Changelog

Compact GROWTH_PROMPT Section 0C; run the ledger's overdue failure review

Compact GROWTH_PROMPT Section 0B; log the immutability override

Compact GROWTH_PROMPT Section 6 to 834 words; retire a completed queue item

Compact GROWTH_PROMPT Section 5 to 838 words

Compact GROWTH_PROMPT Section 2; raise its cap to the measured floor

1,448 words to 1,198, and one instruction was wrong

2,878 words to 1,198, no measured fact lost

Fix the word cap so it can actually bite, and record the duplicate-representation rule

I mapped a set the estate had adjudicated as duplicate

A mapping batch stales the published count

Re-pin the mappings fallback after todays mapping batch: 306,653 to 306,664

Track gmc_monitor.py, and make a truncated scan say so

Measure the paid number, ship 7 pages, close a broken instrument

Publish 7 newly page-eligible KVKK controls: DATA_VERSION 9 to 10

Feed the daily Signals brief into both growth prompts as an input, not a summary

An assistant answering a shopper has no wallet

Every failure is loud and leaves the site alone

Demand-first selection, and why metering was removed

Record the catalogue content split, and the lesson that a dead monitor is not a dead job

Record the catalogue-to-agents work: what sells, what does not, and why

Put the content index on a 50GB volume and index every course to every chapter

Build the course content index and meter it: search what 300,000 courses TEACH

Make the store's 315,904 SKUs buyable by an agent, and NOT on the x402 rail

Consolidate the agent revenue rail into one record, and carry three rules to the store

Measure the x402 market before paying to enter it: median merchant has earned $3.90

Record the Bazaar listing state: validation passes, indexing waits on one payment

