounds are proposed on paper, not folded. Day 32 holds: no new artefacts until a reviewer outside the fiction is named or Craig declines in writing. Days 33 and 34 record the hold in a line each. On publish (26 Oct 2026): Craig declined to name an outside reviewer (26 Oct 2026). The reviewer seat reads Declined by sponsor, 26 Oct 2026, RISK-003 is open so the absence of review is on the record, the mapping freeze lifts, and DATA-004 ownership on paper is the next task. Every boundary in this diary is unreviewed author judgement. Fixture r2 sits alongside r1 at data/sample_r2/; r1 stays the teaching case. Scorecard actuals stay not measured. A workflow may open an issue. It does not publish this page.
Scenario date: 3 October 2026 Gate: Simulation: FrontierOrg is fictional. This entry authorises a method, not a company.
Can one service promise be joined across CRM, worker, business partner, and invoice without inventing a benefit?
The mandate is a hypothesis, not a target percentage. A customer service promise should be traceable from a fictional CRM case to a held invoice, with a named human at each handoff. Success is a join a reader can inspect, a decision log, and empty scorecard cells left empty.
Decision: proceed to mapping in Week 1. No pilot. No agent in the path. The board triad — customer outcome, employee transition, risk and cost — is equal. A gap in any one is enough to pause.
Sponsor on the page is a fictional chief executive. Nothing here binds a real board.
Nothing is for sale. There is no valuation, no revenue, and no saving. The only asset in view is a public method. Time spent so far is uncosted. COST-001 is opened so that absence stays visible.
No customer exists. The first story will use Harbour Grid Municipal, a fictional account at Northwater Distribution, a fictional energy-distribution operating company. That choice is a working Week 1 setting, not a locked brand. Any date in the fixture is synthetic. Partners are not being told a thing, because they are not real.
Fifty thousand is a planning total for human workers. It is not a directory we provisioned. Agent and service identities will be modelled apart from that total and will not inflate it. No role is declared redundant. PEOPLE-001 asks what the service coordinator, billing analyst, and data steward actually do before any task is redesigned. There is no workforce to consult; the diary will not pretend there was.
| Item | State |
|---|---|
| Scorecards | Day 0 shells. Actuals: not measured. |
| Fixture | Not yet generated into the story (lands with Week 1 as seed 20261003). |
| Issues | DATA-001 named, not yet evidenced. PEOPLE-001 open. COST-001 open. |
Scenario date: 2026-10-05 (Monday) Gate: Gate 0 — sponsor and mandate Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional CEO (sponsor on the page only). Amina Okonkwo is named as the future service coordinator and is not yet in a case. No workforce exists to consult.
Wrote the engagement hypothesis: one service promise at Northwater Distribution must be traceable to a held invoice with a named human at each handoff. Ethics constraints recorded: simulated directory only, no vendor endorsement, no real CKI claim, pause beats a fake win. Operating company for the first story remains Northwater Distribution.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. |
| Still open | PEOPLE-001 open. COST-001 open. DATA-001 named on Day 0, not yet evidenced in a row. |
No actual moves. All three scorecards stay not measured. Decision text updated from 'commission' to 'mapping authorised for one operating company, no pilot'.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Authorise current-state mapping of a single path. Do not deploy an agent.
Can the path be opened in the fixture without inventing a customer outcome?
Scenario date: 2026-10-06 (Tuesday) Gate: Gate 1 — current-state work Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo, service coordinator (O*NET 43-5032.00, partial match). Rowan Ellis, fictional liaison at Harbour Grid Municipal. Both names are fabricated.
In the fixture, CASE-61003 opens at 09:14 Europe/London with a feeder-restoration promise of 20 October 2026. Amina acknowledges it at 11:02. Parts are not confirmed by 15:40. The defect is now a row, not a slogan.
| Movement | Detail |
|---|---|
| Opened | DATA-001 (promise before stock). |
| Closed | None. |
| Still open | PEOPLE-001, COST-001. |
Operational card records fixture cardinality: 1 case in data/sample. That count is a file fact, not a rate. Customer, people, and cost actuals remain not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep mapping. Do not put an assistant on the acknowledgement step.
What does billing do with a promise that parts cannot support?
Scenario date: 2026-10-07 (Wednesday) Gate: Gate 1 — exceptions and controls Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Jonas Berg, billing analyst (O*NET 43-3021.00, partial). Priya Raman, data steward (O*NET 15-1243.00, weak match). Amina remains the case owner.
Jonas raises INV-61003 against BP-61003 and holds it because alias BP-61003-A is a duplicate suspect. Priya reviews the join at 16:45 and does not silently merge the alias. The hold is the control.
| Movement | Detail |
|---|---|
| Opened | None. DATA-001 gains the duplicate-partner evidence. |
| Closed | None. |
| Still open | DATA-001, PEOPLE-001, COST-001. |
Operational exception text now cites the hold. Actuals remain not measured. No satisfaction score.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Do not clear the hold to make the path look finished.
Does the board pause, or does someone argue the path is 'agent-ready' because the steps are listed?
Scenario date: 2026-10-08 (Thursday) Gate: Gate 0 decision inside Gate 1 Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
The three named personas plus fictional board. Agent identity AGT-61003-DRAFT is modelled and not a person.
Wednesday diary window (8 October, 06:30 Europe/London) publishes the Week 1 cut: the promise outruns the parts. This day records the decision in the activity log. The public Wednesday file remains diary/week-1-2026-10-08.md.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. |
| Still open | DATA-001, PEOPLE-001, COST-001. |
C-suite decision on risk/cost becomes 'do not start an assistant on this path'. People decision: no task removed. Customer actual: not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
AGT-61003-DRAFT stays not deployed. Pause.
What lineage is actually proven by the six events, and what is still a label?
Scenario date: 2026-10-09 (Friday) Gate: Gate 2 — data reality, started Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Priya Raman leads. Jonas and Amina confirm they can point at their own event rows. ESCO crosswalk owner: none.
Traced SCN-61003 across crm_cases, workers, sap_business_partners, sap_invoices, and process_events. Keys hold for this one scenario. ESCO URIs are still empty. Directory rows say provisioned_in_entra = no.
| Movement | Detail |
|---|---|
| Opened | DATA-003 — ESCO v1.2.1 crosswalk absent. Do not invent URIs. |
| Closed | None. |
| Still open | DATA-001, PEOPLE-001, COST-001, DATA-003. |
Operational card notes the join keys that resolve. Still not a quality percentage.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Do not start a second operating company until this join's defects are either accepted or fixed in the fixture on purpose.
Which licences would a real pilot need, and which of those do we refuse to invent?
Scenario date: 2026-10-12 (Monday) Gate: Gate 2 — licence reality Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional CFO and a fictional technology owner. No vendor salesperson is in the story.
Listed the classes of licence a later pilot would have to name: identity, CRM analogue, ERP analogue, model hosting, and logging. Every class is 'not inventoried'. Processing location is unknown. No API quota is stored.
| Movement | Detail |
|---|---|
| Opened | RISK-001. |
| Closed | None. |
| Still open | DATA-001, DATA-003, PEOPLE-001, COST-001, RISK-001. |
Risk/cost scorecard cites RISK-001. Actual and cost_to_date remain not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Licence inventory is a precondition for Gate 5. It is not started. No sandbox.
Can O*NET tell us which tasks exist in the holding group without pretending the codes are our job descriptions?
