The week the manual pipeline ran twice in one day, then taught the studio to enforce its own grammar.

AHA Pressure Test May 4 + May 7 · 2 sessions cross-cutting

AHA v04 Pressure Test shipped end-to-end Mon 05-04 morning against the new ShurIQ Report Grammar. Three-phase pipeline (Research → Synthesis → Builders) with three parallel Opus subagents in phase 1 (web spine 76 sources, InfraNodus graph aha-v04-pressure-test-2026-05-04 with 8 clusters / mod 0.49 / 7 structural gaps, design-system mining). An in-thread SYNTHESIS.md pinned editorial decisions for phase 2, then two parallel builders for editorial + viz hub. Live at aha-v04.pages.dev with five-viewport viz hub plus v2 alternates preserved on the viz-grammar-b branch.

• Fact-check audit caught four corrections — KHC reach order-of-magnitude error, AHA research $215M → $226M, Black women awareness drop −67% → −69%, Oura framing tightened to $900M Series E
• VP04 negative-space viewport rebuilt twice — v1 cleaned, v2 "What AHA Isn't Saying" lexicon preserved as a team-only alternate via the -b.html filename convention
• Pin/Clear UX redesigned with three explicit affordances: visible button + ESC keyboard + outside-click
• v06 regen Thu 05-07 to aha-pressure-v06.pages.dev: 28 of 30 R-IDs applied, 8 v0.5 violations rewritten in place, LLM-Primer sidecar shipped
• Viz hub regen at aha-real-viz-v06.pages.dev swapped the BPS Pentagon for 5 InfraNodus graph-playground viewports (8 viewports total)

TruData v04 Editorial Brief May 4 evening · 1 session · 8h

Full TruData v0.4 Editorial Brief — Sales pipeline ran against the manual generation playbook with five Opus 4.7 subagents (3 parallel research → editor's-chair synthesis → 2 parallel builders + 1 audit). Phase 1 checkpoint at trudata-v04-checkpoint.pages.dev went to team Slack review; Reframe pinned as Unnamed-Product Firm; cohort = TruData + 3 GSIs + 2 mid-tier + Globant.

• Reader-first transformation ran on a per-project reader-first git branch. Three commits stripped system-self-reference (betweenness, modularity, cluster IDs) from body copy while preserving load-bearing operations (SAS, named SAP products, anchor numbers). Methodology moved to appendix.
• A/B side-by-side at trudata-v04-b.pages.dev confirmed the editor's read; fast-forward merged into production at trudata-v04.pages.dev
• truPath visualization added to Gap 3 (3-loop / 5-phase / per-phase artifact diagram, illustrative-only framing)
• VP04 phantom-layout fix Chrome-verified: tangential-spread algorithm replaced angular spread for radial graph annotation pairs (angular spread collapsed at edge clusters); iteration loop ran Chrome headless screenshots → Read-tool inspection → adjust → re-render across desktop, mobile, and 2x DPI
• TSDR direct lookup corrected truPath status from "registered" to LIVE/PENDING, filed Feb 2026, intent-to-use basis — applied surgically across 16 edits

ShurIQ Report Studio May 6, 7 · 3 sessions cross-cutting

Grammar v0.5 refinement handoff authored Wed 05-06 — four-deliverable spec for a fresh Claude Code session to extract refinements from the 2026-05-05 review meeting using a four-pass extraction lens (Direct Address / Repeated Theme / Anti-Pattern / Structured Rule) with foreground-only agents per the 600s-watchdog learning. Embedded prompt block, source paths verified, dedupe procedure against 19 existing feedback files.

