The memo queue

the file as received · 9dd3064503e2… · 2 KB

Produced. The stories section: why stories, the seven-step workflow with the picture last, the cast of nine as drawn with what each stands for on this site's model and one proposed, the rules for a story, and what is not done · Six stories as data, each with its page: the truth under it, the storyboard panel by panel, the prompt for any image model with a copy button, and one-panel ideas for a shirt. Two drawn by a ChatGPT image model, with what to correct written beside them; four storyboards from the articles: the calendar edit, the pilot, silence moving up, and always allow · A generator with a check in CI, so a story is a JSON file and the pages follow; the cast and the stories can move to stories.sgit.ai without being rewritten

Not done.

• Any model but ChatGPT's. The prompts are written for any; none has been tried on another.
• The four storyboards are not drawn.
• The one who signs, missing from the cast; proposed, not drawn.
• Voice: nothing is read aloud.

the file as received · f3849c17c9fb… · 1 KB

Produced. The article, in the order asked: seven user stories; the six weeks as a story; how to read the six bands, the two colours and a click; each connection alone (CRM, Mail, Calendar), then combined, where the trifecta appears; eight controls one at a time and then all of them, with a table of what each ends and leaves; the business analysis and five questions for a room; the technical analysis down to the one-way walk. Thirty-four captures, each taken by script in the state its caption names

Not done.

• Captures of a real agent. Every picture is of the invented example; running the figure on a published ABP vault is the next step the article names.
• A capture per role for all thirteen of the map's roles: the article shows the ones the argument needs (the rep, Sales, Legal, the CIO, the CEO, the board), not each one.

the file as received · 33d4cba3b317… · 1 KB

Produced. The case, computed like the others from the repository's own pages: what it does in its words, four answers it changes (what it could reach, what it did, whether and how fast it can be stopped), the register before and after, what it adds, what adopting it takes, one place its pages disagree, what the case does not claim · Two sections no other case has: the ABP mapping, object by object, of what a Topgent run fills (the grant, measured) and what it leaves to the ABP (the mandate, the delta, the barrier); and the tool explained from the buyer's desk, with a figure of what a run produces and who reads each part · Seven commercialisation notes from this site and open-source.sgit.ai: sell the maintained build; packaging not open core; the buyer is the person who signs and the product for them is the ABP; a vault per machine in sgit.ai as the custom UI and data storage; the mandate on the screen; OWASP and the summits; what not to do

Not done.

• A Topgent-fed ABP. The case says a run is most of the first pass; no run has been made, because that would mean running it, and the first one should be on a machine of ours with the author watching.
• The lead's own reply to the author. The page is the material; the message is the lead's.
• The two screenshots. They are a private conversation and are not archived; the case uses the repository only.

the file as received · ecff4157737b… · 0 KB

Produced. The case, computed like the others from Cloudgeni's documentation and Opengeni's README: six things they do in their words, five answers changed (changes need a person, what it did is replayable, it can be stopped in one action in minutes, its reach is stated at dispatch), the register before and after, what it adds, what adopting it takes, two places the pages disagree, what the case does not claim · The harness mapped against the published Gmail ABP row by row: which of its four unbounded rows a run inside Opengeni would bound, and how, with the projection that unbounded excess goes from four to nought and two residual risks appear; the Calendar before-image as the pilot's first design decision · The run from the buyer's desk, with a figure; and six notes on the pilot and the case as a product, including a vault per run as the evidence pack

Not done.

• The measurement. Every ABP entry is a projection from the pages; the pilot's first deliverable is the Gmail policy measured inside the harness, and that replaces the projection with its date.
• The Calendar policies themselves, still in draft.
• The message to Cloudgeni, which is the lead's.

the file as received · b0a0b9bcf447… · 0 KB

Produced. Every case page carries the figure: the model's ten roles up to the board, every risk that holds for the stated deployment placed with the roles the model assigns it, the corporate register above them, and each of the product's changes as a switch. The page carries the model and the engine, so a switch recomputes the register in the browser exactly as the build did; checked on three cases against the tables. Red holds, faded retired, amber brought; a role's circles count what it holds and what it no longer holds; click-to-focus with the one-way walk

Not done.

