phData cert progress — source of truth
This file is the canonical progress record read by /cert-study (cron + on-demand).
Update it after any study session, exam attempt, or blocker change.
The skill reads it to derive verdicts; it never overwrites entries — only appends.
Partner-portal access (Cert 2 gate)
| Status | Date confirmed | Notes |
|---|---|---|
| ☐ not yet provisioned | — | Required before Cert 2 (Anthropic) study materials are accessible. Action: follow up with phData once onboarding settles. |
2026-06-13 (iMessage): Founder is about to start a Claude Architect cert prep course with practice exams — so either the partner-portal gate above has effectively cleared or he's using external/public prep. ❓ Confirm gate status on next
/cert-study. Advised diagnostic-first: take ONE practice exam cold before the course (pretesting effect + triages which modules to skim vs sit through — high baseline since he builds agent architectures daily), and save remaining exams for spaced post-course retrieval + a near-final readiness check.
Cert 1 — Snowflake GenAI Specialty (GES-C02)
Deadline: 2026-08-24 (90d from job start 2026-05-26) Exam scheduled: ☐ not yet scheduled Track selected: — (Track A = 7-10 day sprint recommended; Track B = 3-4 week steady) Exam registration: not yet registered
Study log
| Date | Day/Topic | Time spent | Notes |
|---|---|---|---|
| no entries yet |
Weak areas (populated by /cert-study quiz)
none logged
Attempt log
| Date | Outcome | Score | Notes |
|---|---|---|---|
| no attempts |
Cert 2 — Anthropic Claude Certified Architect – Foundations
Deadline: 2026-11-22 (180d from job start 2026-05-26) Exam scheduled: ☐ not yet scheduled (gated on partner-portal access) Current phase: Phase 1 — D1 Agentic Architecture (target: Jun 2026)
Study log
| Date | Phase/Domain | Time spent | Notes |
|---|---|---|---|
| formal materials gated on partner-portal access | |||
| 2026-06-10 | D1+D4+D3 quiz (via /cert-study on-demand) | ~15 min | 4/5 — see quiz log. Weak area: D3 API specifics (Message Batches, tool_choice, streaming) — daily Claude Code use doesn't teach the raw API surface. |
Weak areas (populated by /cert-study quiz)
- D3 API specifics (logged 2026-06-10, quiz 1): Message Batches API — founder "no clue" (taught: 50% discount, ≤24h no-SLA, single-shot/no tool loop). Drill next:
tool_choicemodes (auto/any/forced), streaming event sequence (message_start → content_block_delta → message_stop), extended-thinking token budgeting. Root cause: daily Claude Code use never touches the raw API surface. CONFIRMED on exam 2 (Q7 tool_choice, Q33 prefilling, Q52 MCP notifications) — promote to active drill. - Hook scoping (logged 2026-06-27, exam 2; recurring from exam-1 Cluster A): tool-specific hooks only fire for that tool — Write hook ≠ Bash
echo > file; comprehensive protection needs Write+Bash hooks (+.claudeignore+OS perms); exit-code 0=allow/non-zero=block. Drill from [[2026-06-14-cca-harness-mechanics-cheatsheet]]. - "Most-complete answer" test-craft (logged 2026-06-27, exam 2; ~6 Qs): on BEST/most-robust/MOST-addresses stems, pick the layered/both-halves/one-notch-further option, not the clean single-mechanism one. Not a knowledge gap — a test-taking habit. Highest-ROI fix.
- Act-vs-ask calibration (logged 2026-06-27, exam 2; ~4 Qs, judgment): the autonomy-threshold rubric (additive→accommodate; material change→surface; divergent outputs→present-both-if-cheap; self-param-change w/o goal-check→risk).
- Context economy (logged 2026-06-27, exam 2; ~2 Qs): right-size output to consumer's decision; progressive summarization across sessions.
Attempt log
| Date | Outcome | Score | Notes |
|---|---|---|---|
| 2026-06-14 | PASS (practice, pre-study) | 50/60 = 83% (pass = 72%) | First practice exam, COLD, before the course. Instructor benchmark: most get 50-60% after the course → founder well above post-course cohort pre-study. 10 misses (founder read: only 2 real guesses, 1 broken question). Pattern + fixes → [[2026-06-14-cca-harness-mechanics-cheatsheet]]. |
| 2026-06-27 | practice (borderline) | 42/60 = 70% | Pass line recorded as 72% (⚠️ source unverified — CONFIRM real cut score; founder reports 70% passes). Finished 12 min to spare. Down 13 pts from exam 1 — but it's targeting not regression: misses concentrate in 1 test-craft pattern + 2 knowledge gaps + 1 judgment axis. All 18 misses analyzed → [[2026-06-27-cca-practice-exam-2-miss-analysis]]. |
Miss pattern (practice exam 1, 2026-06-14)
- Cluster A — harness mechanics/timing (~half, highest ROI): Stop-hook per-turn (Q48 ✓), PostToolUse timing (Q10 — exam answer shaky vs current docs: PostToolUse CAN modify output before context), MCP
resources/updatednotify→client-refetch (Q27 ✓), subagents parent-mediated (Q34 ✓), skill triggers intent-not-keyword (Q42 ✓). Verified vs docs.claude.com via claude-code-guide. - Cluster B — plausible-neighbor definitions: idempotency≠atomicity (Q12), episodic/semantic by specificity-not-duration (Q22).
