Skip to content

GR Candidates — Session 2026-05-30/31

Guide

GR Candidates — Session 2026-05-30/31

Prepared by the Personal Assistant per GR-6 (session-close reconciliation). These are flagged candidates only — no rule is added without the Principal’s explicit approval (GR-3). Decisions were triaged against the GR bar: a Golden Rule is project-wide, durable, and prevents a class of failure. Most session decisions are policy-level or already captured in memory/policy; two cleared the bar.


Candidates

Candidate A — Amend GR-4 to clarify the AI-workforce interpretation

Decision: Org size is not a limiter. The GR-4 bar is “real need + clear purpose,” not “workload heavy enough to justify a human hire.” The lean/freeze stance was predicated on a human-staffing cost model that doesn’t apply here.

Why GR-level: GR-4 currently reads “analyze the need before you tool (or build)” with no explicit statement about headcount. Without an amendment, agents (and the Principal) could reasonably interpret the lean-default as still operative. Encoding it in GR-4 makes the AI-workforce stance authoritative and ensures future sessions don’t revert.

Form: amend GR-4 (not a new rule), adding a parenthetical or note. Suggested language to refine together:

For agent/role creation: the bar is clear purpose and defined scope, not workload volume. AI workforce headcount carries nominal cost; the real limiter is scope clarity.


Candidate B — New GR for archive integrity (the “never bulk-replace first/ rows” invariant)

Decision: Archive integrity is enforced by CI. When editing a source file, always refresh archive/documents/<slug>/latest/ and update MANIFEST.md checksums. Never use bulk hash replacement that touches first/ rows — those rows are write-once by design.

Why GR-level: This is a concrete, already-violated invariant (bulk hash replacement touching first/ rows is an irreversible corruption of the record). CI enforcement helps, but CI can be worked around; a Golden Rule makes the prohibition explicit and gives every agent a clear stop signal. It is project-wide (any change touching documented source files) and durable (the archive is permanent by design).

Suggested wording to refine together:

GR-7 — Protect the archive record. When editing a source file that is archived, refresh archive/documents/<slug>/latest/ and update the MANIFEST checksums in the same commit. Never modify or bulk-replace first/ rows — those entries are write-once. The CI archive verify check must pass before merge; never bypass it.


Not escalated (rationale)

Decision Why not a GR
Full org build-out (18 units, 36→35 agents, C-suite) Org-chart fact; policy captured in staffing-and-delegation.md
CPTO structure / Creative unit / Data & Analytics / AI Systems Structural decisions, not behavioral rules
Mentor offboarded One-time event; policy note already in staffing-and-delegation.md
Template descriptor “ideator · incubator · builder · operator” Configuration detail; captured in project memory
gh CLI for GitHub settings (never defer to manual) Feedback memory already written; GR-2 (use Claude Code workflows) covers the spirit of it
Customer Success naming stays Naming convention; policy/template guidance level
Confidant intentional exclusion from ORG.md Operational note; not a behavioral rule

Staffing policy — confirmed current

hr/policies/staffing-and-delegation.md already reflects the AI-workforce model (lines 42–58), the supersession of the lean/freeze stance, and the Mentor offboarding. No changes needed.