Village

Threat, function and solution map

What could go wrong when a small organisation relies on one provider for its mail, files and AI assistant, what must be true to be protected, which layers answer it and how far, and how you can check.

Version 2, 10 November 2026 · proposed by AI agents, not yet reviewed

Summary diagram: the map on one page, for a board

How to read it

Every box is a claim with its own source and an exact quote, checked by a program against that source before this page is built. The 12 threats are stated for any single provider, not as claims about a named company. The tools named, Microsoft 365, Copilot and Outlook, are where the map starts, because that is where most small organisations are. A threat that nothing here answers stays open, and the gap list says so.

Each claim carries a tier: verified in code (the quote is on a named line of the platform's code), documented on a product page, stated by us (our claims register or our guide), or open (no source can be quoted, because the claim is that something does not exist). Each claim also says what would show it wrong. The drawing and the outline below carry the same claims; the outline is the full reading.

This map was drafted by AI agents from our pages, our claims register and the platform's code. Until a person at My Digital Sovereignty Ltd has reviewed a claim, it is marked proposed by AI agents, not yet reviewed.

The map

Views

Each view picks out part of the map. With scripts on, the outline shows only those claims and the drawing dims the rest; with scripts off, each link goes to that view's list further down.

Threat, function and solution map, drawn 51 claims in four columns, left to right: threats, what must be true, layers and evidence, with 73 links between them. Each box is a link to the same claim in the outline below, which says everything the drawing shows. Threats What must be true Layers Evidence A written mandate for each agent reduces AI agents that overstep; does not cover: an agent nobody listed, or an action the mandate wrongly allows A different person approves reduces AI agents that overstep; does not cover: actions an agent takes without asking first A different person approves reduces An email sent by an agent nobody approved; does not cover: mail sent from a mailbox the agent reaches directly A different person approves reduces Messages that sound like someone you know; does not cover: a request that never reaches a gate, such as a phone call A different person approves reduces Confident AI mistakes; does not cover: an approver who does not read closely Permission that runs out reduces AI agents that overstep; does not cover: a wrong action carried out while the permission is still valid An independent witness reduces Your provider is also the only witness; does not cover: what happens inside the provider's systems before it reaches the witness An independent witness reduces A record changed after the event; does not cover: a witness that can itself rewrite what it holds A record anyone can check offline detects A record changed after the event; does not cover: whether the record was true when it was signed A record anyone can check offline reduces Locks that will not hold forever; does not cover: records that are not signed again under newer methods before the old ones weaken Know who owns the provider and holds the keys reduces Foreign law; does not cover: the move itself: knowing the answer does not take the data anywhere Know who owns the provider and holds the keys reduces A provider you cannot leave; does not cover: the work of moving records and habits to another system Know who owns the provider and holds the keys reduces Locks that will not hold forever; does not cover: asking for a provider's plan is not the provider having one Village Gateway linked A different person approves: provides it: a held request waits for a different approver The requester cannot approve its own request linked A different person approves: provides it in code: the requester's approval is refused Four eyes on an agent's email linked A different person approves: provides it for email: the person who registered an agent cannot release its held email Permits run out linked Permission that runs out: provides it: a single-use permit with a time ceiling MIO° Conduct Record linked A written mandate for each agent: provides it: a sealed register of what each agent may do The Village linked An independent witness: provides it, from a second provider: MDSL keeps the signed record An outside clock on each Gateway receipt linked An independent witness: adds an outside clock that sees only a fingerprint Village Gateway linked A record anyone can check offline: provides it: an exported decision checks offline Email Decision Record linked A record anyone can check offline: provides it: a dossier checks offline MIO° Conduct Record linked A record anyone can check offline: provides it: an auditor checks offline Attested Accounts linked A record anyone can check offline: provides it for invoices Provenance linked A record anyone can check offline: provides it: a buyer checks in a browser Signing keys on a separate host, held by MDSL linked Know who owns the provider and holds the keys: answers it for these layers: MDSL holds the signing keys Held in New Zealand or the EU, by providers that are not US-owned linked Know who owns the provider and holds the keys: answers it for these layers: New Zealand or EU hosts, not US-owned Outlook add-in: seal a message and link its reply linked A record anyone can check offline: provides it for a sealed message: its record checks offline Teams tab: seal a copy of pasted text linked A record anyone can check offline: provides it for pasted text: the saved copy checks against its record Seal a file: a document version sealed by its digest linked A record anyone can check offline: provides it for a file: any copy of the same version checks against its record Village Gateway reduces AI agents that overstep; does not cover: agents that do not call the Gateway, and systems that do not check its permits Village Gateway reduces Messages that sound like someone you know; does not cover: a convincing request read and acted on by a person outside the Gateway Village Gateway linked Unfair automated decisions: records which rule applied to each decision, so a decision can be read and appealed; it does not show how outcomes fall by group A spending limit for each agent reduces AI agents that overstep; does not cover: requests that carry no amount, which are not counted Stop every Gateway decision at once reduces AI agents that overstep; does not cover: agent keys, which stay valid after a stop Permits run out reduces AI agents that overstep; does not cover: a wrong action approved and carried out within the permit's life Checking an action against the mandate in force reduces AI agents that overstep; does not cover: agents whose key is not linked to a mandate; the check is off until turned on MIO° Conduct Record detects AI agents that overstep; does not cover: the action itself: the register records authority and blocks nothing Email Decision Record reduces An email sent by an agent nobody approved; does not cover: mail an agent sends straight from a mailbox without proposing it; off by default Email Decision Record reduces Confident AI mistakes; does not cover: the truth of what the email says: the approver sees it, nothing checks it Four eyes on an agent's email reduces An email sent by an agent nobody approved; does not cover: a sole owner who turns the rule off Sending through a Microsoft account with send-only permission reduces An email sent by an agent nobody approved; does not cover: mail sent from Outlook directly; the connection is off by default and set up on request Hash-linked records detects A record changed after the event; does not cover: records from before hash-linking began, removal of the newest entries, and rewriting by the key holder Signing keys on a separate host, held by MDSL reduces A record changed after the event; does not cover: MDSL itself, which holds the keys Attested Accounts detects A record changed after the event; does not cover: whether the figures are right; quotes, expenses and time entries have no bundle yet Provenance detects A record changed after the event; does not cover: whether what a party recorded is true The Village reduces Your provider is also the only witness; does not cover: the second witness is MDSL, which holds the signing keys An outside clock on each Gateway