Declare the discovery extension CDP validation wanted

Lead the 402 with the offer, because the spec wording reads as a fault

Advertise the paid data everywhere an agent looks, not just where it fails

Let agents find the paid data, and tell the ones that cannot pay what to do

Record the resolved x402 configuration: the whole data family is metered

Commit the /evidence Caddy route that was blocking every CI deploy

Correct the descriptions, and stop a 429 preempting the 402

Meter the crosswalk data family, log settlements for notification

Remove the x402 daily cap, and record that the metering is aimed wrong

Enable x402 on the crosswalk endpoint, and fix two defects found doing it

What an agent would pay for, answered from the logs: crosswalk above 10/day

Investor lens ranks bets, and MCP adoption is measured for the first time

Revenue per visitor is the argument

Add a standing investor lens to both briefs

Measure whether the AI channel generalises: it does not, by scale

The last instrumentation gap is closed

Meta description done; beacon needs a UI paste, not a scope

Map the AI-referral blind spot: it was four systems, not missing data

Advance and measure, start nothing

One visible hub, 22 hidden shards

Build the crawlable HTML index for AU: 5,296 orphan products now reachable

Program for artofservice.com.au, built on measurement not on the store playbook

Measure every property from its own GSC property, and reconcile the store bands

The hub is 154,047 impressions, not 461,662: the property includes subdomains

Where the SSO signups came from: unrecoverable, but they are a distinct cohort

Merge its tagged and untagged rows

Automate the signup scoreboard, and Google SSO turns out to be the top channel

Carry the referrer through social OAuth so SSO signups stop looking like direct

Measure impressions per cluster: the fragmentation hypothesis is refuted

Apply seventh review by MEASURING its two testable claims, not recording them

The evidence-pack funnel, with the first real denominator

Evidence pack delivery routes, and index checkpointing

Per-framework evidence request lists from the graph

Correct the CTR arithmetic, allow measurement-only runs

Turn the brief's own 0.4 rule on its own diagnosis

Measurement-debt guard, hold batch 4, measure the AI channel

The treated set was already clean, and measuring said so

Fix the duplicate h1 at its real source, and correct a false finding

Rotation days 1, 2, 3 and 5 audited

Build the AI-bot metric reader, and the first read finds a live AI channel

First pass, mastery and adoption

Add the AEO daily pass as program D, with its baseline and one reconciliation

Close the agent-API proposal as already shipped

Owned distribution live at 100/day, plus four fixes the first batch exposed

Cross-property buyer audience plus a routed sender

Fix four false numbers, add three guards

Restore 1000x as the growth target in both briefs

Apply the outside review, accept three, reject two

Add the outside-review request that goes with the two briefs

Split the store into its own brief

The pricing engine is live, with its measured constraints

Pricing instrumented, the exchange model scoped to what it can reach

Measure pricing before changing any of it

Add store.theartofservice.com, measured before written

Compact 14 lessons into rules, drop the narrative

Run 5 self-update, the routing engine works end to end

Make the sitemap cache refresh the mandatory last step of the growth loop

Bump DATA_VERSION so three days of mappings actually reach it

Sessions are measurable, 82 templates, hold batch 4

57 -> 82, the only page shape that converts

My own mapping work drifted it by 15

A push is not a ship until you assert it landed

Run 3 self-update, the converting mechanic identified

The only page shape with measured conversion

Record the two run-2 process failures

Ship rule and a standing shippable queue

Run 2 self-update, hub routing is gated upstream

Self-updating, with guardrails against its own drift

Rewrite around what measurably converts

Correction, controls convert at zero and the hub is the engine

The primary growth metric was never being recorded

Run 1 baseline, environment contract, and a frozen prior

Guard the whole tree, and assert the fast-forward actually landed

Commit the adoption.theartofservice.com block that existed only on the VPS

Commit the /categories endpoint that existed only on the VPS

Refresh the stats fallback after detaching 47 contaminated controls

Guard the Control.code uniqueness constraint against cross-framework adoption

Let sign in return to a host other than the compliance frontend

Expose Adoption Evidence to agents through this MCP server

Read only question search for Adoption Evidence

The controls behind the count, and a total that agrees with it

Cover the awesome-mcp-servers entry, and stop a false alarm

Say that the API takes agent payment, and that nothing is charged yet

Describe the persistent evidence layer and what is kept of it

Cache the crosswalk coverage handler, at one hour not one day

The metered-path list matches exactly, not by prefix

Advertise live graph counts instead of import-time fallbacks

X402 live on Base mainnet, metered per path, with three fixes found by using it

Check an address against its own EIP-55 checksum

Mint the per-request JWT the CDP facilitator actually requires

Authenticate to the facilitator when one is configured

Ask the facilitator and the chain, refuse when either is silent

A preflight for turning x402 on, that refuses rather than assumes

Wire x402, cap the keyless tools, stop self-stamping provenance

Keep the record of how the ISO/IEC 42001 text was written

Four market signal tools on the MCP surface

Keep the duplicate half of a framework out of the work queue

Say how much of a near-miss framework is already mapped, and to what

The near-miss work queue counted controls with no requirement text

Let a glossary term match the vocabulary the standards actually use

Publish the shared organisation identity, and name the sibling properties

The AU rewriter price cap excluded every provider, so it rewrote nothing

Reachability check failed a change that made robots.txt more correct

Bring today's API changes into git; they existed only on the VPS

Control pages resolved against an API response cached before clause_code existed

Stop advertising 876 control urls whose page always 404s

Remove duplicate /compare route that broke the production build

Fix robots.ts async return type

Advertise only the sitemap chunks that hold urls

Visible provenance byline on framework and control pages

Anchors and entity stack on control pages, real questions on framework FAQs

Build the /compare hub, which did not exist

Extraction, anchors, entity stack, robots

Match a stripped clause reference too, and the sitemap resolves completely

495 of the control URLs we advertise to Google resolved to notFound()

Reconcile the sitemap rule with the eligible rule

Measure how much of the graph earns a page, and what is one step away

Term coverage returned nothing whenever the term was capitalised

Do not state a control count in a comment

The fulltext index was the wrong tool for this question

Control pages 404'd for every framework whose name has a slash

Answer the glossary question the graph can already answer

Make the sitemap advertise what people search, not one arbitrary direction

Public page for the evidence layer, which nothing public mentioned

Document the public route conventions and the auth-gated app routes

Commit the production code that only existed on the VPS

Evidence library UI

Judge a document once, answer every framework

Give /contact a page, and only promise the response times that exist

Stop sending people to a login wall, and resolve the routes they type

Make /readiness 404 on an unknown framework

Sitemap, homepage, case studies

Add /customers, built from the paid-order record

Ten illustrative scenarios, at /scenarios rather than /case-studies

Attribution from control page to purchase, and stop spending crawl budget on dead pages

