wayfinder: resolve TeamBot public-evidence boundary (#321)

decisions/open-unknowns.md

@@ -0,0 +1,35 @@
+# Open unknowns
+
+Deferred decisions, unvalidated assumptions, and domain-standard considerations a grill did not
+reach. Appended by `/grill-with-docs` at the confirmation gate; never rewritten in place.
+
+The third bucket is a checklist of considerations the grill skipped — **not** a guarantee of
+blind-spot coverage. An empty bucket is a legitimate result.
+
+---
+
+## 2026-08-01 — employer-evidence boundary for public FDE artifacts (wayfinder #321)
+
+Ruling: day-job material is publishable only as de-identified engineering patterns passing the
+**stranger test**, enforced by a denylist lint sourced from gitignored `personal-knowledge/`.
+HEAD scrubbed on decision; history rewrite staged. See `adr:employer-evidence-boundary`,
+`adr:boundary-enforced-by-construction`, `adr:staged-history-remediation`.
+
+**Deferred decisions**
+- Git-history rewrite of the 10 commits carrying the protected strings (earliest 2026-06-25) — unblocked by: in-flight PRs landing (#337 among them) plus an agent-quiescent window. Force-pushing `main` against ~23 live agent sessions and 3 worktrees is the risk being deferred around, not the rewrite itself.
+- Whether the denylist lint covers sibling public repos (`daily-logger`, `shell`, `core-library`, `landing`) or only `core` — unblocked by: the lint's roadmap-slice scoping.
+- Whether GitHub support is asked to GC unreferenced objects after any rewrite — a rewrite alone leaves objects reachable by SHA.
+- Whether the primary day-job corpus (unmounted external volume, never inspected) is ever examined. Nothing from it can be ruled publishable until someone has looked at it.
+
+**Unvalidated assumptions**
+- ~~That the employer's employment agreement and publication policy permit de-identified, pattern-level technical writing.~~ **RESOLVED 2026-08-01 — operator confirmed** ("I am good on employment agreement"). The ruling in `adr:employer-evidence-boundary` no longer rests on an unread policy. Recorded rather than deleted: this was the load-bearing assumption, and a future reader should be able to see that it was closed by operator confirmation rather than by document review.
+- That the stranger test is actually satisfiable for the adoption story specifically — no draft exists yet, and none has been tested against a cold reader.
+- That a term denylist is sufficient protection. It catches names; it does not catch situational fingerprints (a distinctive metric, an unusual org shape, a dated incident).
+- That the 9 remaining bare-alias files pass the stranger test now that the binding is removed. Reasoned, not reader-tested.
+
+**Standard considerations not covered**
+- Non-git surfaces were never swept: published daily-logger articles, Notion, selfco exports, and the Vercel-hosted sites may carry the same strings. Only the git surface was checked.
+- LinkedIn, X, and résumé name the employer by design and sit outside this boundary entirely — no ruling was made on how they interact with de-identified public artifacts that a reader could correlate.
+- No human review step for future narrative drafts before publish, beyond the mechanical lint.
+- ~~`core-library/public/graph/selfco.json` — untracked, not gitignored, in a public repo.~~ **FIXED 2026-08-01** — `core-library@ccbf853` ignores `public/graph/` wholesale. The git vector is closed.
+- **Open, and larger than the one just fixed: `core-library` serves the vault graph at runtime.** `apps/web/src/main.ts:82` fetches `/graph/selfco.json`, and the vault-ingest Vite plugin regenerates it into `publicDir` on every build. The repo is not currently deployed (no `vercel.json`, no `.vercel`, no `netlify.toml`), which is the only reason this is contained. **If core-library is ever deployed, the build republishes the entire private vault graph to a public URL, and `.gitignore` does nothing to stop it.** Under `adr:boundary-enforced-by-construction` this wants a build-time guard (de-identify at ingest, gate the fetch behind local-only, or exclude protected nodes from the graph), not a note. Decision, not yet made — deploying core-library is currently blocked on this.

decisions/research/2026-08-01-fde-deliverables-audit.md

@@ -69,7 +69,7 @@ Every role definition across all five segments centers the full arc; harness art
 
 1. (#328) Should the public narrative lead with the one genre the market screens hardest (eval frameworks) or with the rarer combined artifact (full-arc case study)?
 2. (#327) Does dive-briefing's roadmap need an explicit "engagement simulation" slice (discovery doc + stakeholder/plumbing narrative) to cover genre 3/4, or is that a separate artifact?
-3. (#321) Is the pr-quiz buy-in story tellable at pattern level without SAP specifics — and does SAP policy allow even that?
+3. (#321) **RESOLVED 2026-08-01** (`adr:employer-evidence-boundary`): tellable at pattern level under the stranger test — de-identified, no employer or system name, no team size, no internal metrics. Whether employer policy permits it is an unvalidated assumption recorded in `decisions/open-unknowns.md`.
 4. (follow-up cycle, if warranted) The skeptic sweep this cycle lacked: FDE-critical sources, attrition lore, buyer-side outcome evidence.
 