receipt reduces Your provider is also the only witness; does not cover: content: the authority sees a fingerprint only Held in New Zealand or the EU, by providers that are not US-owned reduces Foreign law; does not cover: data left in Microsoft 365, and dependencies outside record storage A second, post-quantum signature reduces Locks that will not hold forever; does not cover: Gateway receipts, invoices, other Village writes, and encryption of stored data Export a Village's shared records reduces A provider you cannot leave; does not cover: files and mail kept in Microsoft 365; there is no screen for the export Check one suspicious email detects Messages that sound like someone you know; does not cover: voice and video, and link reputation, which is shown as not checked Outlook add-in: seal a message and link its reply reduces A record changed after the event; does not cover: who wrote the message, or a message nobody sealed; mail written in Outlook is not gated, and what Copilot does is not seen Teams tab: seal a copy of pasted text reduces A record changed after the event; does not cover: who wrote the text, that it appeared in Teams, or text nobody sealed Seal a file: a document version sealed by its digest reduces A record changed after the event; does not cover: who wrote the file, that it was in SharePoint or OneDrive, or a version nobody sealed; a very small text file's digest can be guessed A Copilot connector to Village Gateway linked AI agents that overstep: gap: a Copilot action is not gated unless the organisation builds the call Reading a Microsoft 365 mailbox through Microsoft sign-in linked Messages that sound like someone you know: gap: replies to a Microsoft 365 mailbox are not recorded through Microsoft sign-in Village Gateway depends on The Village: switched on Village by Village Checking an action against the mandate in force depends on Village Gateway: runs only on requests the Gateway governs Attested Accounts depends on The Village: sits inside the Business Village Outlook add-in: seal a message and link its reply depends on Signing keys on a separate host, held by MDSL: signed like every record, with a key MDSL holds Teams tab: seal a copy of pasted text depends on Signing keys on a separate host, held by MDSL: signed like every record; the signing host is not yet updated for this record kind Seal a file: a document version sealed by its digest depends on Signing keys on a separate host, held by MDSL: signed like every record; the signing host is not yet updated for this record kind The Gateway checking program linked Village Gateway: checks an exported decision The Conduct Record offline check linked MIO° Conduct Record: checks a Conduct Record The Email Decision Record checker linked Email Decision Record: checks a dossier An invoice's offline verification bundle linked Attested Accounts: checks an invoice's history The Provenance check in a browser linked Provenance: checks a lot's chain Each checking program published with its SHA-256 linked The Gateway checking program: confirms the program run is the one published A check never reads as verified without a pinned key linked Signing keys on a separate host, held by MDSL: a pinned key confirms which key signed; MDSL holds it What a check cannot show linked A record anyone can check offline: the limit of every check: unchanged since signing, not true The Email Decision Record checker linked Outlook add-in: seal a message and link its reply: reads a sealed message's record The Email Decision Record checker linked Seal a file: a document version sealed by its digest: reads a document seal AI agents that overstepAI agents that overstep■ On a page · Partly answered An email sent by an agent nobody approvedAn email sent by an agent nobodyapproved● Stated by us · Partly answered Messages that sound like someone you knowMessages that sound like someoneyou know■ On a page · Partly answered Confident AI mistakesConfident AI mistakes■ On a page · Partly answered Unfair automated decisionsUnfair automated decisions■ On a page · No answer on this map A record changed after the eventA record changed after the event■ On a page · Partly answered Your provider is also the only witnessYour provider is also the onlywitness● Stated by us · Partly answered Foreign lawForeign law■ On a page · Partly answered A provider you cannot leaveA provider you cannot leave● Stated by us · Partly answered Locks that will not hold foreverLocks that will not hold forever■ On a page · Partly answered Leaked passwords and member detailsLeaked passwords and memberdetails■ On a page · No answer on this map Pictures you cannot vouch forPictures you cannot vouch for■ On a page · No answer on this map A written mandate for each agentA written mandate for each agent■ On a page A different person approvesA different person approves■ On a page Permission that runs outPermission that runs out● Stated by us An independent witnessAn independent witness● Stated by us A record anyone can check offlineA record anyone can check offline■ On a page Know who owns the provider and holds the keysKnow who owns the provider andholds the keys■ On a page Village GatewayVillage Gateway● Stated by us · Built A Copilot connector to Village GatewayA Copilot connector to VillageGateway○ Open · Not built MIO° Conduct RecordMIO° Conduct Record● Stated by us · Available by arrangement Email Decision RecordEmail Decision Record■ On a page · Off by default Reading a Microsoft 365 mailbox through Microsoft sign-inReading a Microsoft 365 mailboxthrough Microsoft sign-in○ Open · Not built The VillageThe Village● Stated by us · Built Attested AccountsAttested Accounts● Stated by us · Available by arrangement ProvenanceProvenance■ On a page · Available by arrangement The requester cannot approve its own requestThe requester cannot approve itsown request◆ In code · Built Permits run outPermits run out◆ In code · Built Stop every Gateway decision at onceStop every Gateway decision atonce● Stated by us · Built A spending limit for each agentA spending limit for each agent● Stated by us · Built An outside clock on each Gateway receiptAn outside clock on each Gatewayreceipt● Stated by us · Built Checking an action against the mandate in forceChecking an action against themandate in force● Stated by us · Off by default Four eyes on an agent's emailFour eyes on an agent's email■ On a page · Off by default Sending through a Microsoft account with send-only permissionSending through a Microsoftaccount with send-only permission◆ In code · Off by default Outlook add-in: seal a message and link its replyOutlook add-in: seal a message andlink its reply◆ In code · Off by default Teams tab: seal a copy of pasted textTeams tab: seal a copy of pastedtext◆ In code · Off by default Seal a file: a document version sealed by its digestSeal a file: a document versionsealed by its digest◆ In code · Off by default Hash-linked recordsHash-linked records● Stated by us · Built Signing keys on a separate host, held by MDSLSigning keys on a separate host,held by MDSL● Stated by us · Built A second, post-quantum signatureA second, post-quantum signature● Stated by us · Built Held in New Zealand or the EU, by providers that are not US-ownedHeld in New Zealand or the EU, byproviders that are not US-owned● Stated by us · Built Export a Village's shared recordsExport a Village's shared records● Stated by us · Built The Gateway checking programThe Gateway checking program■ On a page Each checking program published with its SHA-256Each checking program publishedwith its SHA-256● Stated by us The Conduct Record offline checkThe Conduct Record offline check■ On a page The Email Decision Record checkerThe Email Decision Record checker■ On a page Check one suspicious emailCheck one suspicious email● Stated by us A check never reads as verified without a pinned keyA check never reads as verifiedwithout a pinned key■ On a page What a check cannot showWhat a check cannot show● Stated by us An invoice's offline verification bundleAn invoice's offline verificationbundle■ On a page The Provenance check in a browserThe Provenance check in a browser■ On a page Systems map: proposed by AI agents, not yet reviewed, for My Digital Sovereignty Ltd. Version 2, 10 November 2026.

Threat (cut corners) · What must be true (rounded ends) · Layer · Evidence (double edge) · Dashed edge: open, not built or no answer

Tier, printed in each box: ◆ In code · ■ On a page · ● Stated by us · ○ Open. Status of a layer: Built · Available by arrangement · Off by default · Not built.