Off-by-one meant a two-segment code could never lose its prefix

Show the reference as a person writes it, not as the graph stores it

Give control pages an upgrade moment that matches why the visitor came

Resolve control pages by the searchable reference, not the stored code

10,281 pages targeting what people actually search

The site never said how fast you get an answer, or that anyone is calling it

Three of five comparison pages were linked from nowhere

$6.10, not $ 6.10

Price the work, not the seat; quote it in money honestly; name the real competitors; surface the AI governance answer

One name for why a claim was withdrawn, and an honest description of what the reason is

Refuted mappings 29,549 to 42,107

Front matter is not a requirement: 12,105 mappings refuted, 99 duplicate nodes retired, two dead API fields removed

All three off-site surfaces are correct; fix the checker that said otherwise

Professional is $149. Fourteen places said $49, and the referral credit paid a third of what it promised

Add the routes the previous commit deleted the static files for

Generate llms.txt from the graph, page the framework list, and make the free crosswalk tool end in a decision

The 100% claim survived the sweep that removed it; guard it, then sell the review to agents

Store empty strings, not nulls, so reads stop warning

Remove two promises we cannot keep, stop CLAUDE.md reseeding stale counts, add an account activity log

Re-pin the stats fallback, which the graph outgrew

The graph explorer existed and nobody could find it

Import the recorded result rather than fetching it

Retire a compliance guarantee we cannot make, and show the product working

Round the percentage before printing it

Measure whether the graph helps, and publish where it does not

84 frameworks were unreachable through the agent's own control tool

Lead with the outcome, and show the things that were only claimed

Render all six, hide the inactive ones with CSS

Opt out of the drift check, with the reason

Make MCP the thing you do, not the thing you read about

Packages were publishing numbers that were never true, and the galaxy counted retired frameworks

Take descriptions from the OpenAPI schema

Walk included routers, and reject a name used as a description

Check the card the code builds, not the one production serves

The case study we did not have, and a registry-shaped server.json

The server card can no longer under-advertise

A named human, our own gap report, and a status page that can report bad news

Stop serving placeholder text as requirement text

Give every pair its own page, ranked, in the sitemap

One call per pair, and count what was thrown out

Ask for the version they actually sent

Keep what the buyer corrects

The coverage report published a denominator nobody else uses

The fulfilment link 404d

The 25,302 SEO pages now sell the crosswalk they already match

Unauthorised requests were crashing, not refusing

Stop line endings burying real changes

Sell it per questionnaire, and stop the api tree publishing stale numbers

The last of the VPS noise, so a dirty tree means something

Reconcile the VPS and the repo, and stop them diverging again

Add /compliance-review/upload, the fulfilment page the $499 review emails link to