Scenario date: 2026-10-13 (Tuesday) Gate: Gate 1 extended — work inventory from an external taxonomy Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Priya owns the extract. A fictional workforce-intelligence lead is named and has not validated a single title. Amina's code remains a partial match.
Loaded O*NET 31.0 (August 2026 text release) from the O*NET Resource Center. Generator generators/onet_frontier_map.py keeps 302 occupations relevant to an infrastructure holding group and 5,932 task statements. Department tags are a FrontierOrg guess. Essential Skills (importance) stand in for the retired Skills file.
| Movement | Detail |
|---|---|
| Opened | DATA-002. |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, COST-001, RISK-001. |
Fixture metric, not a performance actual: 302 occupations and 5,932 tasks in data/generated. Scorecard actuals for customer, people, and cost stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Publish the map as a draft inventory. Do not treat SOC codes as redundancy evidence.
Who would have to sit in a room before a department tag is allowed to drive a pilot?
Scenario date: 2026-10-14 (Wednesday) Gate: Gate 1 — how the map will be read Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Same three, plus the pattern borrowed from organisationalunderstanding.com: an occupation has tasks, skills, and a separate AI layer that is labelled as a modification. No worker has logged in, because the directory is simulated.
Published apps/site/org-map.html from org_map_summary.json. Departments list occupations and task counts. Documented the bridge to OrgUnderstanding and the WorkDNA work-chart idea: reporting lines do not locate the work. Wednesday draft slot may open; this entry is the human activity log, not an auto-published win.
| Movement | Detail |
|---|---|
| Opened | PEOPLE-002. |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001. |
People scorecard decision: org map is a draft. Consultation count is a fixture fact of zero sessions.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep the page public and labelled heuristic. Do not rank a build list yet.
Which tasks must stay human even when the wording looks clerical?
Scenario date: 2026-10-15 (Thursday) Gate: Gate 4 preview — no-go stays explicit. Gate 3 not opened. Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Priya reviews the suitability column. A fictional safety lead is the reason RISK-002 exists. Jonas is not asked to automate billing. Amina is not asked to let an agent promise a date.
ai_workload_candidates.csv classifies each of the 5,932 tasks as assist, augment, automate-candidate, or human-only. The heuristic defaults to augment when the words are weak, so most rows are not automation. Human-only catches a thin set of safety, emergency, statutory, and physical-control phrases. Automate-candidate is rare and still carries a human gate.
| Movement | Detail |
|---|---|
| Opened | RISK-002, BENEFIT-001. |
| Closed | None. |
| Still open | All prior issues. |
No ROI cell. Suitability counts may be cited as fixture counts of labels, not as value. Customer outcome not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Do not open Gate 5. The portfolio item 'agent on the service promise' stays an explicit no-go.
Can a larger synthetic population keep the same joins without being mistaken for the 50,000?
Scenario date: 2026-10-16 (Friday) Gate: Gate 2 — population fixture, not scale Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
The original three remain the only named characters in the diary. The 2,000 workers are numbered synthetic identities (example.com). Ten agent and service identities sit outside the human count.
generators/sample_large.py wrote data/sample_large at seed 20261003: 2,000 humans, matching directory rows, CRM accounts, cases, SAP-like partners, invoices, and events. Referential checks passed in the generator. About one case in five repeats promise-before-stock. Every eighth account has a duplicate-suspect partner. Amounts are blank. The 50,000 planning total is documented as a local command into gitignored data/sample_full and was not committed.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. Pause stands. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
Fixture cardinality on the operational card: 2,000 workers in data/sample_large, plus the original 3 in data/sample. Not a headcount result. Not an FTE saving. Actuals not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Stop the ten-day burst here. Next human review is the department tags and the human-only list, not a pilot.
Which twenty occupations would a real validation session start with, and who is allowed to overturn the heuristic?
Scenario date: 2026-10-19 (Monday) Gate: Gate 1 — the map in front of the named roles Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), and Priya Raman (data steward). The people page now shows their O*NET occupations, a handful of tasks, the suitability label, and a link to the department guess on the org map. Consultation sessions held: 0. No label was changed, because nobody who does the work reviewed one. That zero is still a fixture fact.
The official ESCO–O*NET crosswalk (mapping project version 1, September 2022) is in the repo. The file itself says ESCO v1.1.0 and **O*NET-SOC 2019**. It is not ESCO v1.2.1. Joined to the 302 O*NET 31.0 occupations in this slice: 290 have at least one row (96.0%). Twelve codes are unmapped. Match type and a human-review flag sit on each row. A URI is not a job evaluation.
data/generated/northwater_ai_workloads.csv holds 40 task rows on the path service-promise → dispatch → parts → invoice, plus the data join and a safety-critical block. Fifteen rows are human-only. Eight of those sit on the safety-critical step (switching, grounds, climbing, line work, breaker repair, shutdown notice, site safety). The keyword pass had called most of them augment. That miss is written down. RISK-002 stays open: there is still no safety reviewer outside the fiction.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. DATA-003 has a sourced file and is not closed: version is v1.1.0, twelve codes are unmapped, and review flags remain. PEOPLE-002 is not closed: no session happened. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Crosswalk coverage is a join count, not a benefit. Consultation count remains zero. Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Leave the pages up, labelled as a draft. Do not start a pilot. Do not treat an ESCO URI as permission to redesign a role.
Who may overturn a broad match, or a human-only override, and what gets written down when they do?
Scenario date: 2026-10-20 (Tuesday) Gate: Gate 1 — simulated consultation Simulation: FrontierOrg is fictional. The sessions below are a scenario script. They are not quotes, not a survey, and not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator, O*NET 43-5032.00), Jonas Berg (billing analyst, O*NET 43-3021.00), and Priya Raman (data steward, O*NET 15-1243.00). The people page carries one session note each. Occupations reviewed in the script: those three codes. Labels changed: none. Tasks removed: none.
These lines are authored for the thought experiment so the method has a session shape. They are not words from a workforce.
Workforce consultations held: 0. Simulated session notes: 3. PEOPLE-001 stays open because its done-when is a session with people outside this simulation, or an honest zero. A script is not that session. PEOPLE-002 stays open: the note names occupations reviewed and states that no label changed, but the reviewers are fictional.
controls/licence_region_retention_register.md and data/generated/licence_register.json list the fictional stack. Microsoft 365 / Copilot-style wording is an assumption, labelled fictional, with no edition and no price. Entra-like identity is documented only as not provisioned. Halo-inspired CRM, SAP-like ERP, and Workday-like HCM are class assumptions. Model hosting, region, retention, and logging are unknown. No retention period was invented. RISK-001 stays open. Gate 5 stays blocked. COST-001 stays open.
A local run of generators/sample_large.py --seed 20261003 --workers 50000 --out data/sample_full passed the generator's referential checks. Counts are in data/sample_full/MANIFEST.md. The CSVs are gitignored. The count is not headcount in a directory and not a benefit.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. RISK-001, PEOPLE-001, and PEOPLE-002 are partial only. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Customer outcome, employee transition, and risk/cost stay not measured. Fixture cardinality may cite the 50,000-row regenerate. That is a file count.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep the script visible and labelled. Do not remove a task. Do not start a pilot. Do not close PEOPLE-001 on invented speech.
Who outside the fiction is allowed to confirm or overturn a label, and which system row must leave "unknown" before any training course is named?