Lines: reduces (solid) · detects (long dashes) · depends on (dash and gap) · linked (dots). Every line that reduces or detects says, in the outline, what it does not cover.

Threat, function and solution map · Version 2, 10 November 2026 · proposed by AI agents, not yet reviewed · The current version: https://mysovereignty.digital/ring-map.html

The gap list: what this map does not answer

No answer on this map (3)

Not built (2)

Partly answered, and what is left (9)

  • AI agents that overstep
    • A written mandate for each agent does not cover an agent nobody listed, or an action the mandate wrongly allows
    • A different person approves does not cover actions an agent takes without asking first
    • Permission that runs out does not cover a wrong action carried out while the permission is still valid
    • Village Gateway does not cover agents that do not call the Gateway, and systems that do not check its permits
    • A spending limit for each agent does not cover requests that carry no amount, which are not counted
    • Stop every Gateway decision at once does not cover agent keys, which stay valid after a stop
    • Permits run out does not cover a wrong action approved and carried out within the permit's life
    • Checking an action against the mandate in force does not cover agents whose key is not linked to a mandate; the check is off until turned on
    • MIO° Conduct Record does not cover the action itself: the register records authority and blocks nothing
  • An email sent by an agent nobody approved
    • A different person approves does not cover mail sent from a mailbox the agent reaches directly
    • Email Decision Record does not cover mail an agent sends straight from a mailbox without proposing it; off by default
    • Four eyes on an agent's email does not cover a sole owner who turns the rule off
    • Sending through a Microsoft account with send-only permission does not cover mail sent from Outlook directly; the connection is off by default and set up on request
  • Messages that sound like someone you know
    • A different person approves does not cover a request that never reaches a gate, such as a phone call
    • Village Gateway does not cover a convincing request read and acted on by a person outside the Gateway
    • Check one suspicious email does not cover voice and video, and link reputation, which is shown as not checked
  • Confident AI mistakes
    • A different person approves does not cover an approver who does not read closely
    • Email Decision Record does not cover the truth of what the email says: the approver sees it, nothing checks it
  • A record changed after the event
    • An independent witness does not cover a witness that can itself rewrite what it holds
    • A record anyone can check offline does not cover whether the record was true when it was signed
    • Hash-linked records does not cover records from before hash-linking began, removal of the newest entries, and rewriting by the key holder
    • Signing keys on a separate host, held by MDSL does not cover MDSL itself, which holds the keys
    • Attested Accounts does not cover whether the figures are right; quotes, expenses and time entries have no bundle yet
    • Provenance does not cover whether what a party recorded is true
    • Outlook add-in: seal a message and link its reply does not cover who wrote the message, or a message nobody sealed; mail written in Outlook is not gated, and what Copilot does is not seen
    • Teams tab: seal a copy of pasted text does not cover who wrote the text, that it appeared in Teams, or text nobody sealed
    • Seal a file: a document version sealed by its digest does not cover who wrote the file, that it was in SharePoint or OneDrive, or a version nobody sealed; a very small text file's digest can be guessed
  • Your provider is also the only witness
    • An independent witness does not cover what happens inside the provider's systems before it reaches the witness
    • The Village does not cover the second witness is MDSL, which holds the signing keys
    • An outside clock on each Gateway receipt does not cover content: the authority sees a fingerprint only
  • Foreign law
    • Know who owns the provider and holds the keys does not cover the move itself: knowing the answer does not take the data anywhere
    • Held in New Zealand or the EU, by providers that are not US-owned does not cover data left in Microsoft 365, and dependencies outside record storage
  • A provider you cannot leave
    • Know who owns the provider and holds the keys does not cover the work of moving records and habits to another system
    • Export a Village's shared records does not cover files and mail kept in Microsoft 365; there is no screen for the export
  • Locks that will not hold forever
    • A record anyone can check offline does not cover records that are not signed again under newer methods before the old ones weaken
    • Know who owns the provider and holds the keys does not cover asking for a provider's plan is not the provider having one
    • A second, post-quantum signature does not cover Gateway receipts, invoices, other Village writes, and encryption of stored data

The map in words

The same 51 claims and 73 links as the drawing, band by band. Every name in a link sentence goes to that claim.

  1. Threats: what could go wrong (12)

    1. AI agents that overstep

      Threat Tier: Documented on a product page Partly answered Tools: Copilot, Microsoft 365, Outlook Proposed by AI agents, not yet reviewed

      An AI agent given tools such as email, payments, calendars or files does more than anyone intended, because it was given too much authority, too many permissions or too little oversight.

      Source
      Product page /why-trust.html, at commit 90b235706c: An AI agent given tools (email, payments, calendars, files) does more than anyone intended, because it was given too much authority, too many permissions or too little oversight.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show an agent with access to email, payments or files that cannot do more than one person intended, on any provider, without a gate outside the agent: the threat would then be overstated.
      Audit note
      Risk 6 on the risks page, which rests on the OWASP list of risks for LLM applications. Stated at provider level, not as a claim about any named product.
    2. An email sent by an agent nobody approved

      Threat Tier: Stated by us Partly answered Tools: Outlook, Copilot, Microsoft 365 Proposed by AI agents, not yet reviewed

      An AI assistant that can reach a mailbox sends a message, with or without attachments, to someone it should not, and no person saw it before it went.

      Source
      The guide, Stepping outside the ring: Your rules decide each email from a person or an AI agent before it goes
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Find an assistant with mailbox access whose every outgoing message is seen by a person first, with nothing added: the risk would then be smaller than stated.
      Audit note
      An AI-specific case of risk 6 (AI agents that overstep), added for this map and stated by us: the guide names it through its remedy, a gate on each email from a person or an AI agent before it goes.
    3. Messages that sound like someone you know

      Threat Tier: Documented on a product page Partly answered Tools: Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      A request that copies a trusted person's voice, face or writing style asks for money, access or a change of bank details.

      Source
      Product page /why-trust.html, at commit 90b235706c: Scams that use AI to copy a voice, a face or a writing style, so a request seems to come from someone you trust.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Find a request of this kind that a layer on this map would have held or flagged. If none of them sees it, for example a phone call, the partial mark is too generous.
      Audit note
      Risk 1 on the risks page. The layers here see only what passes through a gate or a check: an email proposed through the Email Decision Record, a request made through Village Gateway, or one message uploaded to the email check.
    4. Confident AI mistakes

      Threat Tier: Documented on a product page Partly answered Tools: Copilot, Outlook Proposed by AI agents, not yet reviewed

      An AI assistant states something false as fluently as something true, in a summary, a draft reply or a quoted rule.