• Studio runtime v0.6 code complete Thu 05-07 — 27 of 30 R-IDs encoded in functions/lib/grammar.ts, both R-LINT linters (R-LINT.1 inversion-rhetoric + R-LINT.2 scaffolding-leakage) in linters.ts, migration 0005_v06_grammar_linter.sql, GRAMMAR_VERSION bumped in three places. Deploy blocked by sandbox shell-auth (agents don't inherit npm/wrangler auth); main session must run.
• v0.7 schema migration Thu 05-07: account → project → report hierarchy committed (193e5564). Editorial status site shipped as data-model-bootstrap (f6425f72); exec-summary visual-grammar versioning landed three independent fixes side by side (4970ccd1).
• shuriq-advisor-w19 weekly recap site built and deployed Thu 05-07 with 10 minutes to spare before the 5 PM advisor call. Iframe-tiled "screenshot" pattern (live iframe scaled 0.5x with pointer-events:none inside an aspect-ratio frame) substituted for actual screenshot capture.
• 20th memory feedback file authored: feedback_no_scaffolding_in_body.md canonicalized "leakage" as the term for prompt-leak text in body copy

ServiceTitan v05 Editorial May 8 · 1 session · pre-meeting

ServiceTitan v05 editorial shipped with AHA-style zingers injected at every anchor point. Driven by Nuri's three notes from the prior session (too technical, not enough hard digs, every acronym needs glossing on first use). The fix: a four-agent parallel synthesis (AHA-DNA-Strict, Plain-Boss, Koan-Brevity, InfraNodus negative-space) producing 180+ candidate lines, distilled into per-anchor winners, then surgically swapped into the canonical HTML.

• Zinger Lab companion site at servicetitan-zinger-lab.pages.dev shipped with 185 candidate cards using the atwist-CopyLab design DNA — the meeting-prep tool that let any line be swapped in 90 seconds before the 5 PM ServiceTitan meeting
• AHA zinger 5-move pattern distilled as a reusable template: contradiction-named (without inversion), evidence-anchored, scale-inversion, acknowledged-but-absent, outcome-over-metric. Persisted to mem0 as workflow-pattern.
• Negative-space pass on servicetitan-trades-millionaire-2026-04 surfaced the apprentice-hour ledger as the convergent finding: every model run (sonnet, opus, both transcend modes) independently identified the licensing board as the last analog chokepoint AI cannot route around
• Folded into Action 04 kicker rather than promoted to a new Action 06 (would have violated v0.3 locked-action-count); the 11 underlying candidates surfaced in the Zinger Lab's dedicated Negative-Space Atlas section

Memory + Skill Infrastructure May 4–8 · cross-session

• 15-doc manual-generation playbook authored Mon 05-04 at projects/shur/manual-generation-playbook/: README + 10 process pattern docs + session retrospective + 3 skill drafts (review-only). Encodes the three-phase pipeline, synthesis template, viz-grammar contract template, section-directive template, anti-slop checklist, fact-check pattern, bug reference library, reader/internal discipline, deploy/cleanup pattern, branching strategy.
• Two GitHub repos pushed Mon 05-04: aha-v04-pressure-test (deployed report, main + viz-grammar-b branches) and shuriq-report-playbook (process docs, single main branch)
• Seven durable learnings persisted to mem0 Wed 05-06: Opus 4.7 directive, Cloudflare Pages branch trap, CF project-name cross-account collision, InfraNodus 3-tool convergence as rigor signal, background agent 600s watchdog, empirical Opus vs Sonnet quality delta, seed graph_ref stubs caveat
• 20th memory feedback file completed the explicit-rule corpus — R-LINT.2 ("§-routing," citation-policy meta-prose, prompt-leak phrasing) now enforced at runtime

Totem Orchestration · Ecosystem Map 5 batches through the week

• 60-node AHA v04 subgraph (Mon morning) · 54-node TruData subgraph (Mon evening) · 9-relation v0.5-handoff batch (Wed) · 12-relation v0.6-lockdown batch (Thu) · 10-relation ServiceTitan-v05 batch (Fri)
• Cumulative graph state at week's end: 11 clusters, modularity 0.65, top gateways cloudflare_pages (BC 0.47, deg 181) → shuriq (BC 0.34, deg 130) → slack (BC 0.18, deg 66)
• Top three structural gaps now showing: Knowledge-Mapping ↔ Typography-Design · Typography-Design ↔ Diagnostic-Queries · Communication-Channels ↔ Typography-Design