• The acceptance loop. The article's figure has holders, intervals and silence; a case's figure has the register only, because the model has no acceptance records. Adding them is the Risk Acceptance Office's work, not a case's.
• A capture guide for the cases, like the visualiser's article. The figure is the same grammar; the guide covers it.

the file as received · 396705a31466… · 1 KB

Produced. The brief, on the console with a markdown twin to hand to the image model: what RiskMandate is in one paragraph, who buys the shirt and why, the brand kit and where the files are, the print rules, forty-odd taglines in six families with the joke explained, twelve design concepts with and without a figure, what a design must never carry, the prompt to paste, and the output the lead wants back

Not done.

• The designs. The image model makes them; the lead chooses.
• Amazon's current artwork specification. The brief states the usual one and says to confirm it on the account before uploading.
• A trade-mark check on the taglines before they go on sale, which the brief asks for.

the file as received · dadd026e61c5… · 2 KB

Produced. How it works, rebuilt in the order the memo gives: six steps with the status of each on the step

Not done.

• Connectors to a customer's own control planes (identity provider, cloud IAM, proxy, CI). Described as built per engagement and productised as they mature; none exists as a product and the page says so.
• The graph's edges upward to risks and the board, and sideways to a customer's internal policies and documents. Standards edges run today; these are marked in design.
• Execution logs and evidence held in the vault beside the policy, as the memo's provenance store. Marked as where this goes, not as something that exists.

the file as received · 0c528b8c5736… · 3 KB

Produced. UK support, in the open: 79 programmes, events, schemes and networks, each read on its official page and dated, with our own status on every row · The same register as data, for other founders to copy

Not done.

• Applying to anything. Every row says not started except Web Summit; which doors to try, starting with the Sovereign AI procurement challenge, is the lead's decision.
• A matching page on sgit.ai's partnerships section. That site is not in this repository; it can link here.
• Pages that refused an automated read (British Business Bank, london.gov.uk, blackhat.com, rsaconference.com, the Turing, the ABI) are marked unclear rather than filled from search results.

the file as received · 762fd1ea85c2… · 7 KB

Produced. The pilot worked. Then somebody asked what else it could do: fourteen surveys and forecasts with the causes each names, seven cases, five variables the lethal trifecta leaves out, and what the data does not show

Not done.

• Data that tests the hypothesis. No survey found asks whether the gap between reach and mandate stopped sign-off; the article says so and asks readers for cases.
• The credit approval at two in the morning has no public source. It is told as an anecdote the author was told, and nothing more.
• The operating limit, headroom and hard maximum the memo recalls from an insurance vault were not found in the published demo vaults; the article presents the shape with invented numbers, labelled as such.

the file as received · 0349fc94e7df… · 5 KB

Produced. A deleted meeting comes back. An edited one does not: what Google keeps after each action, in its own words, for personal and Workspace accounts; three contradictions in Google's pages; and the draft Calendar rows of a behaviour policy

Not done.

• The Calendar behaviour policies themselves. They are in draft; the article quotes their rows and says so.
• A restore for edited events. Google documents none for users; Workspace with Vault keeps earlier versions an admin can export, and the article says the memo's claim holds for personal accounts and needs that qualification for Workspace.
• The early-user conversation (calendar over email) is reported as a handful of conversations, not a survey.

the file as received · 1dabf7f55b1c… · 6 KB

Produced. Business cases, by the risk they change: the method, the rules, and twelve categories computed against the typical deployment · The first case, our own: an Agent Behaviour Policy retires two risks, names one hidden one, and lists three expectations as reductions, never retirements · The engine and a copy of the RiskGraph Explorer model with its provenance, so every register is computed rather than written

Not done.

• The two named cases (agentgateway; Auth0 for AI Agents) are built as drafts, unlisted and marked noindex, until the lead decides to send them to the vendors.
• A semantic graph per product beyond the model's sixteen questions. A case is written as the answers a product changes; products that act on things the questions do not ask about cannot yet be expressed.
• Outreach to the vendors. It is the lead's, after review.

the file as received · 3686c72fb430… · 3 KB

Produced. OWASP, as a graph: the foundation, four families, 52 projects and documents, 110 items of eleven lists by title, and the relationships OWASP states, joined to the risk model through the Agentic Top 10 · Eighteen open-source business cases, OWASP first, each change backed by the project's own words and each with what adopting it takes · Eighteen companies built on or beside an open-source project, with only the quotes that were checked word for word

Not done.

