TeamBot public-evidence boundary

[ojfbot]

Question

What of the SAP/TeamBot work is usable as public FDE evidence — where is the IP / employer-policy boundary, and what is the sanitized publishable form (pattern-level write-ups, no employer specifics)?

Type

grilling (HITL, /grill-with-docs)

Context

The positioning frame leans on TeamBot ("if SAP paid Anthropic, what would an FDE team of 3 deliver, working with a McKinsey consultant for process understanding"). A wrong answer here invalidates any narrative artifact that references day-job work — this must be settled before the narrative ticket, and the decision is the operator's alone.

Blocked by

None — on the frontier.

Resolution (filled at close)

[ojfbot]

Input from the #320 re-audit (decision remains open here): the two most FDE-legible day-job stories are (a) the buy-in/change-management narrative around the Socratic review gate (genre-4 competency evidence — the corpus's dominant cost center), and (b) quota governance as production literacy (genre-6 supporting line — H1 refuted as a headline, so this is a supporting credit, not the lead). Both need this ticket's IP ruling before any public use; the audit doc kept all day-job references pattern-level.

[ojfbot]

Urgent input from the #320 re-audit (second cycle, #337 §7). Ticket body stands and is not invalidated — this only makes it the next ticket to work.

Verified directly against origin/main of public ojfbot/core this session, not taken from a sub-agent report:

• git grep -w -E "SAP|TeamBot" origin/main returns 9 files.
• decisions/wayfinder/fde-operating-presence.md:19-20 binds the internal system nickname to the employer name and quotes a strategic framing sentence, in one place.
• decisions/research/2026-08-01-fde-deliverables-audit.md carries the employer name and a "Fleet + TeamBot scorecard" section.
• The other seven files use the nickname as a bare alias on generic agent-harness mechanisms (/resume + its two scripts, three draft ADRs, day-runner.mjs). On their own that is a blind alias — but once one public file binds the alias to the employer, all of them become retroactively attributed. That compounding is the actual exposure, not any single file.

Timing, stated as fact without inferring intent: this ticket was opened at 2026-08-01T19:00:28Z and is still OPEN. PR #332 — which merged material naming the employer to public main — was created at 19:38:49Z and merged at 19:39:56Z, ~39 minutes later, by an automated session. The ticket that governs the material was open the whole time.

Nothing was edited. This is a grilling ticket; the ruling is yours. An agent stripping strings from public canon ahead of the decision would be the wrong move, and git history retains them regardless — a history rewrite is a separate and much larger decision.

Adjacent hazard worth closing regardless of the ruling: core-library/public/graph/selfco.json is a ~1.5 MB untracked build artifact carrying a TeamBot-tagged node title, sitting in a directory named public/ with no gitignore entry. One git add -A publishes it.

What the audit found is publishable at pattern level, if you rule that way: the harness discipline is already public in core as generic mechanisms (preflight state verification, git-reconstructed session reports, self-contained briefs, the /merge-quiz comprehension gate) — the sanitized version already exists; what would need removing is attribution, not content. The adoption story — how you get engineers to accept a gate that slows them down — is the most FDE-legible thing available (it is checklist 2.2, 2.8 and 2.12 in one narrative), and is tellable only stripped of employer, team size, internal metrics, tool names and org structure.

Not publishable on the evidence: the employer↔system binding in any form; anything from the primary day-job corpus (it lives on an unmounted volume, was not inspected, and if built on company time is plausibly employer-owned); internal metrics, adoption rates, team sizes, org and customer names.

[ojfbot]

Resolution

Grilled via /grill-with-docs, four operator decisions:

	Decision
Boundary	Unnamed, pattern-level. Named-with-sign-off and no-day-job-material both rejected.
Threshold	Stranger test — a reader with no prior knowledge of the operator's employment cannot identify employer, system, or team. Re-identification by someone who already knows is accepted, not defended against.
Enforcement	By construction — denylist lint sourced from gitignored personal-knowledge/, failing any tracked file that matches. Not author discipline.
History	Scrub HEAD now; stage the 10-commit rewrite for a separate quiesced operation.

Why mechanical enforcement: the one documented breach of this boundary was committed by an automated session, which merged employer-naming material to public main 39 minutes after this ticket opened. Agents write to these repos and do not read policy before merging. Discipline has one observed failure here and zero observed successes. Same posture as dive-briefing's private_mount tier — unpublishable by construction, not by care.

Denylist placement: it cannot be committed. A list of the terms being protected is the disclosure it exists to prevent. personal-knowledge/ is gitignored with zero tracked files — the one boundary that held cleanly through this whole episode.

What landed — PR #343

• HEAD scrubbed. Employer name: 3 lines across 2 files → 0 occurrences in tracked files. The load-bearing one was decisions/wayfinder/fde-operating-presence.md:19-20, the only place binding nickname → employer → strategic framing. The 9 bare-alias files stay; unbound they pass the stranger test, and the lint sweeps them.
• decisions/open-unknowns.md created — the grill's durable artifact, first in this repo.
• Map tended: gist appended, this ticket closed, Positioning narrative + venue map #328 flagged partly invalidated, and the TeamBot-pattern-flow fog graduated to Sanitized pattern-flow loop #342 now that the ruling makes its question statable.

Deliberately not done

• History rewrite — 10 commits still carry the strings (earliest 2026-06-25). Force-pushing main against ~23 live agent sessions and 3 worktrees is a bigger risk than a history exposure that 0 forks / 0 stars say nobody holds. Deferred, logged, and it will not happen unless scheduled explicitly.
• The lint — a delivery, not a decision. Belongs on a roadmap, not this map (two-ledger rule).
• ADRs — three stubs drafted for /adr new: employer-evidence-boundary, boundary-enforced-by-construction, staged-history-remediation.

The assumption this rests on

The ruling was made without either party reading the actual employment agreement or publication policy. If that policy requires pre-approval for any work-derived publication, this ruling is void and the boundary collapses to employer-indifferent. Recorded in the ledger rather than buried; worth reading before the first pattern-level artifact publishes.

Unblocks

#328 — but its body needs re-authoring, not trimming: the proposed frame names the employer, and per the re-audit's §2.11 (paired engagement owner, convergence 4) the consultant-pairing instinct is right while the role assignment is inverted — the FDE is the engineer in that pair. A narrative owning ROI and value cases pitches the Technical Deployment Lead role instead.

Live hazard, flagged not fixed

core-library/public/graph/selfco.json — 1.5 MB, untracked, not gitignored, public repo, protected term in a node title. One git add -A publishes it. Different repo; wants a one-line .gitignore entry.