docs(rc1): partner outreach checkpoint (Tier-2 B)

[trilamsr]

Summary

Ships rc1 Tier-2 item B (Partner outreach checkpoint). Adds docs/partner-outreach.md — the outreach-ready package — and flips the cut-criterion status from ☐ planned → ⧗ in progress by replacing the hard-coded artifact_exists: "false" with a real artifact-presence + tracker-row-count gate.

The artifact carries (per the §Tier-2 B rubric):

1. Persona list — 12 operator archetypes (P1-P12: GPU-cloud providers, frontier-model labs, hyperscaler ML infra teams, OTel-adjacent maintainers, mid-cap AI startups, HPC centers, GPU-renting marketplaces, observability vendor PMs, kernel-engineering shops, GPU-renting end customers, CNCF TAG-Observability, MLOps community organizers). Each archetype carries a one-line "why they care".
2. Outreach message template — ~190-word cold-pitch with a specific ask (install the production-preset Helm chart and tell us what broke), plus audit-funding + OSS-maintainer variants.
3. Talking points — seven bullets covering the 15-pattern coverage story, the RFC-0013 OCB-distro pivot, the why-now structural signal, three-layer differentiation (vs Prometheus + DCGM-exporter alone; vs Datadog Watchdog; vs vendor-tied exporters), production posture, what-we-are-NOT framing, and the conversation ask.
4. Tracker table shape — columns (name, role, org-archetype, first contacted, status, notes), six-token status vocabulary (queued / contacted / meeting-set / feedback-received / pilot-deploy-attempted / declined / deferred), and four hygiene rules (one row per person, one-line notes, no private contact info, honesty in declined/deferred rows).
5. Success criteria — SC-1 (≥5 contacted), SC-2 (≥2 feedback responses), SC-3 (≥1 pilot-deploy attempt). SC-3 is the GA blocker.

Why ⧗ in progress, not ☑ shipped

The artifact ships today (file-exists passes); the gate script counts non-placeholder rows in the §Tracker table and requires ≥3 real entries before flipping ☑. Today's count is zero (only the column-shape placeholder row exists), so status correctly reads ⧗ in progress. Honest about the state.

Root cause addressed

The previous YAML hard-coded artifact_exists: "false" with a comment noting "this is tracked on the issue tracker." That made the criterion ungatable — no PR could close it because there was no falsifiable artifact in the repo. This PR ships the artifact at a stable path so the gate becomes file-driven (the same shape as Tier-2 A's RFP) and the path forward to GA is mechanical: add tracker rows, re-render, status flips.

Cross-links (A+ scope)

• docs/NORTHSTARS.md §O5 — adds a "Partner outreach plan" paragraph pointing at docs/partner-outreach.md.
• docs/RELEASE-CHECKLIST.md GA additional gates — adds a "Partner outreach checkpoint cleared" row before "Real-operator integration verified"; SC-3 row is the same operator as the case-study writeup.
• Cross-link from docs/security-audit-rfp.md §Funding plan + gap is already present (the audit-funding variant in §2 carries the pointer back).

Files touched

• docs/partner-outreach.md (new, 389 lines) — the artifact.
• docs/cut-criteria.yaml — Tier-2 B rubric_check now exercises test -f docs/partner-outreach.md + tracker-row awk gate; rubric prose + notes updated.
• docs/v1-rc1-cut-criteria.md — re-rendered via make cut-criteria-render MILESTONE=all. Not hand-edited.
• docs/NORTHSTARS.md — §O5 cross-link to the artifact.
• docs/RELEASE-CHECKLIST.md — GA gate matrix row.

Test plan

• make doc-check clean (includes cut-criteria-check, slo-rules-check, chart-appversion-check, rfc-status-check, alert-check, link-rot, YAML cross-link)
• bash scripts/validator-recipe.sh clean
• make cut-criteria-status MILESTONE=v1.0-rc1 shows B = ⧗ in progress
• make cut-criteria-render MILESTONE=all produces no diff against committed v1-rc1-cut-criteria.md / v1-ga-cut-criteria.md
• No 🤖 Generated with Claude Code footer
• DCO sign-off on commit
• CI doc-check / cut-criteria-check / dco-check green on push

Follow-up

After this PR merges I will file a tracking issue with the rc1-prep label for "execute partner outreach: contact P1-P12 archetypes per docs/partner-outreach.md" so the §Tracker rows accumulate against a real owner. The issue is the unblock signal once SC-1/SC-2/SC-3 land.

Adds the v1.0 rc1 → GA partner-outreach plan at `docs/partner-outreach.md` — 12-archetype persona list, cold-pitch template, talking points, tracker shape, numeric success criteria (≥5 contacted, ≥2 feedback, ≥1 pilot-deploy). Flips rc1 Tier-2 B cut criterion from ☐ planned → ⧗ in progress.