• Business cases for OWASP documents such as the Top 10s, ASVS and SAMM. They change what a team knows, not what an agent can reach, so they sit in the graph rather than as cases; the bridge shows which cases change the answers each Agentic Top 10 item depends on.
• Three Agentic Top 10 items touch nothing in the model: supply chain, memory and context, and agents talking to agents. The model has to grow for those.
• Outreach to OWASP projects, maintainers and the companies. The lead's, after review.

the file as received · db4b98aef049… · 2 KB

Produced. The brief: three vaults (the existing app vault as the UI, a private keys vault, one vault per person), the controls against leaking across, the workflow, and what the feedback loop is meant to learn · The pack for the new agent: instructions, templates, a scaffolder, a leak gate proved against six planted leaks, the site's own builder and every catalogue deployment, and one fictional example built and checked; kept in the repository at packs/dist/, not on the site

Not done.

• The keys vault itself. The first session creates it and gives its key to the lead, so the steps are exercised once with the lead watching.
• A way for a person to send corrections from inside their vault. Today they reply to the lead; a form is a question for the app vault.
• Any person's vault. The pack has been run end to end on a fictional organisation only.

the file as received · c4734574a343… · 25 KB · sgit.ai v0.6.7

Produced. The first interview page, with the six parts in the brief's order, a copy button that copies the prompt exactly, and nothing loaded or sent · The pattern as a template: site/interviews/_template.json and scripts/site/build-interview-pages.mjs, so the next page is a JSON file

Not done.

• Running the prompt once in ChatGPT voice mode. This agent has no ChatGPT account; the lead runs it and checks the summary has all ten sections.
• The prompt is the brief's, with two phrases corrected to what the site states: the home page names the CEO, CTO and CISO, not an insurer; and the Licence to Operate page makes the behaviour policy the instrument, not the evidence.
• Sending the summary back into a vault through a write-only lane, which the brief marks as later.

the file as received · 2219d9a05179… · 1 KB

Produced. The founder interview's first part: six questions on the Agent Behaviour Policy (its name, explaining it back after one hearing, what it adds, whether the market understands the problem, its value to the people who would use it, whether it should sell); thirty minutes, sixteen summary sections · The article: the infographic's thirteen rows quoted and credited; what each company sets and what stays fixed; the middle column as a routing table and the organisation chart as the escalation path; one invented agent from eight ABP rows to the board; seven rules; six weeks; the model as nodes, edges with inverses and computations

Not done.

• The LinkedIn post's own address and date. The article credits the infographic by title and author and says when it reached us; the link is the lead's to add.
• Running the longer interview prompt once in ChatGPT voice mode, and checking the summary has all sixteen sections.
• The seven roles the article needs and the model lacks (COO, CIO / Data, Legal, HR, Marketing, Sales, all employees), authority as data, and an engine that runs the clocks. The article says so in its section 10.

the file as received · 524b59f30d44… · 2 KB

Produced. Three figures drawn by the page's own script from the article's data: the blast radius, which plays the six weeks or takes the reader's own changes (connect, disconnect, boundaries, instructions) and recomputes the rows, the risks, their holders and the counts; the six weeks as a strip; and the ontology, every verb readable from both ends · The instruction switch: the mandate becomes precise about the business process, every row in the gap gains an expectation, and unbounded excess does not move. The mail scope fact is cited from the Gmail record: three consent lines, each for the whole account

Not done.

• Blast radius as a number. The figure shows which roles a risk reaches and who holds it; it does not size the consequence, because the article scores nothing.
• Real grants. The figure runs on the article's invented agent; wiring it to a published ABP vault, so a real grant drives the same picture, is the next step.
• The flows of evidence downward. A ceased risk cites its ABP version in text; the figure does not yet animate the evidence travelling back to the version.

the file as received · 19de617c22e7… · 5 KB

Produced. The blast radius rebuilt: every live risk lights the path from its holder to the board, every role above carries a count, a role's panel lists what it holds, carries and is informed of, and a table lists every risk that holds at that moment with the chain it reaches, its consequence in words and whether it can be undone · The accepted level: row 1 is the rep's own mail and establishes R0, inside the rep's authority and accepted on connecting; a new row 9, the shared sales inbox, is outside the mandate and establishes R7. The article's tables, counts and six weeks follow: nine rows, five in the gap, unbounded excess five, then four, then three · Marketing, Product and HR leave the figure; Legal and the CFO are informed of every risk that touches customer data or a customer. A compound risk is drawn larger by the number of facts it needs

Not done.