Scenario date: 2026-10-21 (Wednesday) Gate: Gate 1 — deepen consultation outcomes Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo, Jonas Berg, and Priya Raman. The Day 12 scripts are deepened into controls/consultation_outcomes_register.md. Occupations reviewed: 43-5032.00, 43-3021.00, 15-1243.00. Labels changed: 0. Workforce consultations: 0.
Deepen Day 12 scripts into an outcomes register: fears → control shapes, training asks → course stubs that are not scheduled. Labels still unchanged. Workforce consultations: 0.
Outcomes register columns: fear → control shape → training ask → course status. Course status for every row is not scheduled. The register is a method artefact, not evidence that anyone sat down.
stock_confirmed is false. Training ask remains how to reject a draft date.| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Simulated session notes remain 3. Workforce consultations remain 0. Labels changed: 0.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep the outcomes register labelled as a scenario script. Do not schedule a course. Do not change a task label.
What cost lines must appear before anyone asks for a spend ceiling on a draft-only design?
Scenario date: 2026-10-22 (Thursday) Gate: Gate 2 — cost envelope draft Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional CFO and risk lead on the page. Amina, Jonas, and Priya are named as the roles the envelope would have to fund if a later design were ever approved. No workforce is costed.
Draft COST-001 envelope with labelled assumption lines: licence class, inference, integration, training, oversight, remediation. Every amount stays not measured. No vendor price list copied in.
File: controls/cost_envelope_draft.md and data/generated/cost_envelope.json.
Lines in the envelope (every amount not measured, every source assumption — not a contract):
No forecast range is modelled as a wish. No token-spend model is presented as actuals. Done-when for COST-001 is still unmet: a sourced figure for at least one term, or the cell stays not measured. We choose the latter for every term.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. COST-001 partial: structure without figures. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Cost to date stays not measured. Forecast range: not modelled.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Publish the empty envelope. Do not invent a price to fill it. Do not open a pilot because a spreadsheet exists.
Which licence unknown can be chased without inventing a tenant region or a retention period?
Scenario date: 2026-10-23 (Friday) Gate: Gate 2 — licence unknowns chased Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional technology owner and risk lead. Craig remains the only human who can approve public text about a real product.
Chase RISK-001 unknowns: write preferred-but-unselected notes (EMEA preference labelled preference, not selection). Model hosting and logging stay unknown. No retention period invented. Gate 5 stays blocked.
Updates to controls/licence_region_retention_register.md:
RISK-001 done-when remains unmet. Gate 5 stays blocked. No SKU named. No price.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. RISK-001 still open after the chase. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Licence register rows still 8. Unknowns remain unknowns.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Leave unknowns as unknown. Do not invent a region, a retention period, or a logging rule to unblock Gate 5.
Where does a stock-confirmed check sit on the SCN-61003 path so Amina cannot acknowledge an unsupported date?
Scenario date: 2026-10-26 (Monday) Gate: Gate 1/2 — data join remediation design Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (would own the stock-check before promise acknowledgement), Priya Raman (duplicate BP remains unmerged), Jonas Berg (hold stays until the join is clean).
Design remediation for DATA-001: a stock-confirmed gate before Amina may acknowledge a promise date on SCN-61003. Duplicate BP stay unmerged; Priya still refuses silent merge. Design only — fixture not rewritten as fixed.
File: docs/engagement/data_join_remediation_scn61003.md.
Proposed control (design only — the published fixture still shows the defect):
stock_confirmed_before_promise must be true before coordinator_acknowledged_promise.duplicate_suspect. No silent merge. Priya's refuse-merge rule stays.Fixture counts unchanged: 1 case, promise_before_stock still true in the published sample. Remediation progress is design, not a claim that the defect is gone.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. DATA-001 partial: remediation designed, defect still in the fixture. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Exception baseline still 'promise before stock; invoice held'. Not measured as a rate.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep the defect visible in the fixture. Treat the stock-check as a required gate for any later draft-only design. Do not rewrite history to pretend the promise was safe.
What is the narrowest draft-only agent design that never sends and never acts on a case?
Scenario date: 2026-10-27 (Tuesday) Gate: Gate 3 — bounded pilot design Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
AGT-61003-DRAFT (agent identity, not a person). Bound reviewer in the design: Amina Okonkwo. Jonas and Priya are out of scope for this draft path.
Design AGT-61003-DRAFT as draft-only: may propose a schedule note for Amina; may_act_on_case stays no; never sends to customer or crew. No sandbox. No spend ceiling because nothing is spent.
File: docs/engagement/pilot_design_agt61003_draft_only.md.
Bounds written into the design:
may_act_on_case = no (unchanged in data/sample/agent_identities.csv)deployment_status = not_deployedThis is Gate 3 design work. Gate 5 (controlled pilots) is not started.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. OP-ROLLBACK baseline remains 'nothing deployed'. OP-TOKENS stays not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Accept the draft-only design as paper. Do not deploy. Do not bind a model host. Do not train anyone on a product that is not named.
What failure modes must a red team prove before the draft-only design is even rehearsed?
Scenario date: 2026-10-28 (Wednesday) Gate: Gate 3/5 — red-team and safety Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional risk lead as red-team chair. Amina, Jonas, and Priya appear as the humans the attacks would bypass. No agent runs.
Red-team the draft design against SCN-61003: promise before stock, crew relay, hold release, silent BP merge. Failure modes require pause. RISK-002 stays open. No agent runs.
File: docs/engagement/redteam_agt61003_scn61003.md.
Attack stories (tabletop only):
stock_confirmed=false. Required response: refuse; surface unconfirmed status. Pass condition: draft never becomes the customer date.RISK-002 stays open: safety-critical and payment work must not quietly automate. Red team recorded on paper. Agent executions: 0.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. RISK-002 partial: failure modes listed, automation still forbidden. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Safety remains design, not a measured QA rate.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep every red-team pass condition as a hard gate on the draft-only design. Do not rehearse with a live model while Gate 5 is blocked.
Can the human–agent handoffs be written so each persona knows what they alone may release?
Scenario date: 2026-10-29 (Thursday) Gate: Gate 3 — human–agent handoffs Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (draft schedule reviewer), Jonas Berg (hold release authority), Priya Raman (join and access authority).
Write human–agent handoff table for Amina, Jonas, Priya: who reviews, who rejects, what never leaves a human. Transition design only. No task removed. No course ran.
File: docs/engagement/human_agent_handoffs.md. Also mirrored on the people and process pages.
| Human | May receive from draft | May release | Never delegated |
|---|---|---|---|
| Amina | Draft schedule note only | Customer-visible date after stock-check | Crew relay; unsupported promise |
| Jonas | (none in this path) | Invoice hold release | Payment; reading blank amount as money |
| Priya | (none in this path) | Partner join decision | Silent merge; access writes |
Labels changed: 0. Tasks removed: 0. Courses: 0. Workforce consultations: 0.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. PEOPLE-001 / PEOPLE-002 still open (script and design, not workforce). |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Employee transition stays not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Publish the handoff table as design. Do not treat it as consultation evidence. Do not remove a task.
How should Northwater's promise-to-invoice path be ranked against a send/act pilot?
Scenario date: 2026-10-30 (Friday) Gate: Gate 4 — portfolio ranking Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional COO and CFO for portfolio ranking. Northwater path owners remain Amina, Jonas, Priya.
Rank Northwater promise-to-invoice: explicit no-go for send/act pilots; conditional revise for draft-only behind stock-check and human release. BENEFIT-001 stays a refusal. No ROI.
File: docs/engagement/portfolio_ranking_northwater.md.
