CTU metering method
What is metered, at which boundary, in what record, how it is checked and how it is reported.
A draft (0.x): read it, build to it and comment on it; it may change without notice. Drawn from the CTU book and the Sovereign Stack page. Changelog · Comment on this specification
Relies on: Correspondence proof format, Ledger entry format · Relied on by: Audit method
This is a draft (0.x). It may change without notice, and no deployment is metered to it yet. Every number on this page is an example, not a measurement.
1 Purpose
Every CRIS system reports the compute and energy it uses in Compute Thermodynamic Units (CTU), so that gains in capability can be weighed against their cost. This is design principle 5: measured cost. This specification sets out what is metered, how, in what record, how a record is checked, and how it is reported. It does not yet define one CTU; Continuity Labs will state that definition, from the CTU book, in a later version. Until then, records keep the raw energy and compute figures, and CTU is left empty.
No CTU definition yet. Records keep the raw energy, compute and cost; the Conversion field reads none and the CTU field stays empty. Continuity Labs will state the definition, from the CTU book, in version 0.3.
2 What CTU is, and is not
CTU is a unit of measurement. In this specification it measures the compute and energy a cycle used. It is not money, a token or an asset, and it cannot be bought, sold, held or exchanged. CTU is not used in any reward or payment calculation; if that ever changes, the method will be published and approved first. On the CES Platform, Participation record (CTU) is a separate record of participation. The one list of what CTU is not, for both senses, is on the CTU page.
The four rules
Kept from version 0.1, with rule 4 reworded (Q2).
- Every cycle is metered, including cycles whose change fails.
- Capability is reported beside its energy and cost, never alone.
- No figure is published until it is measured. This draft contains no figures.
- In a sovereign deployment, each partner sees the compute and energy its own deployment used.
The method in pictures
Sixteen pictures set out the method: what changed, who reads it, how it measures, how a record is checked and reported, what CTU is not, and how the specification reaches 1.0.
What changed. Draft 0.1’s six sections, and what this version did with each finding.
Who reads it. Six readers, and what each now gets.
Raw quantities first. CTU will be derived from measured energy, compute and cost under a versioned conversion.
The boundary. What is inside each tier.
A record’s life. From opening to the ledger.
Verification. Five checks before a record counts.
Reporting. Three roll-ups.
Corrections and disputes. History is never rewritten.
Threats and guards. How a figure is kept honest.
The money line. CTU is a measure, not money.
Conformance. Three levels and eight test vectors.
The road to 1.0.
How this specification is built. The section map, the data files, the page and the build steps.
3 The boundary
Every figure names the boundary it was measured at.
| Tier | Inside | How measured | Required in 0.2 |
|---|---|---|---|
| T1 Devices | Accelerators, CPUs, memory and storage the cycle used | Device energy counters, or a power meter on the node, sampled over the cycle | Yes |
| T2 Facility | T1, plus cooling, power conversion and building overhead | A facility meter apportioned to the cycle, or T1 × a stated overhead factor with its source | When the site can meter it or states a factor |
| T3 Network | T2, plus data moved between sites | Bytes moved × a stated energy rate | No |
From data. Every figure names its tier; figures at different tiers are never added or compared.
4 What is measured
For each cycle: energy in joules at the stated tier, with the meter type and its stated error; compute in device-seconds, by device model; cost in local currency, with its basis; and the cycle’s signed start and end times.
| Quantity | Unit | Source | Accuracy to state |
|---|---|---|---|
| Energy | joules (reported as kWh) | Device counters or node meter (T1); facility meter or factor (T2) | Meter type and stated error |
| Compute | device-seconds, by device model | Scheduler records | Clock source |
| Cost | local currency, with the basis (tariff, amortisation) | The partner's tariff or contract | Basis and date |
| Duration | seconds, start and end | Signed timestamps | Clock source |
A cycle is one pass of the Continuity Recursion Loop: a proposal, its proof and its gate decision. Every cycle gets exactly one metering record, whatever its outcome. Work that is not part of a cycle (idle capacity, maintenance, training of staff) is reported separately and never folded into a cycle’s figure.