• A quantity of impact or potential loss on a risk. The figure gives each risk a consequence in words and an undo property, and no number: the article's own rule is that kinds of consequence do the work of a score, and a loss figure is the lead's decision.
• Other scenarios. The figure runs one agent; the lead said to keep it so for now.

the file as received · d47dbfda7dd3… · 2 KB

Produced. The colour travels up: while anything below a role is unaccepted, that role's path, halo and count are red. Four more controls, one of them an execution proxy that carries out every write under the ABP's rules and bounds every write row at once; two fact controls, processing terms with the provider and the insurer's written answer, which end Legal's and the CFO's risks. Every control has a residual risk, green, inside its holder's authority, accepted when it appears; with every control on, every path to the board is green and nine green risks remain. A button does it in one click · Two more held risks in the article: RL with Legal (customer mail goes to the model provider, and the notice does not say so) and RC with the CFO (nothing in the cover says whether an act of the agent as staff is covered). The tables, the six weeks and the strip follow: nine risks on day 0, nine open and none unaccepted on day 42

Not done.

• The effectiveness of a control. The figure takes every control as working, as the lead said to for now; a control that is on and not working is a fact the ABP's evidence tiers exist to record, and it is not modelled here.
• Residual risks are drawn from a fixed list. In a real record they would be written by each holder in their own words when the control lands.

the file as received · d376355f1747… · 2 KB

Produced. The reviewed level — the workflow, the expectations and the figures with their sources · Who runs your review — the list, and the rule about what gets published · The first reviewer's page, every line read off a published page · The shape of a reviewer page, labelled a placeholder · The manifest the store can build its chooser from

Not done.

• A second, named reviewer: nothing about them is published until they have read their own page and agreed to it. The placeholder stands in the meantime.
• The chooser at checkout, which is the store's workflow rather than ours. The manifest is published for it; the arrangement is written up for the store team and nothing is agreed yet.
• Comments and testimonials. Asked for; none exist, and none will be composed here.
• Availability, a queue or a throughput number: none can be honoured yet, so the pages carry a status and the date it was confirmed instead.
• What the reviewer is paid — agreed in the brief and deliberately not published. A commercial term, and the lead's call.

voice memo, transcribed in the thread, with a screenshot of the Agentics Foundation London group, 21 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: the free rung below the ladder, why a small policy is not a small vault, the six things the memo names beyond permissions mapped against the model, the build order and five decisions · Try it - the seventh top-level section: the four steps, what you end up with, what it honestly is not, the four layers, and where to say it broke

Not done.

• A prompt workflow per shape we already have (T13) - the Gmail one is published at abp.sgit.ai and the other fifteen have only MAP-A-GRANT.md, inside the vault
• A feedback path from the workflow's last step that survives a stranger
• What we count, and where it is published - the site collects nothing, so the honest measure needs the lead
• Volume and instances on the consequence layer (T14): the relabelling, the mass send, the draft flood, the mailbox's state as an asset
• The session-scoped mandate, and the instance-scoped mandate - proposed as a Lab ask rather than a change to the published vocabulary
• Waiting on the lead: the seventh menu slot, what is counted, whether a user-made policy gets published, and how far the free rung goes

voice memo, transcribed in the thread, 16 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: the four storeys (permitted, exposed, practised, consequences), latent capability as the gap between the first two, the barrier gaining a holder and whether it moves without you, the audience spine, the build order and four decisions

Not done.

