Heartbeat Framework

Home›Papers

The papers

Twenty-two open-access records, numbered in the order they were deposited.

Each has a DOI on Zenodo, a PDF, and, where it supplies an instrument, its stage in the sequence. Records 21 and 22 are the implementation study Learning Outside the Model and its companion note; what was built and measured is set out under Implementation.

All twenty-two records

Cite the DOI, not this page. All records: Paul Blatherwick (2026), CC BY 4.0.

  1. 01
    The 24-Hour Team
    Core paper 1 of 3. Managing round-the-clock care as two teams joined by a handover produces a family of predictable failures; the unit is one organism, and its vitals can be read.
    2026-07-07 · Record 1 · doi:10.5281/zenodo.21243118
  2. 02
    The Mixed Shift
    Core paper 2 of 3. What survives, transforms and breaks when some members of the 24-hour team do not sleep, gossip or die. Sets out the five standing rules.
    2026-07-07 · Record 2 · doi:10.5281/zenodo.21245130
  3. 03
    The Long Handover
    Core paper 3 of 3. Turns the framework's own ethic on itself: five confessions the earlier papers omit, the evidence tier, and the uncosted bandwidth of the managers asked to run it.
    2026-07-07 · Record 3 · doi:10.5281/zenodo.21245205
  4. 04
    The Character Pathway
    Stage 3 · SpecifyApplied series 1. Organisations govern budgets, data and uptime and leave ungoverned the variable that decides adoption: the system's character. The Character Specification and Character Vitals.
    2026-07-07 · Record 4 · doi:10.5281/zenodo.21249771
  5. 05
    Same Monsters, New Casings
    Applied series 2. Six recurring patterns of difficult behaviour in human teams and six in deployed machines, mapped onto each other with the same responses.
    2026-07-08 · Record 5 · doi:10.5281/zenodo.21260388
  6. 06
    The Endless Shift
    Machine temperament as a clinical safety property: the public has converged, unprompted, on stable personality descriptions of the major models, and that convergence is data.
    2026-07-08 · Record 6 · doi:10.5281/zenodo.21264935
  7. 07
    The Overseer's Bible
    Stage 4 · OverseeApplied series 3. The seat-holder's practice: accept or refuse the oversight role on four tests — authority, information, time and retained skill — with the worksheet, shift card and templates.
    2026-07-09 · Record 7 · doi:10.5281/zenodo.21279944
  8. 08
    The Second Signature
    Self-written records cannot testify. The double-trust architecture that lets a machine's record be checked by someone who does not trust the machine.
    2026-07-09 · Record 8 · doi:10.5281/zenodo.21282804
  9. 09
    The Second Pair Of Hands
    Single-form omnicompetence is impossible in principle: Ashby's law of requisite variety applied to the frontier training objective, and what a second pair of hands is for.
    2026-07-13 · Record 9 · doi:10.5281/zenodo.21340572
  10. 10
    The AI Interaction Risk Screen
    Stage 2 · ScreenStanding instrument. A practitioner-facing screen for how a person's AI use is functioning in their life, on the axis that matters: displacement or supplementation. Version 1.1 after structured review.
    2026-07-14 · Record 10 · doi:10.5281/zenodo.21357450
  11. 11
    The Cultural Safety Case: Why a licence to deploy is a licence to amplify
    Every regime assesses the capability of the system; none assesses the culture of the organisation deploying it. Deployment risk is their product, and four organisational predictions are registered.
    2026-07-19 · Record 11 · doi:10.5281/zenodo.21443120
  12. 12
    Before the Machine
    Stage 1 · GateApplied series 0, the paper numbered zero because it comes first. Every other document says how to deploy well; this one says when not to deploy at all. The Readiness Gate.
    2026-07-19 · Record 12 · doi:10.5281/zenodo.21445040
  13. 13
    The Floor Test
    Stage 5 · TrialApplied series 4. A 90-day field protocol for testing the framework on a real unit: baseline the vitals, change one thing, read the result — and send it back, especially if it is null.
    2026-07-19 · Record 13 · doi:10.5281/zenodo.21445071
  14. 14
    The Damping Audit
    Stage 6 · AuditApplied series 5. A ward-level containment instrument: maps and measures the discretion layer where staff absorb, translate, correct, sense and escalate, and what a deployment does to it.
    2026-07-20 · Record 14 · doi:10.5281/zenodo.21455540
  15. 15
    The Glass Ledger
    Stage 8 · RecordStanding instrument. A tamper-evident decision journal for AI agents: about 600 lines of Python, MIT-licensed, shipping with a demo that attacks its own ledger four ways. Version 2.
    2026-07-23 · Record 15 · doi:10.5281/zenodo.21515861
  16. 16
    After the Machine
    Stage 7 · MeasureApplied series 6. The AI Interaction Outcome Profile: every control instruments the technology, none instruments the person. Six bipolar domains, read against the person's own baseline; three predictions registered.
    2026-07-23 · Record 16 · doi:10.5281/zenodo.21515932
  17. 17
    The Heartbeat Framework: Reading Map
    Estate document. The index to the estate at twenty-one records: three core papers, seven applied instruments, six extended works, two standing instruments, the instrument deck, two implementation papers, and the order to read them in.
    2026-09-27 · Record 17 · doi:10.5281/zenodo.22994740
  18. 18
    The Instrument Deck
    Estate document. Every worksheet, checklist and instruction sheet from the published estate, lifted intact and bound in the order a deployment meets them: eight instruments, one sequence, 38 pages.
    2026-08-21 · Record 18 · doi:10.5281/zenodo.22045477
  19. 19
    The Tenant's Record: Who the system serves when the system is not yours
    Who the system serves when the system is not yours. Adult social care went from paper to platform in four years; the record of a person's care is now held by a tenant of somebody else's software.
    2026-08-29 · Record 19 · doi:10.5281/zenodo.22164015
  20. 20
    What the Signature Certifies: Four asks for the revision of DCB0129 and DCB0160
    NHS England is revising DCB0129 and DCB0160 for the first time since 2018. Four asks for that revision, accepting the review's own findings on AI governance, monitoring and compliance.
    2026-09-15 · Record 20 · doi:10.5281/zenodo.22771015
  21. 21
    Learning Outside the Model
    A frozen-weight personal agent in which every learning mechanism held outside the weights is paired with a check the model does not perform on itself, with fifty-one safety claims turned into tests that gate the service.
    2026-09-27 · Record 21 · doi:10.5281/zenodo.22994585
  22. 22
    When the Weights Can Move
    Companion note to Learning Outside the Model: what the same governance becomes when the planner can learn in its weights, and where the boundary sits.
    2026-09-27 · Record 22 · doi:10.5281/zenodo.22994587