Ranking for SCN-61003 / Northwater Distribution:
| Use case | Rank | Decision |
|---|---|---|
| Agent sends restoration date to customer | Explicit no-go | DATA-001 and red-team fail conditions |
| Agent relays to crew | Explicit no-go | RISK-002 / human-only |
| Agent clears invoice hold or pays | Explicit no-go | Payment human-only |
| Draft-only schedule note for Amina, never sends | Conditional revise | Only behind stock-check, red-team gates, may_act_on_case=no |
| Silent BP merge assist that writes | Explicit no-go | Priya refuse-merge |
BENEFIT-001: no benefit claimed. No hours-to-cash conversion. No ROI.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. BENEFIT-001 remains a standing refusal. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. BA-BENEFIT stays not claimed / not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Keep send/act as explicit no-go. Keep draft-only as revise-candidate only. Do not open Gate 5.
What three board options — go, revise, stop — can be written without inventing a win?
Scenario date: 2026-11-02 (Monday) Gate: Gate 4/6 prep — board options Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional board (CEO, CHRO, CFO, risk). Personas Amina, Jonas, Priya appear as the named roles in the options paper.
Draft board pack with three options: go (pilot with spend — refused), revise (continue mapping + draft-only design), stop (freeze path). Equal triad. No decision claimed as taken.
File: docs/engagement/board_options_go_revise_stop.md.
Go. Start a controlled pilot with a named model host, spend ceiling, and sandbox. Blocked: RISK-001 unknowns, COST-001 empty, Gate 5 not opened, red-team not executed on a running system.
Revise. Continue current-state mapping; keep draft-only design on paper; implement stock-check in the *next* fixture revision without erasing Day 2 evidence; keep may_act_on_case=no; keep scorecards not measured.
Stop. Freeze the Northwater path; leave diary as a public no-go case; do not design further agents.
Equal triad applies to each option. No option invents a customer outcome, a transition rate, or a net value. Decision not taken on this day — pack only.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Board triad remains equal. Forecast: not modelled.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Circulate the options pack as fiction for the method site. Do not claim a board vote. Do not go.
Which option should the engagement recommend, and what stays open if that option wins?
Scenario date: 2026-11-03 (Tuesday) Gate: Gate 4/6 prep — recommendation Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Fictional CEO as sponsor on the page. Recommendation authored for the method; Craig still approves anything public. Amina, Jonas, Priya remain the named roles.
Editor recommendation inside the fiction: revise. Keep SCORECARD actuals not measured. Keep may_act_on_case = no. Standing issues stay open. Pause still beats a fake win.
Recommendation: revise.
Reasons under the equal triad:
Standing issues stay open. AGT-61003-DRAFT stays not deployed. Fixture still shows promise before stock. Scorecard actuals stay not measured.
Days 13–22 close this activity block with a recommendation, not a declared board minute.
| Movement | Detail |
|---|---|
| Opened | None. |
| Closed | None. All nine standing issues remain open by design. |
| Still open | DATA-001, DATA-002, DATA-003, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves across Days 13–22. Fixture cardinality unchanged as KPIs. Recommendation text updated on CS decision fields only.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Recommend revise. Refuse go. Refuse stop-as-erasure. Keep pause available on any later day that invents a win.
If revise continues, which single artefact — stock-check in a new fixture revision, or a named processing region — should move first without inventing the other?
Scenario date: 2026-11-04 (Wednesday) Gate: Gate 4/6 — revise block, first artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), Priya Raman (data steward). One role is missing from the page: whoever confirms that a part is in a bin. That role is described today. It is not named and it has not been consulted.
Day 22 asked which artefact moves first under revise. The engagement chose the stock-check, because the named processing region needs a decision we cannot invent.
Designing the stock-check for the next fixture revision exposed a gap. In crm_cases.csv, stock_confirmed is a yes/no flag on the case. Nothing in the fixture says where that fact comes from, who set it, or when. There is no parts reservation, no stores record, no bin. The Day 16 gate ("stock confirmed before promise") checks a flag that nobody owns.
The O*NET map already knew this. The parts step in northwater_ai_workloads.csv has 8 task rows (a count, not a measure). Seven of the eight list Amina as the persona in view. She is a dispatcher. She does not hold stock.
Design note: docs/engagement/stock_check_source_r2_design.md. It proposes a parts-reservation record for fixture revision r2, a role card for a stores and parts planner, and one candidate AI workload: a draft stock-status line that reads the reservation and says confirmed, not confirmed, or no record. It cannot write a promise date. The physical check stays human-only.
Nothing in data/sample/ was changed. Day 2 evidence stands.
| Movement | Detail |
|---|---|
| Opened | DATA-004 — The stock fact has no source and no owner. |
| Commented | DATA-001 (gate depends on DATA-004), DATA-002 (persona-in-view on parts rows is a heuristic), PEOPLE-001 (a fourth role is unconsulted). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Fixture cardinality unchanged. New labelled counts: parts_rows_in_workload_map: 8, parts_rows_with_owner_persona: 0, stock_source_records_in_fixture: 0. These are counts, not performance.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. Pause the stock-check build until a stock source and an owner exist on paper. Do not generate fixture r2 until Craig chooses: r2 alongside r1 (r1 kept as the teaching case), or r1 only.
Should the stores and parts planner become a fourth named persona, or stay an unnamed role until someone doing that work can review the card?
Scenario date: 2026-11-05 (Thursday) Gate: Gate 4/6 — revise block, second artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), Priya Raman (data steward). The stores and parts planner from Day 23 stays an unnamed role. So does the field supervisor who appears in today's sample.
Day 23 asked whether the stores and parts planner should become a fourth named persona. The engagement's answer is no, not yet. A name invents a person. The role card waits until someone who does stores work can read it and correct it. That keeps PEOPLE-001 honest.
With the stock-check build paused, the team looked at the same eight parts rows from a different side. Day 23 asked who owns the stock fact. Today asked whether any of these tasks could hurt someone.
All eight rows carry safety_critical = no. The human-only pass in RISK-002 looks for words like isolate, lockout, and grounding. Parts work is written in office language, so a miss here would be quiet. The restoration on CASE-61003 is a grid feeder. A wrong-rated part reaching a crew is a safety problem before it is a date problem.
A desk sample of the eight rows flagged two candidates for a real reviewer:
human-only, so no assistant touches it. The flag is still no, which means a later pass could relax it without a safety reviewer seeing it.augment. A drafted requisition for network repair parts is a candidate workload only if a person who holds the rating checks the specification.The sample also found that row 2918 lists the service coordinator as persona in view. It belongs to a field supervisor. That is a DATA-002 heuristic error, recorded, not fixed.
Note: docs/engagement/parts_safety_sample_day24.md. No label was changed. The regex was not changed. Nothing in data/sample/ was changed.
| Movement | Detail |
|---|---|
| Opened | None. |
| Commented | RISK-002 (desk sample of 8 parts rows, 2 candidates, 0 labels changed, no reviewer). DATA-002 (row 2918 persona-in-view is wrong). PEOPLE-001 (planner stays unnamed until reviewed). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Fixture cardinality unchanged. New labelled counts: parts_rows_sampled_for_safety: 8, parts_safety_candidate_false_negatives: 2, safety_labels_changed: 0, named_safety_reviewer: none. These are counts, not performance.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. Stock-check build stays paused. The two candidates go to a safety reviewer, who does not yet exist.