• Storey one for this shape (data/permitted.json, the scopes and the methods they unlock)
• held_by and moves_without_you on every barrier; splitting not_reachable into bounded and not exposed
• The audience spine in the reading app, with the executive view authored
• The storeys drawn as one diagram
• Waiting on the lead: whether grant keeps naming storey two, and the audience names

chat instruction with eight screenshots, 16 Sept — no digest; given as speech, chat text or a link

Produced. The template vault claude-gmail-connector: six rows, four measured on the deployer's own account, the evidence transcribed with the address redacted; pushed as vault oc433z3m and read live on its own page · The customer's draft instance, anonymised, with a question on every mandate line (vaults-instances/) · The purchase workflow, run once: the brief · The tile in the library, in Mail & files connectors, measured 4 of 6 · The customer's draft instance pushed as a private vault (xjir6m0c); the key handed to the lead in the session, never on a page

Not done.

• The twelve screenshots as redacted image files in evidence/: they arrived inline and could not be edited from the session; the transcriptions stand in for them
• The customer's correction of the mandate, and the signed licence

chat instruction, 16 Sept — no digest; given as speech, chat text or a link

Produced. The Gmail-connector vault pushed (oc433z3m) and read live on its page and in the library · The admin console: a rail with counts, what needs the lead, the board, the memo queue, every brief as a page, the vaults, the records, the tooling — written by build-admin.mjs from the repository · The Gmail workflow brief as a console page, one of every document under docs/

chat instruction with a PDF and a link, 16 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: what the mock-up says about the vault, the feature list read off the store's table, the dual licence as a file and a data block, the redesign of the reading app with a left navigation, what changes globally, four decisions for the lead, and the plan · Built: the dual licence in the build and LICENCE.md; the v5 left-navigation reading app in a new app vault (vbhmlulo); oc433z3m rebuilt and re-pushed; the vault page repositioned (v1.24.0)

Not done.

• The live host view of oc433z3m was still resolving the new sub-vault mount at merge; the vault data is verified correct and the renderer proven via a static preview

voice memo (Otter transcript, about nine minutes), 16 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: the memo in its own order, where the vault already is, the consequence layer (assets, consequences, standards mini-graphs, the triple), eleven first consequences for oc433z3m, what it settles for the MVP brief, the build order, what needs the lead · Task briefs T11 (consequences and assets) and T12 (standards mini-graphs); T04 amended with Who are you? · Built: assets.json and consequences.json for oc433z3m (eleven consequences, six assets, two routes out), CONSEQUENCES.md derived, the What follows view, the standards mini-graphs, two scenarios (v1.24.0)

Not done.

• The research list documented from Google's pages — sending limits, suspension, the modify tools — is still open (T05/research-vault)
• Who are you? ships as a live filter; the authored per-audience views and projections remain T04

the file as received · ab6f9486ecae… · 3 KB

Produced. The home page rebuilt around who is arriving and what is sold, with the insurance argument moved off it · Make agents insurable — the whole insurance argument kept in full on a page of its own · Licence to Operate — the middle rung, given the page the brief asked for · You run agents today · You are a founder · You are a startup · The menu restructured to seven top-level entries with Behaviour policy, Who it's for and Insurance as groups

Not done.

• MULTILINGUAL DELIVERY FOR TIER 3 AND TIER 4. The brief asks for the £500 correction and the £1,500 session to be deliverable in the buyer's own language, since both have a person in the loop. Nothing is built and nothing is claimed anywhere on the site. It is recorded here so it is not lost.
• The four offerings are not yet listed on the site as offerings. The store publishes four tiers and two add-ons; pricing.html still describes the free/paid line rather than the ladder.
• An investor page. The brief names three audiences; the printed sheets, the store and the Lisbon page all carry a fourth (investors, or 'you back companies that do'). Three pages were built and the fourth is flagged on the home page as missing.
• The vault's per-audience views are described on the home page as the reason for audience pages, but no ABP vault is published on this site yet.

the file as received · d93b0fdc381a… · 4 KB

Produced. The direction brief: the policy is a graph, and every stakeholder gets a projection of it (docs/briefs/direction__abp-as-a-graph-and-stakeholder-views.md) · The by-behaviour facet on the library: one click answers which policies can delete files

Not done.