      Source
      Product page /why-trust.html, at commit 90b235706c: An AI assistant states something false with the same fluency as something true.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show a layer on this map that checks whether what an AI wrote is true. None is claimed: the answer here is only that a person sees the exact message first.
      Audit note
      Risk 3 on the risks page. A held email is shown to its approver exactly as it would be sent; nothing on this map checks its content.
    5. Unfair automated decisions

      Threat Tier: Documented on a product page No answer on this map Tools: Copilot, Microsoft 365 Proposed by AI agents, not yet reviewed

      Software that sorts, ranks or screens people learns patterns from old data, and those patterns can disadvantage some groups.

      Source
      Product page /why-trust.html, at commit 90b235706c: Software that sorts, ranks or screens people learns patterns from old data, and those patterns can disadvantage some groups.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show a layer here that reviews outcomes by group. None does: Village Gateway records which rule applied to each decision, which makes a decision readable, not fair.
      Audit note
      Risk 5 on the risks page. No layer on this map is offered as the answer: recorded reasons make a decision open to appeal, which is not a review of outcomes by group.
      • Linked from Village Gateway: records which rule applied to each decision, so a decision can be read and appealed; it does not show how outcomes fall by group.
    6. A record changed after the event

      Threat Tier: Documented on a product page Partly answered Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      A minute, invoice or decision is edited, back-dated or removed after the fact, and the copy that remains reads as though it was always so.

      Source
      Product page /attested-accounts.html, at commit 90b235706c: so a later change to a record can be detected by you, your accountant or an auditor
      Why this tier
      The quote is on /attested-accounts.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Change a signed record on this platform and present it to a checker: if the checker passes it, the detection claims attached to this threat are wrong.
      Audit note
      An AI-specific case added for this map. The source names the risk through its remedy on the Attested Accounts page.
    7. Your provider is also the only witness

      Threat Tier: Stated by us Partly answered Tools: Microsoft 365, Copilot, Outlook Proposed by AI agents, not yet reviewed

      The provider that runs the platform is also the only party able to say what happened on it, so nobody outside can check its account.

      Source
      The guide, Stepping outside the ring: A provider can protect its own platform. It cannot also be the independent witness of what happened on it.
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Find a provider whose own records of activity on its platform can be checked by an outsider without trusting that provider: the threat would then not hold for it.
      Audit note
      Our framing, from the guide. Stated at provider level; it applies to any single provider, this one included.
    8. Foreign law

      Threat Tier: Documented on a product page Partly answered Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      Data is held by companies, and a company answers to the law of the country that owns it, not only to the law of the country where its servers stand.

      Source
      Product page /why-trust.html, at commit 90b235706c: Data is held by companies, and companies answer to the law of the country that owns them, not only the country where the servers stand.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show that a provider's parent company cannot be required by its home country to produce data held abroad: the threat would then be overstated for that provider.
      Audit note
      Risk 7 on the risks page, which cites the US CLOUD Act text. The map does not say how often this happens to small organisations.
    9. A provider you cannot leave

      Threat Tier: Stated by us Partly answered Tools: Microsoft 365, Outlook Proposed by AI agents, not yet reviewed

      Records, mail and working habits sit with one provider, so leaving it, or even checking what it holds, depends on that provider.

      Source
      The guide, Stepping outside the ring: Most small organisations put their trust in the provider they already use and may even expect it to supply protection and sovereignty.
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Show that an organisation can take all its records and mail from its provider, in a form another system reads, without that provider's help: the threat would then be smaller.
      Audit note
      An AI-specific case added for this map, in our words. The quote shows the dependence; the difficulty of leaving is our inference from it.
    10. Locks that will not hold forever

      Threat Tier: Documented on a product page Partly answered Tools: Microsoft 365, Outlook Proposed by AI agents, not yet reviewed

      Today's encryption and signatures rest on mathematics that a large enough quantum computer could undo, and records stored today could be read later.

      Source
      Product page /why-trust.html, at commit 90b235706c: rest on mathematics that a large enough quantum computer could undo
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show that the signature and encryption methods in common use will stay sound for the lifetime of a long-lived record: the threat would then be overstated.
      Audit note
      Risk 8 on the risks page, which cites NIST and the UK National Cyber Security Centre. A horizon, not an alarm.
    11. Leaked passwords and member details

      Threat Tier: Documented on a product page No answer on this map Tools: Microsoft 365, Outlook Proposed by AI agents, not yet reviewed

      A breach at any service people use exposes passwords and personal details, which are then tried on other services.

      Source
      Product page /why-trust.html, at commit 90b235706c: A security breach at any service your members use, which exposes passwords and personal details that are then tried elsewhere.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show a layer on this map that stops a leaked or reused password being used. None is claimed here, so this threat stays open.
      Audit note
      Risk 2 on the risks page. Two-step sign-in and a different password for each service are the advice there; no layer on this map is offered as the answer.
      • Nothing on this map answers it.
    12. Pictures you cannot vouch for

      Threat Tier: Documented on a product page No answer on this map Tools: Copilot, Microsoft 365 Proposed by AI agents, not yet reviewed

      Images, audio and video are generated or altered well enough that the eye cannot tell.

      Source
      Product page /why-trust.html, at commit 90b235706c: Images, audio and video that are generated or altered well enough that the eye cannot tell.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show a layer on this map that signs a photograph or a recording when it is made. None does, so this threat stays open.
      Audit note
      Risk 4 on the risks page. Provenance signs records of goods along a supply chain, not media, so it is not drawn against this threat.
      • Nothing on this map answers it.
  2. What must be true to be protected (6)

    1. A written mandate for each agent

      What must be true Tier: Documented on a product page Tools: Copilot, Microsoft 365 Proposed by AI agents, not yet reviewed

      Each AI agent has a written statement of what it may do, kept where it can be shown later.

      Source
      Product page /why-trust.html, at commit 90b235706c: A careful community writes down what each agent may do
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Find a case where a written mandate made no difference to what could be shown afterwards about an agent's authority: the need would then be overstated.
      Audit note
      From the advice under risk 6 on the risks page. The guide's first step this week is the same list.
    2. A different person approves

      What must be true Tier: Documented on a product page Tools: Copilot, Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      An action with real consequences waits for a person other than the one, or the agent, that asked for it.

      Source
      Product page /why-trust.html, at commit 90b235706c: require a person to approve high-impact actions
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show that a second approver adds nothing when the first is the requester or the requester's agent: the need would then be overstated.
      Audit note
      One of the OWASP remedies quoted under risk 6 on the risks page. The word different is ours: the page says a person.
    3. Permission that runs out

      What must be true Tier: Stated by us Tools: Copilot Proposed by AI agents, not yet reviewed

      Permission to act covers one action for a short time, so a stolen or replayed permission is of little use.