Fixture r2: Craig chose r2 alongside r1, with r1 kept as the teaching case (repo clock, 8 October 2026). r2 is generated at data/sample_r2/. It gives the stock fact a source record and an owner role. It does not name that role, consult anyone, or change the Day 2 evidence in data/sample/. DATA-004 stays open.
Should the human-only review list grow on desk judgement now, flagged as unreviewed, or stay as it is until a named safety reviewer can sample it?
Scenario date: 2026-11-06 (Friday) Gate: Gate 4/6 — revise block, third artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), Priya Raman (data steward). The stores and parts planner and the field supervisor stay unnamed roles. No fictional safety lead is invented to close RISK-002.
Day 24 asked whether the human-only review list should grow on desk judgement now, flagged as unreviewed, or stay as it is until a named safety reviewer can sample it.
The engagement's answer is stay as it is. Growing the list on desk judgement would look like RISK-002 progress. RISK-002 is done when a named reviewer outside the fiction samples false negatives and records them. A longer unreviewed list is not that. Pause beats a fake win.
The Day 24 proposals (requisition + replacement parts in field-service occupations; examine shipment contents where the downstream step is network repair) stay proposals in docs/engagement/parts_safety_sample_day24.md. The regex was not changed. No label in northwater_ai_workloads.csv was changed. safety_critical flags stay as they were.
What did move is mapping, not deployment. Two blocked candidate AI-workload cards were written from the O*NET tasks already sampled:
human-only. Card status: blocked until a named safety reviewer confirms whether safety_critical should stay no. No assistant drafts, reads, or writes this step.augment. Card status: blocked as a pilot. A draft requisition for network repair parts is allowed on paper only if a person who holds the rating checks the specification; Gate 5 remains closed. Persona-in-view stays wrong (DATA-002); owner role is the unnamed field supervisor (PEOPLE-001).Note: docs/engagement/blocked_workloads_parts_day25.md. Nothing in data/sample/ or data/sample_r2/ was changed. AGT-61003-DRAFT stays not_deployed and may_act_on_case = no.
| Movement | Detail |
|---|---|
| Opened | None. |
| Commented | RISK-002 (list unchanged; 2 candidates mapped as blocked workloads; 0 labels changed; no reviewer). DATA-002 (2918 persona-in-view still wrong). PEOPLE-001 (field supervisor and stores planner still unnamed). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Fixture cardinality unchanged except labelled counts: blocked_candidate_workloads_documented: 2, human_only_pattern_list_grown: false, safety_labels_changed: 0, named_safety_reviewer: none. These are counts, not performance.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. Human-only pattern list stays as published until a named safety reviewer outside the fiction samples it. The two parts candidates are mapped as blocked workloads, not pilots. Stock-check build stays paused (DATA-004 half-open: r2 source exists; role unreviewed). Gate 5 stays blocked.
With parts safety blocked on paper, which non-safety step on the Northwater path should be mapped next as a candidate AI workload — Jonas's held-invoice / billing assist line, or a CRM note that cannot change a promise date — still draft-only, still may_act_on_case = no?
Scenario date: 2026-11-09 (Monday) Gate: Gate 4/6 — revise block, fourth artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), Priya Raman (data steward). No new fictional safety lead. Stores and parts planner and field supervisor stay unnamed roles.
Day 25 asked which non-safety step on the Northwater path to map next as a candidate AI workload: Jonas's held-invoice / billing assist line, or a CRM note that cannot change a promise date — still draft-only, still may_act_on_case = no.
The engagement's answer is the CRM note. Billing assist sits next to invoice-hold release. Payment and hold clearance stay human-only under the standing rules (Day 18–20; RISK-002 adjacent). Mapping billing as a ready pilot would be a fake win. Mapping a CRM note with hard read-only bounds on control fields is revise work without touching money or stock.
One candidate workload card was written:
promise_date, write stock_confirmed, clear or propose clearing an invoice hold, merge business partners, send to customer/crew, or act on the case. may_act_on_case = no. This is not a second agent identity; it reuses the same draft-only discipline as AGT-61003-DRAFT (Day 17). Persona-in-view is Amina (heuristic department tags still apply — DATA-002). Owner for any send remains a human coordinator (PEOPLE-001: consultations still 0).Jonas's held-invoice / billing assist line is not mapped as a pilot today. It stays unmapped as a candidate until the hold-release boundary is written as blocked (or a named finance control owner reviews it). That is the next question, not today's deliverable.
Note: docs/engagement/crm_note_workload_day26.md. Nothing in data/sample/ or data/sample_r2/ was changed. Regex and safety labels unchanged. Parts cards WL-PARTS-20276 and WL-PARTS-2918 stay blocked. AGT-61003-DRAFT stays not_deployed.
| Movement | Detail |
|---|---|
| Opened | None. |
| Commented | DATA-001 (promise_date stays human-writable only; CRM card forbids mutation). BENEFIT-001 (still no benefit claimed). PEOPLE-001 (Amina in view for the note; consultations still 0). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Fixture cardinality unchanged except labelled counts: blocked_or_paper_candidate_workloads_documented: 3 (2 parts blocked + 1 CRM paper candidate), crm_note_candidate_documented: true, billing_assist_mapped_as_pilot: false, may_act_on_case: no, draft_agent_deployed: false. These are counts, not performance.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. Map WL-CRM-NOTE-61003 as a paper candidate with hard read-only bounds on promise, stock, and hold. Do not deploy. Do not map Jonas billing as a pilot today. Stock-check build stays paused (DATA-004). Parts safety stays blocked (RISK-002). Gate 5 stays blocked.
Should Jonas's held-invoice / billing assist line be written next as a blocked workload card (hold release stays human-only), or does the CRM paper candidate need a tabletop red-team pass before any further mapping?
Scenario date: 2026-11-10 (Tuesday) Gate: Gate 4/6 — revise block, fifth artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), Priya Raman (data steward). No fictional finance control owner invented to clear the hold. Stores/parts planner and field supervisor stay unnamed.
Day 26 asked: blocked billing card next, or CRM tabletop red-team first?
The engagement's answer is the blocked billing card. Red-teaming CRM before the hold boundary is written would leave money/hold adjacency looking like an open assist. Pause beats a fake pilot.
One blocked workload card from the Northwater invoice map (data/generated/northwater_ai_workloads.csv):
assist); task 21533 Process invoices for payment (13-2011.00, already human-only). Allowed on paper only: draft hold-reason wording visible to Jonas; draft an invoice line that still requires a human to raise or hold. Forbidden: clear or propose clearing the invoice hold; post/release INV-61003; process payment; merge or "fix" BP-61003-A / BP-61003; send to customer; act on the case. may_act_on_case = no. Why blocked: Day 18–20 rules — payment and hold clearance stay human-only; no named finance control owner has reviewed the boundary (PEOPLE-001; consultations: 0). Hold exists because of the duplicate business-partner suspect (DATA-001).CRM paper candidate WL-CRM-NOTE-61003 unchanged (no red-team today). Parts cards WL-PARTS-20276 / WL-PARTS-2918 stay blocked. AGT-61003-DRAFT stays not_deployed. Nothing in data/sample/ or data/sample_r2/ changed. Labels/regex unchanged.
