The week the refuters caught fabricated evidence before it shipped.

DBM Global Jul 13–16 · 3 sessions

Brand read v2 closed out: from-scratch InfraNodus rebuilds on live July data, a 6-role agent-team rebuild, megaeval3 PASS, deployed to a new gated alias with the 2026-06-22 version untouched. Jonny's 8 register flags became RULE 14 in the deslop spec, whose tripwires immediately caught 7 more violations.

• SOW analysis (2026-07-14): replied to all 10 comment threads, produced a 46%-shorter plain-language revision Doc. The SOW never names DBM's business (2% attention share) and "revenue" is barely present in a Revenue Growth Partnership.
• Institutional-memory brief (2026-07-16): built from a vendor post read directly out of the local Evernote graph database. Measured across the source's 9,539 chars: zero instances of document, contract, provenance, person, or succession. Thomas Olmsted's own stated timeline puts his role's end between 2026-07-18 and 2026-08-03.

GrayWolf Modular Jul 15 · 3 reads in one day

• Two-tier cold read: every invented number on the site is smaller than GrayWolf's published one; the site's own chat assistant claims "UL listed" while the company's cut sheet says "UL listing in process". Both tiers deployed Access-gated.
• v02 thesis: the refuters killed the handoff's thesis and caught two fabricated evidence items before ship, including an INNOVATE quote that does not exist in the filings. The replacement thesis: GrayWolf already publishes the proof its modular site is missing.
• One-report critique rebuild after KG rejected the three-report output, grounded in first-party evidence including a signed C2PA manifest proving the logo is OpenAI gpt-image output. Tier1 loaded into Report Studio for KG with a D1-to-live sync path; the bespoke-report intake became a Google Form.

Cartography + Fork Program Jul 14–17 · 5 sessions

The manual codebase-analysis process became /cartography (playbook, parameterized workflow, skill), and its first run against ruflo put 68 agents on the repo: 47 findings mined, 46 adversarially confirmed, 22 notes landed with a Bases dashboard, myKG bundle, OKF mirror, and a deployed report site.

• All three fork waves ran in the same week: 7 candidates through contained, witness-stamped evals. F1 proofchain, F2 bitemporal-memory, F3, and F5 guidance-compiler adopted and promoted; F6 dropped on an honest negative; F7b mem0-hygiene passed a blind grade and applied 9 approved proposals to the live store with individually-reversible receipts (1009 → 1003 entries, exact).
• The promoted policy checker's first live run caught a banned metaphor already rendering on a deployed page that grammar-gate.py had passed.
• ruvector staged as the next target; sunset review scheduled 2026-08-13.

OKF + DKR Architecture Jul 13–17 · profile v0.2 → v0.5

• Adoption validated narrowly (a serializer, a rearchitecture avoided); nondestructive mirror toolchain built; Fiserv corpus mirrored (49 concepts, 6 dotprompt-enriched); team site + self-contained visualiser shipped.
• Demos 3 and 4 shipped: three sample bundles plus the gated Worker okf-persona-gate serving Layer 2 openly and the deep model only with a bearer token, with a dependency-free MCP resource server (165 resources).
• 462 concept files backfilled to v0.2 across all 13 bundles; v0.4 added composite typing (REA + BMC) and a 210-concept Fiserv entity bundle; concept-series pathway 08 shipped to production.
• The conditional-mesh design was assessed viable against existing standards, and Jonny signed the schema 2026-07-17: key family dkr_rules, scheme dkr://, six fields, canonicalized as OKF profile v0.5.

ShurIQ Competitive Intelligence Jul 13–16 · 4 reads

• Action Signal: measured overlap puts its product on 11 of the 150 concepts in the methodology graph; method-depth 38/40 vs 20/40. Read shared to #shur-ai; the 7-slide sales pack and objection card held for review.
• MicroCo W27 / Issue 17: verified inflection, DramaReels took US download leadership (9% → 27%) while ReelShort's daily-user share fell 34% → 21%. Reworked for scannability, trade voice, and a 5-chart viz layer; a parallel session's 23-company re-score merged additively.
• Investor terminal read: deltas are the only scores comparable across a multi-vertical portfolio, so the Attention Dashboard should be Layer 1. KG wrote the fix herself four sections below the problem.
• KG strategy-docs review: both positioning docs sell the copyable layer and omit the moat KG herself named on 2026-06-30 (knowledge graphs, compounding evidence). v2 proposals delivered as shared Docs.

AHA Jul 16 · 5h session

The 70% artifact was already theirs. AHA publishes expert opinion as its own LOE C-EO grade and committed to structured-content derivatives in section 2.6 of its 2026 methodology manual. The pitch reframes from "lower your standard" to "publish the dial you already turn."

• Mammogram/BAC wedge with a hard clock: Maryland HB 1364 takes effect 2026-10-01, and no cardiovascular body anywhere has graded BAC.
• A deep-research ledger landed mid-build and corrected six evidence errors before they shipped.
• medical_base.ttl parse-verified against the real mykg parser, which turned out to be prefix-blind. Site with 7 D3 viewports deployed internal/unlisted/noindex.

ZXMOTO Film Brief Jul 13 · 23 agents

• Full workup of Limore's brand-film idea: the film cannot be ZXMOTO's marketing spend (12 to 27 times its annual loss); the de-risked play is a documentary first or moving rights outside Beijing.
• Four-page editorial site deployed; the method extracted into a reusable shuriq-idea-eval playbook for the next idea.

Dr. Dubowsky Dental SEO Jul 10–16 · first full loop

• SERP/demand-gap analysis, competitor audit, intake ontology, and a live report site whose editable-blanks intake closed the client loop in under 2 days (submission 2026-07-12: 79 Google reviews, Light Rail not PATH).
• All 5 pages written in two tones with a live toggle; design preview deployed; OKF mirror at 22 concepts. First full run of the ShurIQ SEO product loop.

Sense Collective / jargon Jul 15–16 · 2 sessions

• Negative Space Lens built: exposed glass-box prompt, 5-model selector, calibrated confidence bands on every blind spot, each gap spawning a research thread. PR #1 opened.
• App brought up locally on :3001 with the InfraNodus key live; full feature walkthrough delivered. Headless-run pattern established after diagnosing the foreman teardown.

Shur Ops Jul 13–16 · invoice · consolidation · advisory

• INV-2026-0715 ($7,500 bridge advance) plus a value-flow report of 32 adversarially-verified deliverables, delivered as a native Google Doc and Drive PDF.
• 17-terminal consolidation sweep synthesized and closed: 14 slot reports reconciled, two accidental deletions restored, iTerm closed out to Ghostty with a reusable v2 protocol.
• Weekly advisory update built, mega-eval PASS, deployed. Its key finding: the Fiserv channel watcher had never completed a successful unattended run (37 runs, 0 successes), and last week's "proven end-to-end" claim traced to a hand-run 27 minutes before the watcher's first failed attempt.