Where a cycle shares hardware, energy is allocated by the device-seconds it held, and the rule is recorded. On testnet nodes, such as those of the Botswana Global Testnet Initiative under a written agreement, the node operator’s meter is used and named; cycles on testnet nodes are labelled testnet and kept apart from deployment figures.
The Botswana Global Testnet Initiative. An organisation also led by our Founder. Any work with CRIS is under a separate written agreement on arm’s-length terms. Related organisations
Correspondence gain per CTU. The ratio needs a numerator. The proposal is that the correspondence proof states the cycle’s measured correspondence gain, and the metering record refers to that proof. Until the proof format defines the gain, the ratio is shown only as a form, gain (from the proof) ÷ CTU (from this record), with no example value (Q4).
5 The record
| Field | Required | Meaning | Example (illustrative) |
|---|---|---|---|
| Record ID | Yes | Unique and permanent | example: mr-0001 |
| Cycle ID | Yes | The cycle metered, as on the ledger | example: cy-0042 |
| Outcome | Yes | The gate decision: pass, staged, reject, or failed before the gate | example: reject |
| Deployment | Yes | Deployment or testnet, and which; never a person | example: testnet |
| Boundary tier | Yes | T1, T2 or T3 | example: T1 |
| Start, end | Yes | Signed timestamps | example: 2027-01-01T10:00:00Z / 10:04:10Z |
| Energy | Yes | At the stated tier, with meter type and stated error | example: 9000000 J ± 2% (node meter) |
| Compute | Yes | Device-seconds by device model | example: 1000 device-s, model A |
| Cost | Yes | Amount, currency and basis | example: 12.50 in local currency, tariff |
| Allocation rule | If any | How shared hardware was split, if any | example: by device-seconds |
| Conversion | Yes | The CTU definition version used, or none | example: none |
| CTU | If any | Derived value; empty while no definition exists | example: empty |
| Proof reference | Yes | The correspondence proof for the cycle | example: proof-0042 |
| Corrects | If any | The record this one corrects, if any | example: empty |
| Signature | Yes | The metering agent that wrote it | example: agent-07 signature |
Draft 0.2. Field names may change. Every value in the last column is an example, not a measurement.
Rules: one record per cycle, whatever its outcome, including cycles that fail; a signed record is never edited; a correction is a new record that points to the one it corrects, as the ledger entry format requires; a record names a deployment, never a person; raw values are kept after a CTU figure is derived.
6 Verification
Before it counts toward any report, every signed record must pass five checks. A record that fails any check is marked unverified and excluded.
| # | Check | Fails when | Result of a fail |
|---|---|---|---|
| 1 | Signature | The signature is invalid, or the signer is not a registered metering agent | Unverified; incident |
| 2 | Cycle | No ledger entry, proof or gate decision exists for the cycle | Unverified |
| 3 | Time | The record's times fall outside the cycle, or two records overlap on one device | Unverified |
| 4 | Plausibility | Energy is outside the published range for the devices and duration | Unverified; review |
| 5 | Allocation | Allocated energy across records on a node does not sum to the node meter within the stated error | Unverified for every record on that node and period |
The metering agent signs each record with a registered key that can be revoked. A partner can run the same checks on its own records with the Continuity Audit Suite, without CRIS. An auditor samples records under the audit method, re-reads raw meter logs and reports the share that matched. A figure may be called verified only when every record behind it passed, and audited only after an audit has covered it.
7 Reporting
| Roll-up | Who sees it | Contains | Never contains |
|---|---|---|---|
| Partner report | The partner, for its own deployment | Energy, compute, cost and CTU per cycle and per month, by tier; failed cycles; idle time apart | Other partners' figures |
| CRIS totals | CRIS; the Board; auditors | All deployments, by tier; testnet kept separate | Unverified records |
| Benefit indicator | The public, in the benefit report | One figure per reporting period, with scope and tier | Any figure not yet measured |
The Knowledge Ledger’s entry for each cycle will carry its metering record’s fingerprint as an evidence reference, so any later change is detectable (Q6).