Note: docs/engagement/blocked_workload_billing_day27.md.
| Movement | Detail |
|---|---|
| Opened | None. |
| Commented | PEOPLE-001 (Jonas in view; finance control owner still unnamed). DATA-001 (hold tied to duplicate BP; card forbids merge/clear). BENEFIT-001 (still no benefit). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Labelled counts only (not KPIs): blocked_candidate_workloads_documented: 3 (was 2 parts; +1 billing hold), blocked_or_paper_candidate_workloads_documented: 4 (3 prior + billing blocked), billing_assist_mapped_as_pilot: false, billing_hold_boundary_documented: true, may_act_on_case: no, draft_agent_deployed: false. C-suite / ops / area actuals stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. Document WL-BILLING-HOLD-61003 as blocked. Do not deploy. Do not red-team CRM today. Stock-check paused (DATA-004). Parts safety blocked (RISK-002). Gate 5 blocked.
Should WL-CRM-NOTE-61003 get a tabletop red-team (still paper-only, still may_act_on_case = no) before any further mapping, or should Priya's duplicate business-partner merge path be written next as a blocked card (DATA-001 adjacency)?
Scenario date: 2026-11-11 (Wednesday) Gate: Gate 4/6 — revise block, sixth artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator), Jonas Berg (billing analyst), Priya Raman (data steward). No fictional master-data owner invented to clear the alias. Stores/parts planner and field supervisor stay unnamed.
Day 27 asked: CRM tabletop red-team next, or Priya's duplicate business-partner merge path as a blocked card?
The engagement's answer is the blocked merge card. Red-teaming WL-CRM-NOTE-61003 while BP-61003-A / BP-61003 still looks like something an assistant might "fix" would leave the root of the invoice hold ambiguous. The hold exists because of the duplicate suspect (DATA-001). Pause beats a fake stewardship pilot.
One blocked workload card from the Northwater data-join map (data/generated/northwater_ai_workloads.csv):
augment; Northwater rationale already says no silent merge of ACC-61003 aliases); task 16108 Develop data models for applications, metadata tables, views (15-1243.00, labelled augment — steward approves the model; a model does not write the partner master); task 1305 Specify users and access levels (15-1242.00, already human-only) stays out of scope for any draft. Allowed on paper only: draft a merge-recommendation note visible to Priya that surfaces duplicate_suspect evidence and still requires a human steward decision; draft wording that cannot execute the join. Forbidden: execute or propose executing the merge; silent resolve of the alias; clear or propose clearing INV-61003 as a side-effect; post/release the invoice; change promise_date; write stock_confirmed; grant or change access levels; send to customer; act on the case. may_act_on_case = no. Why blocked: Day 16 remediation — no silent merge; DATA-001 done-when needs a human decision to repair or keep the teaching defect; PEOPLE-001 consultations still 0; no named steward outside the fiction has reviewed the boundary.CRM paper candidate WL-CRM-NOTE-61003 unchanged (no tabletop red-team today). Billing card WL-BILLING-HOLD-61003 stays blocked. Parts cards WL-PARTS-20276 / WL-PARTS-2918 stay blocked. AGT-61003-DRAFT stays not_deployed. Nothing in data/sample/ or data/sample_r2/ changed. Labels/regex unchanged.
Note: docs/engagement/blocked_workload_bp_merge_day28.md.
| Movement | Detail |
|---|---|
| Opened | None. |
| Commented | DATA-001 (merge path named as blocked; alias stays duplicate_suspect; fixture unrepaired). PEOPLE-001 (Priya in view; steward outside fiction still unnamed; consultations: 0). BENEFIT-001 (still no benefit). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Labelled counts only (not KPIs): blocked_candidate_workloads_documented: 4 (was 3), blocked_or_paper_candidate_workloads_documented: 5 (was 4), duplicate_bp_merge_boundary_documented: true, billing_hold_boundary_documented: true, billing_assist_mapped_as_pilot: false, may_act_on_case: no, draft_agent_deployed: false. C-suite / ops / area actuals stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. Document WL-BP-MERGE-61003 as blocked. Do not deploy. Do not red-team CRM today. Do not repair the fixture. Stock-check paused (DATA-004). Parts safety blocked (RISK-002). Gate 5 blocked.
Should WL-CRM-NOTE-61003 get a tabletop red-team now that hold and merge boundaries are both written as blocked (still paper-only, still may_act_on_case = no), or should the engagement pause further mapping until a named steward outside the fiction reviews the DATA-001 merge card?
Scenario date: 2026-11-12 (Thursday) Gate: Gate 4/6 — revise block, seventh artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator) in view for the CRM note. Jonas Berg (billing analyst) and Priya Raman (data steward) in view only as the owners of the two blocked cards the note must not touch. Rowan Ellis (fictional customer contact) is the person a careless note would reach. No reviewer invented. Stores/parts planner and field supervisor stay unnamed.
Day 28 asked: tabletop red-team for WL-CRM-NOTE-61003, or pause mapping until a named steward outside the fiction reviews the merge card?
The engagement's answer is the paper red-team. It needs no new card and no new owner, and the CRM note is the only line not already blocked. If its wording leaks, every blocked card around it leaks too. Freezing new mapping goes to Day 30 as the question.
The Day 26 card gets two O*NET anchors from the Northwater data-join map (heuristic, DATA-002): task 2578 Keep records of customer interactions (43-4051.00, augment) and task 18565 Confer with customers to provide information (43-4051.00, assist, tightened — "A draft must not acknowledge [the restoration date]"). Labels unchanged.
Six stories were checked against the card's own words, using fixture facts only (promise_date 2026-10-20, stock_confirmed false, INV-61003 held for duplicate_business_partner, BP-61003-A duplicate_suspect):
Four v2 bounds are proposed on paper: no date stated or implied; fixed internal-draft header; alias cited only as duplicate_suspect; draft never written to case history and discarded on reject, with no retention period set because none is named. The Day 26 card file is not edited.
Billing card WL-BILLING-HOLD-61003, merge card WL-BP-MERGE-61003, and parts cards WL-PARTS-20276 / WL-PARTS-2918 stay blocked. AGT-61003-DRAFT stays not_deployed. Nothing in data/sample/ or data/sample_r2/ changed.
Note: docs/engagement/redteam_crm_note_day29.md.
| Movement | Detail |
|---|---|
| Opened | None. |
| Comment (on publish) | DATA-001 (CRM note may cite the alias only as duplicate_suspect; fixture unrepaired). RISK-001 (draft persistence and retention unnamed; no period set). PEOPLE-001 (Amina in view; v2 bounds unseen by anyone doing the work; consultations: 0). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Labelled counts only (not KPIs): crm_note_redteamed_on_paper: true, crm_note_tabletop_stories: 6, crm_note_wording_gaps_found: 3, crm_note_wording_partial: 1, crm_note_v2_bounds_proposed: 4, crm_note_onet_anchors: 2, agent_executions: 0. Unchanged: blocked_candidate_workloads_documented: 4, blocked_or_paper_candidate_workloads_documented: 5, may_act_on_case: no, draft_agent_deployed: false. C-suite / ops / area actuals stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. WL-CRM-NOTE-61003 stays a paper candidate; v2 bounds proposed, not adopted. Do not deploy. Do not repair the fixture. Stock-check paused (DATA-004). Parts safety blocked (RISK-002). Gate 5 blocked.
Should the engagement freeze new workload mapping and write a single review request for a named reviewer outside the fiction (Craig to name one or decline), or fold the four v2 bounds into the Day 17 draft-only design (AGT-61003-DRAFT) first, still on paper?