[trilamsr]

Independent Adversarial Review: PR #470 (rc1 Tier-2 B partner outreach)

B/A/A+ Criteria for this PR:

• Artifact completeness: 12-archetype persona list (✓), cold-outreach pitch ≤200 words (✓), talking points crib (✓), tracker shape (✓), numeric success criteria (✓).
• Falsifiability: persona list is concrete with org examples; success criteria are numeric (5/2/1 ladder); gate_script is executable.
• Style parity: matches security-audit-rfp.md structure (status table, "read in order", companion artifacts); cross-links to NORTHSTARS, RELEASE-CHECKLIST, cut-criteria.
• Simplification + comment discipline: outreach doc is proportional (297 lines); cut-criteria.yaml notes section is justified but borders redundancy.

Critical Issues:

1. BUG: gate_script threshold misaligned to success criteria — Line 102 of cut-criteria.yaml checks n >= 3 to flip Tier-2 B status to ☑, but success criterion SC-1 (line 379 of partner-outreach.md) defines "≥5 operators contacted" as the actual threshold. The comment on lines 80–84 (cut-criteria.yaml) incorrectly claims "at least three real rows." This breaks the automated cut-criteria render gate; the criterion will flip to ☑ when only 3 operators are contacted, not 5. The 5/2/1 ladder is stated as a contract (partner-outreach.md §5 bottom), so lowering to 3 violates the Tier-2 rubric.

   • Fix: Change line 102 to exit (n >= 5 ? 0 : 1) and correct the comment to read "at least five real (non-placeholder) rows."

2. RISK: talking-points count inconsistency — Line 261 (partner-outreach.md) claims "Six points; each anchors a question…" but the numbered list contains 7 bullets (points 1–7, lines 264–320). The cut-criteria.yaml notes (line 110) correctly state "seven live-conversation talking points." This is confusing for someone implementing the strategy; an implementer reading §3 intro may prepare for 6 and miss point 7.

   • Fix: Change line 261 to "Seven points; each anchors a question the persona is likely to ask back."

3. RISK: "queued" status excluded from SC-1 but not explicit — The gate_script correctly excludes "queued" from the 6 valid status tokens (lines 96–98), but success criterion SC-1's definition doesn't explicitly enumerate the valid set; it says Status ∈ {contacted, meeting-set, feedback-received, pilot-deploy-attempted, declined, deferred} (line 379), which omits "queued." A tracker row marked "queued" will not count toward the 5-row threshold, even though "queued" is a legitimate status. This is actually correct behavior (queued ≠ contacted), but the SC-1 definition should match the gate_script logic for clarity.

   • Fix: Ensure SC-1 definition (line 379) explicitly lists all 6 tokens: add "queued" to the set or note that queued rows are explicitly not counted toward SC-1.

4. NIT: link style inconsistency — Partner-outreach.md uses both bare relative links and inconsistent markup. Compare to security-audit-rfp.md, which consistently uses backtick-markdown style. Not load-bearing but violates parity.

5. NIT: status table redundancy in cut-criteria.yaml notes — Lines 105–119 (cut-criteria.yaml §notes) repeat nearly all artifact-summary claims from partner-outreach.md lines 130–138. Consider condensing notes to link and summary. Saves ~70 bytes and deduplicates.

---

Verdict: BLOCK pending fix of critical bug

The 3 vs. 5 threshold mismatch is a show-stopper — it would cause the GA release checklist gate to falsely flip ☑ when the actual success criterion has not been met. Fix lines 102 and 80–84 (cut-criteria.yaml), confirm line 261 (partner-outreach.md) is updated to "Seven points," and re-render the cut-criteria table to validate gate correctness. Once done, this is A (ship after simplification).

The redundancy nit and link-style nit are optional cleanup; the risk on queued-status clarity is low-severity (behavior is correct, just under-documented).

[trilamsr]

BLOCK fixes applied: (1) gate_script 3→5 matches SC-1; (2) header point-count fixed; (3) queued explicitly excluded from gate. Re-requesting review.

[trilamsr]

Independent Adversarial Review: PR #470 POST-BLOCK-FIX (commit 3f419a6)

Per-Fix Verification

1. gate_script n >= 3 → n >= 5 ✓

   • Line 494 (docs/cut-criteria.yaml): END { exit (n >= 5 ? 0 : 1) }
   • Executable gate now matches SC-1 threshold correctly.

2. Header "Six points" → "Seven points" ✓

   • Line 133 (docs/partner-outreach.md): "Seven points; each anchors a question the persona is likely to ask back."
   • Consistent with §3 content (7 numbered bullets).

3. SC-1 row clarifies "queued doesn't count toward 5" ✓

   • Lines 251–252 (docs/partner-outreach.md SC-1): queued rows do not count toward the 5 — outreach is "actually sent", matching the executable gate.
   • Binds written contract to gate script.

Threshold Flip-Test: PASS

• 4 tracker rows → gate FAILS (⧗ in progress)
• 5 tracker rows → gate PASSES (☑ complete)

Simplification Sweep: NO NEW ISSUES

All BLOCK findings closed. Optional nits (redundancy + link style) remain unfixed per builder choice.

VERDICT: A (Ship)