      Source
      The guide, Stepping outside the ring: An allowed action gets a single-use permit that lasts at most five minutes.
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Show a replayed or late permission that a protected system still accepts: the need would then be unmet by the layer that claims it.
      Audit note
      Stated by us in the guide as a property of Village Gateway; the need itself is general.
      • Reduces AI agents that overstep; does not cover a wrong action carried out while the permission is still valid.
      • Linked from Permits run out: provides it: a single-use permit with a time ceiling.
    4. An independent witness

      What must be true Tier: Stated by us Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      Someone other than the provider keeps a signed, dated record of what was done, by whom and with whose approval.

      Source
      The guide, Stepping outside the ring: This guide adds that witness around the tools you already use
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Show that a second party's record adds nothing to what the provider's own record shows: the need would then be overstated.
      Audit note
      Our framing, from the guide. The witness offered on this map is MDSL, a second provider, not a neutral third party.
    5. A record anyone can check offline

      What must be true Tier: Documented on a product page Tools: Microsoft 365, Outlook Proposed by AI agents, not yet reviewed

      A record can be checked by whoever holds it, without an account and without asking the provider.

      Source
      Product page /why-trust.html, at commit 90b235706c: prefers signatures that can be checked without the provider
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Show a record from one of the layers here that cannot be checked without an account or a network call to MDSL: the need would then be unmet for that layer.
      Audit note
      From the advice under risk 8 on the risks page.
    6. Know who owns the provider and holds the keys

      What must be true Tier: Documented on a product page Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      Before relying on a provider, an organisation knows who owns it, which laws reach it and who holds the keys.

      Source
      Product page /why-trust.html, at commit 90b235706c: A careful community asks a provider who owns the company, which laws reach it, and who holds the keys.
      Why this tier
      The quote is on /why-trust.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Ask the question of the layers on this map: if MDSL's answer is not on its own pages, the need is unmet here too.
      Audit note
      From the advice under risk 7 on the risks page. The answer for this map's layers is on it: MDSL holds the signing keys.
  3. Layers: what answers it, and how far (24)

    1. Village Gateway

      Layer Tier: Stated by us Status: Built (register GTW-01) Tools: Copilot, Microsoft 365 Proposed by AI agents, not yet reviewed

      An agent or a person asks before acting; the Village's policy answers allow, hold or deny; a held request waits for a different approver; an allowed request gets a one-use permit.

      Source
      Claims register entry GTW-01: an approval-gated request waits for a different approver, and an allowed request gets a one-use permit that a resource redeems exactly once
      Why this tier
      The quote is in our claims register, entry GTW-01: our statement, not an outside check.
      What would show it wrong
      Find a Gateway decision held for approval that was released by the requester, or a permit redeemed twice: either would show this claim wrong.
      Audit note
      Register status built. It is turned on Village by Village and governs only agents that call it and systems that check its permits. No customer has used it. The register records it as built.
    2. A Copilot connector to Village Gateway

      Layer Tier: Open: no source Status: Not built Tools: Copilot Proposed by AI agents, not yet reviewed

      No connector makes a Copilot action ask Village Gateway first, and the Village has no Copilot integration: it does not see or govern what Copilot does. A Copilot action reaches the Gateway only if something the organisation builds calls it.

      Source
      None: a gap, shown by absence.
      Why this tier
      No source can be quoted: this is a gap, shown by absence (see the audit note). Map version 2.
      What would show it wrong
      Find a Copilot plugin, declarative agent or Copilot Studio connector in the platform that calls Village Gateway: this gap would then be closed.
      Audit note
      A gap, shown by absence: a code census of the platform found no Copilot integration of any kind. The Outlook, Teams and file connectors do not change this. An adapter for a Microsoft agent framework exists in the repository but is not served or offered, so it is not claimed.
    3. The requester cannot approve its own request

      Layer Tier: Verified in code Status: Built (register GTW-13) Tools: Copilot Proposed by AI agents, not yet reviewed

      Village Gateway refuses an approval when the approver is the requester.

      Source
      Code: src/services/gateway/gatewayService.js, line 1811, at commit 90b235706c: throw err('APPROVER_IS_REQUESTER', 'the approver may not be the requester');
      Why this tier
      The quote is on line 1811 of src/services/gateway/gatewayService.js at commit 90b235706c, checked by the validator against that commit.
      What would show it wrong
      Approve a held Gateway request while signed in as the owner of the requesting key: if it is released, the claim is wrong.
      Audit note
      The line before compares the approver with both the requesting principal and the owner of the requesting key. The register adds that agents have no approve action at all.
    4. Permits run out

      Layer Tier: Verified in code Status: Built (register GTW-09) Tools: Copilot Proposed by AI agents, not yet reviewed

      Village Gateway refuses to issue a permit that would live longer than its ceiling of 900 seconds, fifteen minutes.

      Source
      Code: src/services/gateway/permit.js, line 67, at commit 90b235706c: throw new GatewayError('PERMIT_TTL_TOO_LONG', `a permit may live at most ${MAX_TTL_SECONDS} seconds`)
      Why this tier
      The quote is on line 67 of src/services/gateway/permit.js at commit 90b235706c, checked by the validator against that commit.
      What would show it wrong
      Ask for a permit lasting longer than fifteen minutes: if one is issued, the claim is wrong.
      Audit note
      The ceiling is set at line 18 of the same file. The register adds that the default permit lasts five minutes and never outlasts its decision.
    5. Stop every Gateway decision at once

      Layer Tier: Stated by us Status: Built (register GTW-11) Tools: Copilot Proposed by AI agents, not yet reviewed

      The Village owner can stop Village Gateway in one step: outstanding permits are withdrawn and nothing is decided until the stop is lifted.

      Source
      Claims register entry GTW-11: The Village owner can stop Village Gateway in one step: outstanding permits are withdrawn, and nothing is decided, approved or redeemed until the owner lifts the stop.
      Why this tier
      The quote is in our claims register, entry GTW-11: our statement, not an outside check.
      What would show it wrong
      Stop the Gateway and redeem a permit issued before the stop: if it works, the claim is wrong.
      Audit note
      Register status built. Agent keys are not revoked by the stop, and held requests keep their hold.
    6. A spending limit for each agent

      Layer Tier: Stated by us Status: Built (register GTW-12) Tools: Copilot Proposed by AI agents, not yet reviewed

      The Village owner can set a spending limit on each agent key, per currency, per day, week or month; a request over the limit is refused, not held.

      Source
      Claims register entry GTW-12: The Village owner can set a spending limit on each agent key, per currency, per day, week or month; a request over the limit is refused, not held.
      Why this tier
      The quote is in our claims register, entry GTW-12: our statement, not an outside check.
      What would show it wrong
      Make a request over an agent key's limit: if it is allowed or held, the claim is wrong.
      Audit note
      Register status built. A request with no amount is not counted against the limit.
    7. An outside clock on each Gateway receipt

      Layer Tier: Stated by us Status: Built (register GTW-07) Tools: Copilot Proposed by AI agents, not yet reviewed

      Village Gateway sends every receipt it signs to an outside time-stamping authority, so the time of a decision does not rest on MDSL's clock alone.