- Energy per unit of capability gain, in CTU: lower is better; used on this specification (0.1) and the Sovereign Stack page (recommended: it is the one gain per CTU supports).
- Useful compute per unit of energy, in Compute Thermodynamic Units: higher is better; used on data/benefit.json B2.
The second sovereignty indicator, deployments running on local infrastructure with local data, is a count. It is reported beside CTU and is not measured in it. No figure is published until it is measured.
8 Corrections and disputes
| Error | Example (illustrative) | Fix |
|---|---|---|
| Meter fault | A node meter stops reporting for an hour | Records for that hour marked unverified; an estimate is never substituted silently |
| Calibration | A meter reads 3% high against a reference | Correction records for the affected period, with the factor and its source |
| Allocation | Shared hardware split by the wrong rule | Correction records for every cycle on that node and period |
| Clock | Timestamps out of order | Unverified until re-signed with a corrected time source |
| Conversion | A new CTU definition version | Not an error: CTU recomputed under the new version; both shown |
Anyone with a stake in a figure (a partner, an auditor or CRIS) can raise a dispute. The records concerned are flagged, investigated against the raw logs and meter calibration, and either corrected by new records that point back or confirmed with a written reason. A published figure that changes is reissued with a note. An estimate is always marked as one, with its method, and never called measured. Appeals go to someone outside Systems Engineering, to be named (Q8).
9 Security and privacy
Records are signed by registered keys and timed from a stated, synchronised source. A revoked key’s later records are rejected. Raw meter logs are kept for [a period to confirm] so they can be re-read. In a sovereign deployment, records and logs stay on the partner’s infrastructure and only agreed roll-ups leave. Records hold no personal data: they describe machines and cycles. Published figures are totals by tier and scope; a deployment small enough to identify a partner is grouped with others or withheld.
10 Conformance
A meter conforms at a level when it gives the expected result for every test vector at that level.
| Level | A meter at this level | Needed for |
|---|---|---|
| 1 Record | Emits one complete record per cycle, with every required field and unit, including failed cycles | Testnet cycles |
| 2 Verified | Level 1, plus signed records that pass the five checks, with T1 measured | Partner reports |
| 3 Facility | Level 2, plus T2 metered rather than estimated, and raw logs kept for audit | Benefit-report figures |
11 Open questions
- The definition of one CTU (Continuity Labs, version 0.3).
- How the correspondence gain in gain per CTU is stated in the proof format (Q4).
- Which benefit indicator, in which direction (Q3).
- Who hears appeals (Q8), and how long raw logs are kept (Q7).
| Version | Adds | Gate to the next | When |
|---|---|---|---|
| 0.1.0 | Purpose, what is metered, four rules, reporting, two open questions | — | Published 11 Nov 2026 |
| 0.2.0 | Boundary tiers, record fields, lifecycle, checks, the money line, disputes, test vectors | Systems Engineering review; Founder approves the money line | This draft |
| 0.3.0 | The definition of one CTU and the first conversion, from Continuity Labs | Continuity Labs and Systems Engineering both sign | Q1 2027 (proposed) |
| 0.4.0 | Testnet cycles metered at Level 2; outside review of the method | Review published | 2027 (proposed) |
| 1.0.0 | Licence confirmed; a deployment metered at Level 3 | Sovereign Stack status set with evidence | After a metered run |
12 Changelog
- 0.2.0 · 12 November 2026 · minor · Adds the boundary tiers, the record, verification, reporting, corrections, security, conformance and test vectors; states what CTU is and is not; rewords rule 4 from bought to used. Awaiting Systems Engineering’s review and the Founder’s approval of the money line.
- 0.1.0 · 11 November 2026 · minor · First draft, from the CTU book and the Sovereign Stack page.
Printed from the developer hub changelog, the one record of every version.