 ## Key sources

decisions/wayfinder/fde-operating-presence.md

@@ -16,10 +16,11 @@ published FDE-genre deliverables (discovery documents, eval harnesses with measu
 deployment runbooks, process-understanding artifacts of the "McKinsey consultant" kind), and a
 **running content flywheel** — BlogEngine emitting to GitHub + Medium + blog.jim.software +
 LinkedIn + X + Instagram + a new dedicated tech-blog domain — fed by **scheduled trend ingestion**
-(thought leaders, model releases, harness-architecture changes) into selfco. TeamBot at SAP is the
-day-job instance of the same posture ("if SAP paid Anthropic, what would an FDE team of 3
-deliver, working with a McKinsey consultant for process understanding"); this map covers the
-personal-stack half only, with the employer boundary explicitly decided. Offers are a lagging
+(thought leaders, model releases, harness-architecture changes) into selfco. The day job is the
+enterprise-scale instance of the same posture ("what would an FDE team of 3 deliver, working with
+a McKinsey-style consultant for process understanding"); this map covers the personal-stack half
+only, and the employer boundary is decided in `adr:employer-evidence-boundary` — day-job material
+enters public artifacts only de-identified, at pattern level. Offers are a lagging
 indicator, not the success object. Serves `ns:l2-ojfbot#P1` (usable surfaces — the published
 artifacts ARE surfaces shown where they live), `ns:l2-ojfbot#P2` (work is legible — the flywheel
 publishes the legibility), and `ns:l3-shared#P1` (a stranger can be shown the cluster doing real
@@ -49,6 +50,12 @@ work end-to-end).
   its credibility ceiling); `wiki/synthesis/value-proposition-canvas-bmc-fit` and
   `concepts/value-proposition-design` (positioning-narrative instrument for the venue-map
   ticket).
+- **#328's ticket body is partly invalidated by #321's ruling (2026-08-01).** Its proposed public
+  frame names the employer, which `adr:employer-evidence-boundary` now forbids; and the re-audit's
+  §2.11 (paired non-engineering engagement owner, convergence 4) shows the consultant-pairing
+  instinct is right but the role assignment inverted — the FDE is the *engineer* in that pair, so a
+  narrative owning ROI and value cases pitches the Technical Deployment Lead role instead. The venue
+  map half of that ticket is unaffected. Re-author the frame; do not trim it.
 - Ingestion automation must respect the sequential-research rule (one deep-research cycle at a
   time; 2026-06-05 saturation failure) and shadow-first discipline (observe-only sweeps before
   unattended vault writes — ADR-0089 idiom).
@@ -64,14 +71,15 @@ Pre-map rulings (grilled at charting, recorded here — not closed tickets):
 
 Closed tickets:
 
-- Primary-source FDE checklist built (6 genres; eval frameworks = the hiring gate, "not a demo role"); H1 cost-orchestration REFUTED as FDE deliverable, H2 eval-gating MIXED (core supported, pr-quiz sub-claims unevidenced, capture loop runs vendor-ward), H3 customer-facing core SUPPORTED — FDE deliverables checklist + fleet re-audit (#320) → decisions/research/2026-08-01-fde-deliverables-audit.md
+- Primary-source FDE checklist built (6 genres; eval frameworks = the hiring gate, "not a demo role"); H1 cost-orchestration REFUTED as FDE deliverable, H2 eval-gating MIXED (core supported, pr-quiz sub-claims unevidenced, capture loop runs vendor-ward), H3 customer-facing core SUPPORTED — FDE deliverables checklist + fleet re-audit (#320) → decisions/research/2026-08-01-fde-deliverables-audit.md (+ independent second cycle, corroborating and correcting → decisions/research/2026-08-01-fde-deliverables-audit-cycle-2.md)
+- Day-job material is publishable only as de-identified patterns passing the **stranger test**, enforced by denylist lint (not discipline); HEAD scrubbed on decision, history rewrite staged — TeamBot public-evidence boundary (#321) → `adr:employer-evidence-boundary`, `adr:boundary-enforced-by-construction`, `adr:staged-history-remediation`
 
 ## Tickets
 
 | Ticket (title, refer-by-name) | Type | Blocked by | Status |
 |-------------------------------|------|------------|--------|
 | FDE deliverables checklist + fleet re-audit (#320) | research | — | closed |
-| TeamBot public-evidence boundary (#321) | grilling | — | open |
+| TeamBot public-evidence boundary (#321) | grilling | — | **closed** |
 | Flywheel first loop (#322) | grilling | — | open |
 | Ingestion source roster + cadence (#323) | grilling | — | open |
 | Ingestion runtime + host (#324) | grilling | Ingestion source roster + cadence | open |
@@ -81,6 +89,7 @@ Closed tickets:
 | Positioning narrative + venue map (#328) | grilling | FDE re-audit; Public identity + domain; TeamBot boundary | open |
 | BlogEngine distribution seam (#329) | grilling | Flywheel first loop; Public identity + domain | open |
 | Account provisioning (#330) | task | Public identity + domain | open |
+| Sanitized pattern-flow loop (#342) | grilling | — | open |
 
 Frontier at charting: FDE re-audit, TeamBot boundary, Flywheel first loop, Ingestion source
 roster, Handle availability survey. Account provisioning is **user-performed** (agents never
@@ -99,9 +108,8 @@ create accounts or handle credentials); Handle availability survey is survey-onl
   pipeline"; the Destination ruling (presence, not business) leaves the commercial question
   unstated, not ruled out. Revisit after the narrative lands; graduating or ruling it out is the
   operator's call.
-- **TeamBot↔personal-stack pattern flow** — whether sanitized TeamBot patterns become fleet
-  artifacts (and vice versa) as a standing loop; statable only after the TeamBot boundary is
-  decided.
+- ~~**TeamBot↔personal-stack pattern flow**~~ — **graduated to a ticket 2026-08-01**: the boundary
+  ruling (#321) made the question statable. Now "Sanitized pattern-flow loop" (#342).
 
 ## Out of scope