      Source
      Claims register entry GTW-07: Village Gateway sends every receipt it signs to an outside time-stamping authority (Certum)
      Why this tier
      The quote is in our claims register, entry GTW-07: our statement, not an outside check.
      What would show it wrong
      Export a Gateway decision after its stamp has arrived and run the checking program: if the stamp does not verify, the claim is wrong.
      Audit note
      Register status built. The stamp arrives after the receipt is signed, so an early export reads incomplete; the authority sees only a fingerprint.
    8. MIO° Conduct Record

      Layer Tier: Stated by us Status: Available by arrangement (register CON-01) Tools: Copilot, Microsoft 365 Proposed by AI agents, not yet reviewed

      A sealed register of what an organisation's AI agents are authorised to do, and which agents exist; a change to a mandate is a new dated record and the old one is kept.

      Source
      Claims register entry CON-01: MIO° Conduct Record is a sealed register of what an organisation's AI agents are authorised to do, and which agents exist.
      Why this tier
      The quote is in our claims register, entry CON-01: our statement, not an outside check.
      What would show it wrong
      Find a mandate change that replaced the earlier record instead of superseding it: the claim would then be wrong.
      Audit note
      Register status available on request. It records authority and blocks nothing itself; the mandate check sits in Village Gateway and is off until turned on.
    9. Checking an action against the mandate in force

      Layer Tier: Stated by us Status: Off by default (register CON-07) Tools: Copilot Proposed by AI agents, not yet reviewed

      Where a policy requires a mandate, Village Gateway denies a request whose action, resource, amount or dates fall outside the agent's mandate in force.

      Source
      Claims register entry CON-07: Village Gateway checks any request it governs, where a policy requires a mandate
      Why this tier
      The quote is in our claims register, entry CON-07: our statement, not an outside check.
      What would show it wrong
      Turn the check on and make a request outside the mandate in force: if it is allowed, the claim is wrong.
      Audit note
      Built; turned off until a Village adopts it. It acts only for agents whose key the owner has linked to a mandate; no customer has used it.
    10. Email Decision Record

      Layer Tier: Documented on a product page Status: Off by default (register EDR-03) Tools: Outlook, Microsoft 365, Copilot Proposed by AI agents, not yet reviewed

      Each email a person or an AI agent proposes is decided by the owner's policy before it goes: send, hold for an approver, or refuse. It sits beside the existing mailbox and leaves a signed record.

      Source
      Product page /email-decision-record-product.html, at commit 90b235706c: It works beside the mailbox you already use: Gmail, Microsoft 365 or any SMTP mail service. It adds a decision and a record. It does not replace your mail.
      Why this tier
      The quote is on /email-decision-record-product.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Find a proposed email that went out without a decision record, or a held email released without an approver: either would show this claim wrong.
      Audit note
      Built; turned off until a Village adopts it. No customer is live. It decides only mail proposed to it; mail sent straight from a mailbox is not seen.
    11. Reading a Microsoft 365 mailbox through Microsoft sign-in

      Layer Tier: Open: no source Status: Not built Tools: Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      No route reads a Microsoft 365 mailbox through Microsoft sign-in, so replies are not recorded on their own. The capture reader signs in with a username and password; the Outlook add-in records a reply only when a person seals it.

      Source
      None: a gap, shown by absence.
      Why this tier
      No source can be quoted: this is a gap, shown by absence (see the audit note). Map version 2.
      What would show it wrong
      Find a capture path in the platform that reads a Microsoft 365 mailbox with Microsoft sign-in: this gap would then be closed.
      Audit note
      A gap. The code census found the capture reader uses password sign-in over IMAP only, with no Microsoft sign-in path. Whether a given Microsoft 365 mailbox accepts password sign-in is a setting on Microsoft's side and was not tested, so automatic capture from Microsoft 365 is not claimed.
    12. Four eyes on an agent's email

      Layer Tier: Documented on a product page Status: Off by default (register EDR-02) Tools: Outlook, Copilot Proposed by AI agents, not yet reviewed

      By default, the person who registered an agent cannot approve that agent's held email.

      Source
      Product page /email-decision-record-product.html, at commit 90b235706c: the person who registered an agent cannot approve that agent's held email.
      Why this tier
      The quote is on /email-decision-record-product.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Release an agent's held email while signed in as the person who registered the agent, under a default policy: if it goes, the claim is wrong.
      Audit note
      Built; turned off until a Village adopts it. A sole owner can turn the rule off when adopting a policy, and a policy can require two approvers.
    13. Sending through a Microsoft account with send-only permission

      Layer Tier: Verified in code Status: Off by default (register EDR-05) Tools: Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      To send a released email through a connected Microsoft account, the Email Decision Record asks Microsoft for permission to send mail, not to read the mailbox.

      Source
      Code: src/services/emailDecisionRecord/mailAdapters/microsoftGraph.js, line 61, at commit 90b235706c: const REQUESTED_SCOPE = 'Mail.Send offline_access openid email';
      Why this tier
      The quote is on line 61 of src/services/emailDecisionRecord/mailAdapters/microsoftGraph.js at commit 90b235706c, checked by the validator against that commit.
      What would show it wrong
      Connect a Microsoft account and read the permissions Microsoft shows for it: if reading mail is among them, the claim is wrong.
      Audit note
      Built; turned off until a Village adopts it. The connection is set up with the organisation on request; whether one exists on either host was not read.
    14. Outlook add-in: seal a message and link its reply

      Layer Tier: Verified in code Status: Off by default Tools: Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      Built and turned off until a Village adopts it. A person opens a message in Outlook and seals it: Outlook digests the message file and each attachment, and the Village signs digests and pseudonyms, not the message. A later reply is linked to the sealed message it answers.

      Source
      Code: src/services/emailDecisionRecord/outlookAddin.js, line 320, at commit 36283f026e: const answered = await findAnswered(tenantId, record, receipt.receipt_id);
      Why this tier
      The quote is on line 320 of src/services/emailDecisionRecord/outlookAddin.js at commit 36283f026e, checked by the validator against that commit.
      What would show it wrong
      Seal a message, then seal a reply to it in the same Village: if the reply's seal does not name the first record, or the Village holds the message body, the claim is wrong.
      Audit note
      Built, turned off, not yet tried in a real Microsoft 365 account: EDR_OUTLOOK_ADDIN_ENABLED is not set. Tested in tests/integration/emailDecisionRecordOutlookAddin.test.js. It seals mail a person opens; it does not gate mail written in Outlook and does not see what Copilot does. No claims register entry of its own yet.
    15. Teams tab: seal a copy of pasted text

      Layer Tier: Verified in code Status: Off by default Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      Built and turned off until a Village adopts it. In the Teams tab a person pastes text and the tab builds a sealed copy with a random nonce. Only the copy's digest and size, and a digest of the chat identifier, reach the Village, which signs a pasted-text seal.

