> Redacted excerpt. Employer-identifying detail and job-search working notes have been removed.
> The process structure and every self-correction are intact. The full document is not public.
> Source: `FLAGS.md` (private repo).
>
> Companion to [/writing/agentic-delivery/](https://jasonstiltner.com/writing/agentic-delivery/).

# Open questions

Where the brief asked for something the ground-truth inventory didn't support, the agent wrote
the question down instead of filling the gap. This is that list.

Its header, verbatim:

> Numbers, claims, and framing decisions I could not verify against the §5 ground-truth
> inventory, or where I made a judgment call that's worth double-checking. Nothing here is
> broken; everything here is a "please confirm" or "your call," not a bug.

The full list has 19 numbered items. **Six are reproduced below, verbatim. Thirteen are
withheld**, in four categories:

| Category | Items | Why |
|---|---|---|
| Identifies the current employer by description | 4 | Same reason the name itself is withheld. A unique description is a name. |
| Job-search working notes and private history | 5 | Resume and interview-prep drafts, and one personal-history item the brief explicitly instructs is not to be asserted anywhere. Not process. |
| Unresolved questions about specific live claims | 3 | Publishing "this number may be broader than reality, we're checking" next to the page making the claim is worse than either resolving it or removing it. These are open, and remain open. |
| Method used to probe repository visibility | 1 | Security method. |

Stating the count and the reason is the point. A list that silently published six of nineteen
and called itself "the open questions" would be the exact failure mode this process exists to
catch.

---

## Needs a decision

**1. Human throughline sentence** — three drafts in the review document, not yet applied to the
site. Pick one, pick none, or ask for different options.

**2. Work-history timeline (Accenture, Atiba Software, JLS Langues).** §5 lists these as
available facts, and I used the *arc* they describe (humanities → executive-communication
coaching → automation → ML) in the new Education/Background framing sentences. I did **not**
build a formal, dated Experience/work-history section listing them, because §5's own ordering
is ambiguous: "Earlier: Atiba Software" doesn't say earlier than what — before Accenture,
before Paris, or both. Guessing an order for real employment history felt like the wrong kind of
guess to make silently. If you want a proper timeline added (About page, most likely), tell me
the actual chronological order and I'll add it.

**3. HALA version history: 20/+8 vs. your brief's 18/+6 (which doesn't reconcile to 28
anyway).** I resolved this from the data's own `isNew` flags (8 patterns flagged new in v2 —
that's the actual per-pattern source of truth, not a hardcoded prose string). If that's not the
real v1/v2 split, tell me the correct numbers and I'll fix the version history string to match.

## Left untouched, unverifiable against §5

**6. "Based in Nashville"** in the root layout's meta description — pre-existing text I carried
forward unchanged (I only edited the surrounding sentence for the identity fix). Not in §5. Left
as-is since I didn't introduce it and have no reason to think it's wrong, but flagging per your
"don't assert what isn't covered" instruction.

**7. Mellifera, Agentic Orchestration, Strategy Tournament — content depth.** I surfaced these
three (linked them, fixed their SEO bugs, added a ProvenanceTag/repro check where relevant) but
did **not** do the same line-by-line stat-and-claim calibration pass on their body content that
GCL/CNL/Aegis got, since none of the three are in §5 and I had no ground truth to calibrate
against. I did confirm no dead links or "coming soon" stubs exist on any of them. If you want
the same §3 treatment applied to these three, say so and I'll do a follow-up pass.
**Update (Amendment 1):** Agentic Orchestration's worst issue — explicit job-seeking language in
two sections — is now fixed. The rest of that page's technical claims (line counts, specific
tech choices) still haven't been calibrated against any ground truth beyond what's visible in
its own code samples.

**9. Twitter handle, GitHub/LinkedIn URLs** (`@jasonlstiltner`, `github.com/jstiltner`, etc.) —
pre-existing, unchanged, not in §5. Assumed correct since they predate this pass.

## Scope notes (not really flags, just stated for completeness)

- Per-page OG images: `og:image` is per-page-correct now (via `buildMetadata()`) but every page
  currently points at the same shared `/og-image.png` — no custom per-project OG images exist as
  assets. Brief said "where available"; none were available.
- "Updated Jan 2026" recency markers were added to the project entries on `/research` only, not
  to the individual project pages themselves or to About/Homepage.
- The `/research/hala` → `/writing/hala` and `/tools` → `/about` redirects return HTTP 308, not
  301 — Next.js's `redirects()` has no literal 301 option; 308 is the standard permanent-redirect
  equivalent and preserves SEO link equity the same way.