- Cluster C — "BEST/canonical" vs pragmatic-builder answer: fan-out/barrier/fan-in (Q24), subagent-delegation for context isolation (Q35).
- Low-signal near-duplicate-answer traps: Q26 (has an editing artifact in its own explanation), Q42.
- D4: clean sweep (0 wrong).
Miss pattern (practice exam 2, 2026-06-27)
All 18 misses → full breakdown [[2026-06-27-cca-practice-exam-2-miss-analysis]]. Five threads, ROI-ranked:
- "Most-complete answer" under-selection (test-craft, ~6 Qs, highest ROI): Q4/Q5/Q19/Q56/Q18 (+Q23 flavor). On BEST/most-robust/MOST-addresses stems, founder picks a clean single-mechanism answer; key is the layered/both-halves/one-notch-further option. Fix = a test-taking rule, zero study.
- Hook scoping (knowledge, ~3 Qs, RECURRING): Q42/Q5/Q23. Tool-specific hooks only intercept that tool — a Write hook does NOT catch Bash
echo > filewrites; exit-code 0=allow/non-zero=block. Was Cluster A on exam 1 → persisting. Re-read [[2026-06-14-cca-harness-mechanics-cheatsheet]]. - API surface (knowledge, ~3 Qs): Q7/Q33/Q52.
tool_choiceforces a tool call (+downstream validation); prefilling{locks JSON vs override framing; MCP capability-change notifications are protocol-level/transport-agnostic. = the "D3 API specifics" gap flagged 2026-06-10 (tool_choice named then), now confirmed. - Act-vs-ask calibration (judgment, ~4 Qs): Q8/Q15/Q17/Q55. Miscalibrated autonomy threshold, errs both ways. Rubric: additive discovery → accommodate+note; material scope/resource change → surface options; divergent outputs unspecified → present both if cheap else ask; self-modifying params w/o goal-check → anti-pattern.
- Context economy (knowledge-judgment, ~2 Qs): Q46/Q9. Match output granularity to consumer's decision (summary not 500 rows; progressive summarization across sessions).
Exam guide update (2026-07-02)
Founder shared updated official exam guide (v0.2, June 30 2026) post-maintenance-window. Domain weights/structure unchanged vs. 05-27 baseline. New: full exam mechanics now confirmed (60 Q, 120 min, $125 fee, 12-month validity, 720/1000 pass bar confirmed official), 27 granular task statements, 6 named scenarios, 12 worked sample Qs w/ explanations, and an explicit Out-of-Scope list (drops streaming API/SSE, computer use, vision, fine-tuning, MCP hosting/infra, pricing/rate-limit calc, auth/billing, tokenization). Action: the "D3 API specifics" weak-area label above is mislabeled — tool_choice/Message Batches are D4 (Prompt Eng & Structured Output) under current numbering, not D3. Drop the queued streaming-event-sequence drill (now confirmed out-of-scope). Full diff → [[2026-07-02-cca-exam-guide-update-analysis]].
Strategic note
At 83% cold, founder could sit Cert 2 EARLY and bank the +$5k months ahead of the Nov-22 deadline — and it's the natural candidate for the "pass within first 3 months" escalator (by ~2026-08-26). Surfaced via iMessage 2026-06-14. ❓Confirm partner-portal gate is cleared (he has practice-exam access, so likely is — update the gate table when confirmed). 2026-06-27 caveat: exam 2 at 70% (borderline vs the 72% on record) tempers "sit it now" slightly — but the gap is ~3 cheap patterns (test-craft + hooks + API), not broad weakness, so a short targeted pass before sitting is the move. Re-confirm the real cut score either way.
Quiz log (appended by /cert-study quiz)
| Date | Cert | Scope | Result |
|---|---|---|---|
| 2026-06-10 21:32 ET | Anthropic CCA-F | D1 ×3 (stop_reason, subagent isolation, hook enforcement) + D4 ×1 (tool scoping/descriptions) + D3 ×1 (Batches API) | 4/5 — D1 perfect (3/3), D4 perfect, D3 missed (Q5 Batches API, founder: "no clue"). Taught inline. Next quiz: D3-heavy (tool_choice modes, streaming events, extended-thinking budgeting). |
Related
- Skill:
~/.claude/skills/cert-study/SKILL.md - Milestones + negotiated comp lift:
~/rdco-vault/01-projects/phdata/milestones.md - Memory:
~/.claude/projects/-Users-ray/memory/project_phdata_cert_escalator_path.md - Snowflake study hub:
~/rdco-vault/01-projects/certifications/snowpro-genai-c02/ - Anthropic study hub:
~/rdco-vault/01-projects/phdata/anthropic-certification-study-guide.md