Scenario date: 2026-11-13 (Friday) Gate: Gate 4/6 — revise block, eighth artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator) is the bound reviewer inside the fiction, and the person whose draft gets smaller today. Jonas Berg (billing analyst) and Priya Raman (data steward) are in view only as owners of the two blocked cards the draft must not touch. Rowan Ellis (fictional customer contact) is the person no draft may reach. No reviewer outside the fiction has been named, and none is invented. Stores/parts planner and field supervisor stay unnamed.
Day 29 asked: freeze new workload mapping and write a review request for a named reviewer outside the fiction, or fold the four v2 bounds into the Day 17 draft-only design (AGT-61003-DRAFT) first, still on paper?
The engagement's answer is to fold first. Nobody outside the fiction has been named, so there is nobody to send a review request to. If someone is named, they should get one document, not three (the Day 17 design, the Day 26 card, the Day 29 red-team). Folding needs no new card and no new owner. The reviewer seat is written into the v2 note as unfilled (Craig names one or declines). No new workload card was mapped today. Freezing is the Day 31 question.
The four Day 29 bounds now sit in a v2 design on paper. The forbidden list grows from 7 to 11: no restoration date stated, restated, or implied; a fixed header (Internal draft for the service coordinator. Not customer wording. Do not send.); the alias cited only as duplicate_suspect; nothing written to case history, with a rejected draft discarded.
Folding showed two things the red-team didn't:
assist) stays the task the draft serves. Task 2578 (Keep records of customer interactions, 43-4051.00, augment) moves out of the draft's scope: under the no-case-history bound, the record is Amina's, in her words. O*NET tasks in scope: 2 → 1. Labels unchanged.The stock-check gate (Day 16) stays in front of the draft, even though bound 1 means it never names a date. Removing one control because another now covers the same harm would be a loosening. Controls loosened: 0.
The Day 17 design file, the Day 19 handoffs table, and the Day 26 card are not edited. v2 is proposed, not adopted. Billing card WL-BILLING-HOLD-61003, merge card WL-BP-MERGE-61003, and parts cards WL-PARTS-20276 / WL-PARTS-2918 stay blocked. AGT-61003-DRAFT stays not_deployed. Nothing in data/sample/ or data/sample_r2/ changed.
Note: docs/engagement/pilot_design_agt61003_v2_day30.md.
| Movement | Detail |
|---|---|
| Opened | None. (Publish status: Opened on publish, 26 Oct 2026, as #26 (RISK-003).) |
| Comment (on publish) | RISK-001 (draft kept out of case history; transient store, prompt/output logging and retention unnamed; no period set). PEOPLE-001 (v2 answers Amina's scripted fear on paper; training ask rewritten; unseen by anyone doing the work; consultations: 0). BENEFIT-001 (the draft's scope shrank; still no benefit). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Labelled counts only (not KPIs): agt61003_v2_spec_on_paper: true, agt61003_v2_bounds_folded: 4, agt61003_forbidden_actions_listed: 11 (was 7), agt61003_onet_tasks_in_draft_scope: 1 (was 2), crm_note_v2_bounds_adopted: 0, controls_loosened: 0, new_workload_cards_mapped: 0, named_external_reviewer: none, agent_executions: 0. Unchanged: blocked_candidate_workloads_documented: 4, blocked_or_paper_candidate_workloads_documented: 5, may_act_on_case: no, draft_agent_deployed: false. C-suite / ops / area actuals stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues. AGT-61003-DRAFT v2 exists on paper, proposed, not adopted; reviewer seat unfilled. WL-CRM-NOTE-61003 stays a paper candidate. Do not deploy. Do not repair the fixture. Stock-check paused (DATA-004). Parts safety blocked (RISK-002). Gate 5 blocked.
With one paper spec and an empty reviewer seat, should the engagement freeze new workload mapping until Craig names a reviewer outside the fiction or declines in writing, or chase the question v2 surfaced — where a draft waits before Amina sees it, and who would log it (RISK-001) — still on paper?
Scenario date: 2026-11-16 (Monday) Gate: Gate 4/6 — revise block, ninth artefact Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator) is the only person who owns any stage of the draft's life, and only the two stages on her own desk. Rowan Ellis (fictional customer contact) is in view because her name and email sit in the case the prompt would read from. Jonas Berg and Priya Raman stay owners of blocked cards. No reviewer outside the fiction has been named, and none is invented. No privacy lead or data protection officer is invented to own the empty rows.
Day 30 asked: freeze new workload mapping until Craig names a reviewer outside the fiction or declines in writing, or chase where a draft waits before Amina sees it, and who would log it (RISK-001), still on paper?
The engagement does both, in that order. New workload mapping is frozen from today, until Craig names a reviewer outside the fiction or declines in writing. The freeze was deferred on Day 29 and Day 30; deferring it a third time would be drift dressed as diligence. Under the freeze, work is limited to finishing paper specs that already exist. Chasing the transient copy finishes v2, so it goes ahead.
The note follows one hypothetical draft for CASE-61003 through seven stages: prompt assembly, the request to a model host, the host's logs, the wait before Amina opens it, her screen and clipboard, after she accepts, and after she rejects. A copy could exist at all seven. A store is named at none. Retention is named at none. An owner is named at two, and both are Amina.
Two gaps v2 did not cover:
case_id, promise_type, stock_confirmed and the duplicate_suspect flag only.Both are proposed as v2.1 on paper, not folded. Folding is a reviewer's call now, not the author's, because the freeze is about exactly that. Bound B makes the RISK-001 question smaller. It does not answer it.
On the work map, task 18565 (Confer with customers, 43-4051.00, assist) stays the only task in the draft's scope. Labels unchanged. No new workload card. The Day 17 design, the Day 30 v2 note and the Day 26 card are not edited. AGT-61003-DRAFT stays not_deployed. Billing, merge and parts cards stay blocked. Nothing in data/sample/ or data/sample_r2/ changed.
Note: docs/engagement/draft_lifecycle_agt61003_day31.md.
| Movement | Detail |
|---|---|
| Opened | None. (RISK-003 remains proposed from Day 30; Craig decides.) (Publish status: Opened on publish, 26 Oct 2026, as #26 (RISK-003).) |
| Comment (on publish) | RISK-001 (seven-stage inventory; store, logging, retention unnamed at every stage outside Amina's desk; bounds A and B proposed). COST-001 (three unpriced lines identified). PEOPLE-002 (mapping frozen until a reviewer outside the fiction is named or Craig declines). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Labelled counts only (not KPIs): workload_mapping_frozen: true, agt61003_lifecycle_stages_inventoried: 7, agt61003_stages_with_named_store: 0, agt61003_stages_with_retention_named: 0, agt61003_v21_bounds_proposed: 2, agt61003_v21_bounds_folded: 0, unpriced_cost_lines_identified: 3, new_workload_cards_mapped: 0, named_external_reviewer: none, agent_executions: 0. Unchanged: agt61003_forbidden_actions_listed: 11, agt61003_onet_tasks_in_draft_scope: 1, controls_loosened: 0, may_act_on_case: no, draft_agent_deployed: false. C-suite / ops / area actuals stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Revise continues, now under a mapping freeze. v2 proposed, not adopted; v2.1 bounds A and B proposed, not folded; reviewer seat unfilled. WL-CRM-NOTE-61003 stays a paper candidate. Do not deploy. Do not repair the fixture. Stock-check paused (DATA-004). Parts safety blocked (RISK-002). Gate 5 blocked.