      Source
      Code: src/services/emailDecisionRecord/teamsConnector.js, line 55, at commit 36283f026e: const SEAL_KEYS = 'contentSha256,contentSizeBytes,context,conversationSha256';
      Why this tier
      The quote is on line 55 of src/services/emailDecisionRecord/teamsConnector.js at commit 36283f026e, checked by the validator against that commit.
      What would show it wrong
      Seal pasted text in the Teams tab and read what the Village stored: if it holds the text, or the saved copy's SHA-256 differs from the signed record, the claim is wrong.
      Audit note
      Built, turned off, not yet tried in a real Microsoft 365 account: EDR_TEAMS_CONNECTOR_ENABLED is not set, and the signing host has not yet been updated for this record kind. Tested in tests/integration/emailDecisionRecordTeamsConnector.test.js. No claims register entry of its own yet.
    16. Seal a file: a document version sealed by its digest

      Layer Tier: Verified in code Status: Off by default Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      Built and turned off until a Village adopts it. A person picks a file downloaded from SharePoint or OneDrive; the browser works out its SHA-256 and only the digest and size reach the Village, which signs a document seal. It seals the copy chosen, not the original in Microsoft 365.

      Source
      Code: src/services/emailDecisionRecord/documentSeal.js, line 37, at commit 36283f026e: const WRITTEN_SOURCES = Object.freeze(['console_file', 'teams_file']);
      Why this tier
      The quote is on line 37 of src/services/emailDecisionRecord/documentSeal.js at commit 36283f026e, checked by the validator against that commit.
      What would show it wrong
      Seal a file, then work out the SHA-256 of another copy of the same version: if it does not match the signed record, or the Village holds the file or its name, the claim is wrong.
      Audit note
      Built, turned off, not yet tried in a real Microsoft 365 account: EDR_DOCUMENT_SEAL_ENABLED is not set, and the signing host has not yet been updated for this record kind. Tested in tests/integration/emailDecisionRecordDocumentSeal.test.js. No claims register entry of its own yet. Folder watching through Microsoft Graph: not built, by design.
    17. The Village

      Layer Tier: Stated by us Status: Built (register VIL-01) Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      A members-only platform, built and operated by MDSL in New Zealand, in which these layers sit; each Village is held separately in New Zealand or the EU.

      Source
      Claims register entry VIL-01: Village is a members-only community and organisation platform built and operated by MDSL in New Zealand
      Why this tier
      The quote is in our claims register, entry VIL-01: our statement, not an outside check.
      What would show it wrong
      Find a Village whose records are readable from another Village: the separation claim would then be wrong.
      Audit note
      Register status live. It does not replace everything Microsoft 365 does: there is no hosted mailbox and no forms builder.
    18. Signing keys on a separate host, held by MDSL

      Layer Tier: Stated by us Status: Built (register EST-16) Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      A Village's signing key sits on a separate signing host, apart from the process that writes records, once MDSL has moved it there. MDSL operates that host and holds the keys; the organisation does not.

      Source
      Claims register entry EST-16: Signing keys sit on a separate signing host, outside the process that writes records
      Why this tier
      The quote is in our claims register, entry EST-16: our statement, not an outside check.
      What would show it wrong
      Find a Village record signed by a key the organisation itself holds: the custody statement would then be wrong.
      Audit note
      Register status built. In the code a Village signs with a local key on MDSL's servers unless MDSL has moved it to the signing host; this map does not say which Villages have been moved. Either way MDSL holds the key, so a verified record is not protected against rewriting by MDSL.
    19. A second, post-quantum signature

      Layer Tier: Stated by us Status: Built (register EST-37) Tools: Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      A second signature, ML-DSA-65, is added after the classical one on Provenance seals, Governance API receipts, Email Decision Record receipts and compaction summaries.

      Source
      Claims register entry EST-37: A second signature using ML-DSA-65 (NIST FIPS 204) is added to Provenance seals, Governance API receipts, Email Decision Record receipts
      Why this tier
      The quote is in our claims register, entry EST-37: our statement, not an outside check.
      What would show it wrong
      Check a recent Email Decision Record receipt from a Village where the signature is turned on: if it carries one signature only, the claim is wrong.
      Audit note
      Register status built, turned on per account. It does not cover Gateway receipts, invoices or other Village writes, and it is a hedge, not a statement that records are safe against a quantum computer.
    20. Held in New Zealand or the EU, by providers that are not US-owned

      Layer Tier: Stated by us Status: Built (register VIL-02) Tools: Microsoft 365, Outlook Proposed by AI agents, not yet reviewed

      Each Village's data is held in New Zealand, in Wellington, or in the EU, in Strasbourg.

      Source
      Claims register entry VIL-02: Each Village's data is held in New Zealand (Wellington) or the EU (Strasbourg)
      Why this tier
      The quote is in our claims register, entry VIL-02: our statement, not an outside check.
      What would show it wrong
      Find a Village record stored with a US-owned provider: the claim would then be wrong.
      Audit note
      Register status live; the register's EST-07 states neither host is US-owned. It covers where records are stored, not every dependency.
    21. Export a Village's shared records

      Layer Tier: Stated by us Status: Built (register VIL-23) Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      A Village owner or moderator can export the Village's shared content records, one record type at a time, through its programming interface.

      Source
      Claims register entry VIL-23: A Village owner or moderator can export the Village's shared content records, one record type at a time
      Why this tier
      The quote is in our claims register, entry VIL-23: our statement, not an outside check.
      What would show it wrong
      Ask for an export of one record type as a Village owner: if it is refused for a Village whose records all carry a provenance hash, the claim is wrong.
      Audit note
      Register status live. There is no screen for it, and a record type is refused whole if any record lacks a provenance hash.
    22. Attested Accounts

      Layer Tier: Stated by us Status: Available by arrangement (register ACC-01) Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      An evidence layer over the Business Village's invoicing, quotes, recurring invoices, expenses and time tracking: each write is signed as it is made.

      Source
      Claims register entry ACC-01: Attested Accounts is an evidence layer over the Business Village's invoicing, quoting, recurring invoicing, expense and time tracking.
      Why this tier
      The quote is in our claims register, entry ACC-01: our statement, not an outside check.
      What would show it wrong
      Find a write to an invoice that left no signed entry: the claim would then be wrong.
      Audit note
      Register status available on request. It is not an accounting system and does not say whether figures are right.
    23. Provenance

      Layer Tier: Documented on a product page Status: Available by arrangement (register PRV-02) Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      Each party in an export chain signs its own records of a lot and passes them on with the lot, so the last holder's file shows every party back to the first.