• One page per behaviour, generated from the catalogue
• Edges per path rather than per row, in every vault
• Metrics (speed, volume, blast) and outward links (ATT&CK, GDPR) on the 23 primitives
• Views per audience — CEO, CFO, CTO, investor, buyer, operator, engineer, project manager — with the prompt and the script shipped in the vault so the projection can be regenerated
• The two asks to the model site, on Lab 03

Notes. The memo was worked from the transcript on 15 September before the file itself was registered; the file arrived the same day with D8. The transcriber writes the acronym as "ADP" throughout; it is the spoken "ABP", and the archived bytes are kept as received. The task briefs for the unbuilt items are in the repository under .claude/briefs/ (T01–T04, T07).

the file as received · 8457a6637e21… · 4 KB

Produced. The direction brief: a policy per use case, and the £500 level is a prompt the customer runs (docs/briefs/direction__use-case-driven-policies-and-the-prompt-workflow.md) · The pricing page rebuilt around the four levels, with the level-3 prompt workflow written out step by step and every level linked to the store · The homepage and every vault page point at viewing a policy and buying one at the store

Not done.

• The two Voice Debrief use-case vaults (the web flow through a model-routing service; the WhatsApp flow through n8n), written by the agent that knows those workflows and packaged here
• A page of its own for the £500 workflow, once the store's side has a return address that is not a mailbox
• A use-case group on the library page, and the combination of several vaults' grants into one
• The per-shape header on MAP-A-GRANT.md naming the vocabulary and the order reference
• Synchronising the level-3 wording with the store agent's page

Notes. Received with D7 on 15 September. The store's four levels went live the same day (store.sgit.ai v0.1.7); the memo's four price points match them.

the file as received · 99e75be33f09… · 31 KB · v0.33.71

Produced. Level 1 after payment: the page a £5 buyer lands on, where the zip of the template vault for the shape bought is downloaded, with its size and sha256 stamped from the file by the build and checked in CI; the page hashes the download in the browser on request. Unlisted and noindex; the payment link's success address, with ?shape=<slug> · Level 2 after payment: a working vault; a person follows up within 24 hours · Level 3 after payment: the prompt step and where to send what it produced; a person follows up within 24 hours · Level 4 after payment: the first session booked by a follow-up within 24 hours · Pricing: the plus-one-thing rule stated on the page, and an after-you-pay section with a definition of done per level, linking the four pages · The homepage: a four-level strip in the behaviour-policy section, and the £10 that was still on two pages corrected to £5

Not done.

• The payment links themselves, and the success address set on each of them (level 1 passes ?shape=<slug>) — the store's, not this site's
• The level-3 what-you-do text on the store's own product page, and the opinion add-on page
• The A5 and the physical stand

Notes. Received in chat on 15 September. The brief's site-side items are built; its store-side items are named and left to the store. The lead's follow-up the same day set the two things the brief left open: the £5 sale lands on a download, and levels 2–4 are a follow-up within 24 hours.

chat instruction, with two public comments supplied as the first two questions — no digest; given as speech, chat text or a link

Produced. Questions — real questions, answered with a date and no name attached

Not done.

• A third-party screenshot convention. The instruction anticipated detailed examples and screenshots; screenshots of our own demos are fine, and screenshots of anybody else's product run into the no-probing and no-verdicts rules, so the Lab's drawn mockups remain the pattern until that is decided.

the file as received · 1b2dfa45d228… · 29 KB · v0.33.70

Produced. Lab 01 — the grant is user-shaped, not data-shaped · Lab 02 — what buying a behaviour policy would look like, including the twelve-stage flow · Lab 03 — the request list against abp.sgit.ai, including the proposed material property

Not done.

• The five first policies themselves. Lab 01 documents the grants; no policy document exists for any of the five shapes.
• The template vault and the shape library — named in the brief as the actual product, and not started.
• The instrumentation table timing the first five, which the brief calls the only pricing input anybody will have.

Notes. Uploaded twice, byte-identical. The second arrival was in the same message as D3 and D4 and was not reprocessed.

the file as received · 11ff8f5f9eeb… · 31 KB · v0.33.70

Produced. Lab 04 — seven things to build, and one word we have not earned

Not done.

• All seven items on Lab 04's own build list. The page is the specification; none of it is built.

Notes. One correction to the brief is stated on the page and marked as ours: banding is the necessary first move and not the whole answer, because a banded submission still carries roughly the entropy of the fingerprint study the brief benchmarks against.

the file as received · b54eddae92bc… · 34 KB · v0.33.70

Produced. Lab 05 — your agent can commit as you, and no instruction stops it · The free/paid line on the pricing page, which now sends a reader to Lab 05's four free settings before asking for money

Not done.