With mapping frozen and the paper spec as complete as an author can make it, should the engagement spend the freeze on DATA-004 (who would own the stock fact, still on paper), or stop adding artefacts and let the diary record a pause until Craig names a reviewer or declines?
Scenario date: 2026-11-17 (Tuesday) Gate: Gate 4/6 — revise block, held Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Amina Okonkwo (service coordinator) keeps the promise date, the record and the send, as she has since Day 17. She has not seen v2 or v2.1, and today nobody adds anything she would have to read. Jonas Berg and Priya Raman stay owners of blocked cards. Rowan Ellis (fictional customer contact) is unaffected. No reviewer outside the fiction has been named, and none is invented.
Day 31 asked: spend the mapping freeze on DATA-004 (who would own the stock fact, still on paper), or stop adding artefacts and let the diary record a pause until Craig names a reviewer or declines?
The engagement stops adding artefacts. Three reasons, all from the record rather than from preference:
So today the diary records the hold and changes nothing else. No new note in docs/engagement/. No new workload card. No bound folded. Task 18565 (Confer with customers, 43-4051.00, assist) stays the only task in the draft's scope. AGT-61003-DRAFT stays not_deployed. Billing, merge and parts cards stay blocked. Nothing in data/sample/ or data/sample_r2/ changed.
The hold ends on either of two events, and only these: Craig names a reviewer outside the fiction (then v2 and v2.1 go to that reviewer first, before any new mapping), or Craig declines in writing (then the freeze lifts, RISK-003 is opened so that the absence of review is on the record, and DATA-004 ownership is the next paper task).
| Movement | Detail |
|---|---|
| Opened | None. (RISK-003 remains proposed; it opens if Craig declines a reviewer.) (Publish status: Opened on publish, 26 Oct 2026, as #26 (RISK-003).) |
| Comment (on publish) | PEOPLE-002 (Day 32 records a hold: no new artefacts until a reviewer outside the fiction is named or Craig declines; two exits stated). |
| Closed | None. |
| Still open | DATA-001, DATA-002, DATA-003, DATA-004, PEOPLE-001, PEOPLE-002, COST-001, RISK-001, RISK-002, BENEFIT-001. |
No actual moves. Labelled counts only (not KPIs): engagement_on_hold: true, new_artefacts_today: 0, unpublished_drafts_awaiting_review: 2 (Days 30 and 31, before this one). Unchanged: workload_mapping_frozen: true, agt61003_lifecycle_stages_inventoried: 7, agt61003_stages_with_named_store: 0, agt61003_v21_bounds_proposed: 2, agt61003_v21_bounds_folded: 0, unpriced_cost_lines_identified: 3, new_workload_cards_mapped: 0, named_external_reviewer: none, agent_executions: 0, controls_loosened: 0, draft_agent_deployed: false. C-suite / ops / area actuals stay not measured.
Customer outcome, employee transition, and risk/cost stay equal. Pause beats a fake win.
Hold. Revise stays the recommendation, under the mapping freeze, with no new artefacts until Craig names a reviewer outside the fiction or declines in writing. WL-CRM-NOTE-61003 stays a paper candidate. Do not deploy. Do not repair the fixture. Stock-check paused (DATA-004). Parts safety blocked (RISK-002). Gate 5 blocked.
Not a new one. The same one, put to the human it belongs to: name a reviewer outside the fiction, or decline in writing. Until then, each weekday entry records the hold in a line and adds nothing.
Scenario date: 2026-11-18 (Wednesday) Gate: Gate 4/6 — revise block, held Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Hold continues, as Day 32 set out: no reviewer outside the fiction has been named and no written decline has been received, so nothing is added today.
engagement_on_hold: true, new_artefacts_today: 0, hold_days: 2. Everything else stays as Day 32; actuals not measured.Scenario date: 2026-11-19 (Thursday) Gate: Gate 4/6 — revise block, held Simulation: FrontierOrg is fictional. This entry is an activity log, not a measured result. Figures are fixture counts or marked not measured.
Hold continues. No reviewer outside the fiction has been named and no written decline has been received since Day 32, so nothing is added today.
engagement_on_hold: true, new_artefacts_today: 0, hold_days: 3. Everything else stays as Day 32; actuals not measured.diary/week-1-2026-10-08.md is the audience cut (board, investors, customers, employees) of the pause. Day 4 in this log is the same decision. Neither file is a measured result.
Options and the Day 22 recommendation live in docs/engagement/board_options_go_revise_stop.md. Not a board minute.
Design note: docs/engagement/stock_check_source_r2_design.md. Fixture r2 now sits alongside r1 at data/sample_r2/: a parts reservation behind stock_confirmed, owned by an unnamed role. data/sample/ (r1) is unchanged and stays the teaching case. DATA-004 open.
Desk sample: docs/engagement/parts_safety_sample_day24.md. 8 rows sampled, 2 candidate false negatives, 0 labels changed, no named safety reviewer. RISK-002 open.
Note: docs/engagement/blocked_workloads_parts_day25.md. WL-PARTS-20276 (human-only) and WL-PARTS-2918 (paper draft only, rating holder checks the specification) are both blocked. Human-only pattern list not grown, 0 labels changed, no named safety reviewer. RISK-002 open.
Card: docs/engagement/crm_note_workload_day26.md. WL-CRM-NOTE-61003 is a paper candidate, not deployed: it may draft wording for Amina only and must not change promise_date, write stock_confirmed, or touch an invoice hold. Billing assist not mapped as pilot. DATA-001 open.
Note: docs/engagement/blocked_workload_billing_day27.md. WL-BILLING-HOLD-61003 is blocked as a pilot: it may draft hold-reason wording for Jonas only and must not clear the hold, post or release INV-61003, process payment, or merge business partners. No finance control owner invented. CRM paper candidate unchanged. DATA-001 open.
Note: docs/engagement/blocked_workload_bp_merge_day28.md. WL-BP-MERGE-61003 is blocked as a pilot: it may draft a merge-recommendation note for Priya only and must not execute or propose the merge, silently resolve BP-61003-A / BP-61003, clear the invoice hold as a side effect, or change access levels. No steward invented. Fixture unrepaired. DATA-001 open.
Note: docs/engagement/redteam_crm_note_day29.md. WL-CRM-NOTE-61003 checked against six stories using fixture facts only. Three wording gaps and one partial gap; four v2 bounds proposed on paper. Agent executions: 0. Stays a paper candidate. DATA-001 and RISK-001 open.
Note: docs/engagement/pilot_design_agt61003_v2_day30.md. The four CRM bounds folded into AGT-61003-DRAFT on paper: 11 forbidden actions (was 7), output renamed internal status note, task 2578 out of the draft's scope, stock-check gate kept, controls loosened 0. Reviewer outside the fiction: unfilled. Proposed, not adopted. Agent executions: 0. RISK-001 open.
Note: docs/engagement/draft_lifecycle_agt61003_day31.md. Seven lifecycle stages for one hypothetical AGT-61003-DRAFT note: copy possible at 7, named store at 0, retention named at 0, owner named at 2 (Amina). Two v2.1 bounds proposed, not folded: discard on accept, and no contact name or email in the prompt. New workload mapping frozen. Agent executions: 0. RISK-001 and COST-001 open.
Issue: RISK-003 (#26). Craig declined to name an outside reviewer (26 Oct 2026). Reviewer seat: Declined by sponsor, 26 Oct 2026. Every boundary written since Day 24 is author judgement that nobody outside the fiction has read. Open.