      Source
      Product page /provenance-product.html, at commit 90b235706c: Each party in an export chain, from grower to processor, exporter and buyer, signs its own records of a lot
      Why this tier
      The quote is on /provenance-product.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Find a Provenance dossier in which a substituted record passes the check: the claim would then be wrong.
      Audit note
      Register status available on request. It is about goods in a supply chain, not about mail or files.
  4. Evidence: how you can check (9)

    1. The Gateway checking program

      Evidence Tier: Documented on a product page Tools: Copilot Proposed by AI agents, not yet reviewed

      An exported Gateway decision can be checked by whoever holds it, with the Village's trust bundle and Node 20, with no account and no network.

      Source
      Product page /gateway.html, at commit 90b235706c: Anyone you give the record to can check it with the checking program, using the trust bundle the Village supplies and Node 20 or newer. It needs no account and no network.
      Why this tier
      The quote is on /gateway.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Run the checking program on an export with one byte changed: if it reports verified, the claim is wrong.
      Audit note
      The program is served at mysovereignty.digital/downloads/ with its SHA-256. Run on the synthetic sample it exits 1 by design, because the sample's receipt has no stamp.
    2. Each checking program published with its SHA-256

      Evidence Tier: Stated by us Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      Each offline checking program is published with its SHA-256, so a reader can confirm the file they run is the file MDSL published.

      Source
      Claims register entry EST-38: Each offline checking program is published at mysovereignty.digital/downloads/ with its SHA-256
      Why this tier
      The quote is in our claims register, entry EST-38: our statement, not an outside check.
      What would show it wrong
      Download a checking program and its .sha256 file and compare: a mismatch would show the claim wrong.
      Audit note
      Register status built. Publication for reading and checking is not an offer of any product.
    3. The Conduct Record offline check

      Evidence Tier: Documented on a product page Tools: Copilot Proposed by AI agents, not yet reviewed

      An auditor can check a Conduct Record offline, without an MDSL account.

      Source
      Product page /mio-conduct-record.html, at commit 90b235706c: An auditor can check a Conduct Record offline, without an MDSL account.
      Why this tier
      The quote is on /mio-conduct-record.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Check a Conduct Record with the network off and no account: if the check cannot run, the claim is wrong.
      Audit note
      The page says the order in which mandates were recorded is not yet tamper-evident; the lookup says so with each answer.
    4. The Email Decision Record checker

      Evidence Tier: Documented on a product page Tools: Outlook, Microsoft 365 Proposed by AI agents, not yet reviewed

      A standalone checker verifies an Email Decision Record in a browser or from a file, with no account and no call to MDSL.

      Source
      Product page /email-decision-record-product.html, at commit 90b235706c: A standalone checker verifies a record in a browser or from a file, with no account and no call to us.
      Why this tier
      The quote is on /email-decision-record-product.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Remove a receipt from inside a chain in a dossier and run the checker: if it does not report a gap, the claim is wrong.
      Audit note
      The checker returns a plain result for each record and says what it could not check.
    5. Check one suspicious email

      Evidence Tier: Stated by us Tools: Outlook Proposed by AI agents, not yet reviewed

      An email saved from Outlook as .eml can be uploaded for nine checks on the message itself, with no account.

      Source
      The guide, Stepping outside the ring: Save it from Outlook as .eml and upload it at mysovereignty.digital/scam-check.html. No account is needed. Nine checks run on the message itself
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Upload a message with a lookalike sender domain: if no check flags it, the claim is wrong.
      Audit note
      A tenth check, link reputation, is shown as not checked. It looks at one message; it does not watch a mailbox.
    6. A check never reads as verified without a pinned key

      Evidence Tier: Documented on a product page Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      Without a pinned key, the checker reports a record as valid under a key the record supplied, never as verified, because a record with swapped keys would otherwise pass.

      Source
      Product page /evidence.html, at commit 90b235706c: It never reads as verified, because a record whose keys had been swapped would otherwise pass.
      Why this tier
      The quote is on /evidence.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Run a checker on a record with no key pinned: if it reports verified, the claim is wrong.
      Audit note
      Pinning a key confirms which key signed; MDSL holds the signing keys today.
    7. What a check cannot show

      Evidence Tier: Stated by us Tools: Microsoft 365, Outlook, Copilot Proposed by AI agents, not yet reviewed

      A check shows that a record is unchanged since it was signed, and when. The signature says nothing about whether the content was true.

      Source
      The guide, Stepping outside the ring: The signature says nothing about whether the content was true.
      Why this tier
      The quote is in our guide, Stepping outside the ring: our statement, not an outside check.
      What would show it wrong
      Find a page of ours that presents a passed check as proof that a record's content is true: that page would contradict this node.
      Audit note
      The limit that applies to every checker on this map.
    8. An invoice's offline verification bundle

      Evidence Tier: Documented on a product page Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      An invoice has an action-history view and a downloadable bundle that a third party can check offline, without a Village account.

      Source
      Product page /attested-accounts.html, at commit 90b235706c: An invoice has an action-history view and a downloadable verification bundle that a third party can check offline, without a Village account.
      Why this tier
      The quote is on /attested-accounts.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Change one figure in an invoice bundle and check it offline: if it passes, the claim is wrong.
      Audit note
      Invoices only: quotes, expenses and time entries do not yet have a history view or bundle.
    9. The Provenance check in a browser

      Evidence Tier: Documented on a product page Tools: Microsoft 365 Proposed by AI agents, not yet reviewed

      A buyer checks every party's records in a browser, on its own machine, with no account and no connection to MDSL.

      Source
      Product page /provenance-product.html, at commit 90b235706c: A buyer checks every party's records in a browser, on its own machine, with no account and no connection to us.
      Why this tier
      The quote is on /provenance-product.html at commit 90b235706c, a page whose claims have been corrected against the register and the code.
      What would show it wrong
      Open a dossier with the network off: if the check needs a connection, the claim is wrong.
      Audit note
      It does not show that what a party recorded is true.

The map by view

By tier

Verified in code (6)

Documented on a product page (22)

Stated by us (21)

Open: no source (2)

By layer

Village Gateway (18)

MIO° Conduct Record (7)

Email Decision Record (15)

The Village (19)

Attested Accounts (5)

Provenance (4)

By threat

AI agents that overstep (13)

An email sent by an agent nobody approved (6)

Messages that sound like someone you know (8)

Confident AI mistakes (4)

Unfair automated decisions (3)

A record changed after the event (16)

Your provider is also the only witness (6)

Foreign law (4)

A provider you cannot leave (4)

Locks that will not hold forever (5)

Leaked passwords and member details (1)

Pictures you cannot vouch for (1)

By tool

Microsoft 365 (38)

Copilot (29)

Outlook (26)

Open claims only

Gaps: not built, or no answer on this map (5)

Claims with a limit

Claims on a link that says what it does not cover (37)

Recently changed

Changed in the latest version (10 November 2026) (51)