• The comparison experiment. Lab 05 specifies it — three arms, twenty runs each, violations counted as repository queries — and nobody has set up a repository to run it against.
• A signed-commits rule on this site's own repository. Lab 05 says in as many words that the argument is demonstrated and not adopted until that is on.
• The provider and connector pages the brief maps onto the four layers, and the community incident repository behind them.

Notes. Read in full; the finding and the prompt are built, the experiment is not. All six load-bearing quotations were fetched and checked against their sources rather than relayed, which produced two corrections to the brief that are stated on the page: the partially-verified state additionally requires the author to have enabled vigilant mode, and the claim that the attribution renders a profile picture and link could not be found on the page cited.

the file as received · af73131b6a72… · 29 KB · v0.33.70

Produced. The urgency section on the front page — three vendor sentences with dates and no adjective, and the two dated changes of this year · The free/paid line on the pricing page, with the mistakes-not-attacks label and the authorise question in place of the insurance word

Not done.

• The price experiment as redesigned — charge one price and count, with a certainty question after. The pricing page states the approach; no price is set and nothing is charged.
• The generic prompt as a published, installable artefact. The pricing page says it is free and open; the file does not exist yet.
• The early access group: a dozen people running one of the five shapes, used for objections rather than numbers. No list exists.

Notes. Read in full. Its two rulings are now on the site: the entry product says it reduces accidents and does not stop an attacker, and the behaviour policy is not sold as a skill — the skill is the free half, because the portable frontmatter cannot carry a constraint. The word for the narrowing cover does not appear on the page carrying a price; the authorise question stands in its place.

the file as received · d13e3f08729f… · 2432 KB

Produced. The booth working page, with the guide embedded and the materials to hand over · Three corrections to the Lisbon messaging brief: no banner, the portal rather than email, and power must be requested

Not done.

• Logo and name into the exhibitor portal — the organisers' own deadline was 15 September.
• The power request, which goes through the portal and not by email.

Notes. Third-party material, republished on a working page rather than a public one. Whether it stays fetchable under site/ is an open decision.

the file as received · b85fdbcb235b… · 68 KB · v0.33.70

Produced. Lab 06 — every routable address is in the grant, and the one rule everybody signs has no barrier

Not done.

• The two proposals to the model site — the barrier's companion fields (above whom, deterministic, what voids it, evidenceable) and the six composition rules. Lab 06 states them; they are not yet on the request list in Lab 03.
• The provider comparison table, which the brief says writes itself from the sources: network on by default locally and hosted, allow list present, all egress paths or one, survives being written into a repository, fails open or closed, resolution covered, and a log the operator can read.
• The reconciliation tool between an agent's several destination lists — open question 3, and possibly the smallest useful thing we could ship on this capability.
• The generic pasteable document as an artefact. Lab 06 drafts it in a table; open question 4 is whether it should be published before the custom generator exists.

Notes. Seven load-bearing quotations were fetched and checked rather than relayed, including the reference container's firewall script. Two departures from the brief are stated on the page: the brief reports observing the two egress paths diverge in one session, and in ours they did not — both reached every host tried — so the finding rests on the vendor's own sentences rather than on that observation; and we published our own measurement of this machine's egress instead, which found a raw outbound socket to an arbitrary public address and working name resolution with no allow list in either path.

voice memo, transcribed in the thread — no digest; given as speech, chat text or a link

Produced. The direction brief · The Agent Behaviour Policy page · Pricing, rebuilt around the store's four levels and linked to it (v1.19.0)

Not done.

• The homepage's second panel, which should become "the grant you did not enumerate" and still does not.

chat instruction — no digest; given as speech, chat text or a link

Produced. Working with us · Brief B1 — power user and tester of Agent Behaviour Policies

Not done.

• Rate, hours, start date, escalation route, publication rules and repository access — deliberately absent because the page is public, and listed on it as owed.

chat instruction — no digest; given as speech, chat text or a link

Produced. Dated PDF editions of every Lab entry, and the whole Lab as one file · The edition register

link, in the thread — no digest; given as speech, chat text or a link

Produced. The Lisbon 2026 page · The messaging and strategy briefs