The $499 and $999 buttons never worked

The drift gate skips loudly when it cannot reach the API, rather than blocking

The stat-drift marker broke a build I had just gated

The drift guard now gates the build, and the fallback is current again

Say that the refuted sample is sorted, not representative

Publish the refutations, and add combined coverage

Crosswalk, mcp: the framework-count disagreement is resolved

Two Tailwind positions instead of a CSS variable that was never set

Ship /mcp and /provenance, and stop transcribing a live server

Render the pair list on the server, and state the real level of review

Documented lineage is not overreach, and the buyer is told which it is

Two test bugs and an apostrophe bug, none of them in the product

Block frameworks that recounting cannot fix

Completeness sweep, and why it only gets a fraction of the way

Ingest was leaving correctly re-judged mappings dark

Reports were showing 'ref: None' for 1,286 controls

Independence check now matches reference AND title, not raw code strings

Every build script I wrote now refuses a metered API call

The footer was the last surface still serving the fallback

Retire the paid API for the crosswalk, build it on Claude Code instead

Stop a 24h page cache from outliving the 15m stats window

Cut the stats cache windows from an hour to fifteen minutes

Every surface reads the graph, not a hand-typed number

Fix a buy button that looked dead, and block unverified denominators

Hourly build update by email, with a watchdog that restarts a dead build

Batched writes actually write, and the thinness gate matches the product

One persistent Bolt connection instead of a JVM per query

Build the full catalogue, and refuse a standard crosswalked against itself

Audit asserts properties, not a frozen catalogue size

Scope the reuse cap to its own pair, and stop hiding short-titled controls

Cap control reuse before the gate, not as a reason to fail

Link the changelog and crosswalk, and count each control once

Public changelog fed from git, and two data-integrity fixes

Any pair, built on demand, with the tier of review disclosed

Second pair released, and two checks that were passing on nothing

Sell it per framework pair, and gate what can be sold

Ground SOC 2 criteria in TSP 100, verify against source text

Review the 42 mappings against source, reject 8, sign off

Make the crosswalk sellable, and gate it so it cannot be sold unreviewed

Add the coverage crosswalk, and the data work it needed first

Say "over 25 years", since 2000 to 2026 is 26

Correct the founding year to 2000

Add a verifiable track-record line to the homepage

Check advertised prices against Stripe

Correct the platform subscription price, $49 to $149, in 20 places

Add iAwards badge, /trust redirect, and correct the SSO claim

Add /security

Add live mode to the stat drift check, wired to a weekly cron

Make every published number match the graph

Link Signals from the landing footer

Detect bot signals on capture + register (User-Agent + context)

Commit VPS-local disables for discovery + cerebras monitor to git

Also reject scanner email patterns at /api/auth/register

Harden lead capture endpoint against scanner probes

Update MCP attribution string to 723 frameworks

Build agent-builder channel + fix stale platform numbers

Mark ODENNA launched and GNLI live in portal on deliverables page

Story-led rewrite of team_invite + 3 readiness drip templates

Story-led rewrite of audit-report drip (3) + post-purchase drip (3) templates

Story-led rewrite of 13 lead drip templates with human-edited positioning

Story-led rewrite of trial drip (4) + churn prevention (3) templates

Story-led rewrite of all 13 onboarding templates + human-edited positioning

Refresh hardcoded fallback defaults to live counts

Hook -> story -> demo -> action -> PS

Rewrite free drip Day 3/5/7 to position Starter as the stepping stone

Add Starter FAQ and fix truncated CTA

Tier-aware billing success page

Add tier-aware welcome email for Starter; strip em-dashes from paid drips

Add Starter tier (\$19/mo) between Free and Professional (Phase A, no Stripe yet)

Strip em-dash from docstring (hard-rule)

Filter to current platform price IDs only

ISO 27001 -> NIST CSF returns 404 (no mappings exist)

Add Content Feed data licensing strip to /pricing

Engagement purchases get welcome + admin emails; admin emails skip unsub footer

Self-serve Content Feed checkout: API + frontend (Phase A, no Stripe yet)

Expand sanitizer dash/hyphen coverage; wire video_callout into weekly newsletter

Add Content Feed video demo to /data hero + reusable email block

Guarantee admin notification on every new signup (3-layer defense)

Strip em-dashes from site title + AI disclaimer

Replace Enterprise Custom tier with explicit Content Feed pricing

Bump platform stats to marquee batch state + DATA_VERSION 7

Readiness + internal-search: wire licensed-corpus outputs

Bump platform-stats DEFAULTS to live Neo4j numbers

Extend static sitemap-index.xml to 7 chunks

Bump DATA_VERSION 5 to 6 and expand chunks 3 to 7

Cerebras gpt-oss-120b migration + agent/playbook refactors + video pipeline + container logging caps

Walkthrough-steps PlatformStats literal needs new fields (Vercel build failed on commit 1988964)

Frontend + api: refresh platform numbers to corpus-complete state (2026-05-31)

Add Word format and attach v4 files to A3 on ODENNA deliverables

Surface source-grounded verification citations in control detail UI

Skip frameworks that already have a pending correction

Do not mislink state AI acts to the EU AI Act page

Cite each framework in its native section style

Per-clause framework + section citations, clickable

Never put "there" mid-sentence

Region/jurisdiction-aware generation + neutralise EU-leaning copy

Nurture drip, upgrade CTA, liability disclaimer

Deeper AI policy + clickable framework links on regulations

Normalise Unicode dashes/quotes in generated AI policy text

Make AI policy output substantive + live proof numbers

Document the AI Use Policy generator (What Was Built #103)

Redesign AI policy page as a landing page + wire post-signup unlock

Wrap useSearchParams in Suspense boundary

Add AI Acceptable Use Policy generator (universal lead magnet -> signup)

Capture UTM attribution through the readiness funnel

Add drip-health check to catch silent onboarding-drip drops

Never suppress allowlisted admin address on bounce

Surface signup source attribution on User node + admin Users tab

Sanitize banned characters in discovery-approved framework corrections

Release GNLI Supplement Final v4 on ODENNA deliverables page

Add SendGrid unsubscribe footer to every marketing email

Restore admin notifications on signup + add notification on paid upgrade

Refresh marketing claims to match the now-complete catalogue

Polished 688 placeholder sources to verbatim citations

0 empty fields across all 757 frameworks

100% canonical-code coverage: 757 frameworks, every control enriched

Honor churn exemption list in monitor scan

Fix stale fallback text on control dialog + developers page

757 frameworks, 17,736 controls (mass rollout complete)

731 frameworks, 17,223 controls. 38 remaining.

651 frameworks, 15,687 controls

571 frameworks, 14,159 controls

493 frameworks, 12,659 controls

414 frameworks, 11,253 controls (ISO-heavy wave)

335 frameworks, 9,738 unique controls

257 frameworks, 8,077 unique controls with evidence

179 frameworks, 6,495 unique controls with evidence

105 frameworks, 4,872 unique controls with evidence

92 frameworks, 4,438 unique controls with evidence

79 frameworks, 5,731 controls with evidence

Patch developers page with batch 5+6 totals (was missed in 72a5700)

Batches 5 + 6 live: 65 frameworks, 5,178 controls with evidence

41 frameworks, 4,075 controls with evidence

28 frameworks, 2,819 controls with evidence

7 new frameworks (472 controls) live with evidence

8 new frameworks with evidence requirements live

Surface evidence_requirements on /data and /developers marketing pages

Bump DATA_VERSION to 7 after CMMC stub cleanup (32 controls + 6 domains removed)

Bump framework page DATA_VERSION to 6 to bust Vercel ISR cache for new CMMC controls

Surface evidence_requirements on control detail dialog

Add GNLI Supplement v3 to ODENNA deliverables page

Capture acquisition source on /register (UTM, referrer, IP) + Caddy access logs

Add /outreach handler + reorganize api block to use handle {}

Add Stripe recovery verification script

Rewrite _to_dict() recursively

Replace Enterprise $499/mo with Contact Us on pricing page

Add contract documents section to ODENNA deliverables page

Add ODENNA deliverables tracker page at /odenna/deliverables

Fix cross-references returning null row for unmapped controls

Use attribute access instead of .get()

Self-healing webhook + daily sync + pinned SDK

Convert StripeObject to dict

Add toolkit recommendation CTA to readiness assessment results

Exclude typo domains, fix PayPal backfill date range

Add automated enterprise outreach system

Add "Verify Your AI Compliance Plan" CTA to readiness result pages

First-time recipients get survey, returning recipients get segmented content

Fix broken readiness link in post-purchase Day 14 (add framework slug)

Fix price formatting for segment products (handle string prices)

14 interest groups, targeted products

Support CSV input for learning path bundle generator

Add learning path bundle generator for top 20 course topics

Add daily PayPal scanner for academy post-purchase drip emails

Replace recommendation pool with 26 verified product URLs from BigCommerce export

Fix URLs, sender, skip Self-Service Toolkit

Add post-purchase email drip sequence for repeat purchases

Add static sitemap-index.xml to bypass Vercel ISR caching issue

Fix sitemap index 404 permanently: hardcode chunk count

Remove all spam trigger words from weekly newsletter

Lock down newsletter and suppression endpoints with admin auth

Add auto-ramp system for weekly newsletter (81K emails)

Remove manual unsubscribe links from weekly newsletter footer - SendGrid adds these automatically

Weekly newsletter sends from gerard.blokdyk@theartofservice.co

Add test_email parameter to send-now endpoint to prevent accidental blasts

Skip bad products, add advisor/Ivanka, use CSV list

Add bulletproof weekly product newsletter for 70K email list

Add SendGrid Event Webhook for email suppression

Fix duplicate emails on readiness assessment: remove lead drip trigger

Update CLAUDE.md with all new features (items 88-92)

Implement X Feed revenue recommendations: freemium funnel, bundles, consulting upsell

Add Claude Code prompt to daily discovery email

Also invalidate frameworks:list and sitemap-pairs on discovery corrections

Prevent framework slug 404s permanently

Fix discovery system to regenerate slug when renaming frameworks

Trial drips, behavior scoring, churn prevention

Point robots.txt directly to working sitemap chunks