# Use Member
> Per-class consume positions, commit-% listen, edge workload mapping, self-dispatch and telemetry MUSTS, ramp limits, and firm-limit fallback.
- Repository: g2tf-org/g2tf-standards
- GitHub: https://github.com/g2tf-org/g2tf-standards
- Human docs: https://grok-wiki.com/public/docs/g2tf-org-g2tf-standards-436ba3f3e0ff
- Complete Markdown: https://grok-wiki.com/public/docs/g2tf-org-g2tf-standards-436ba3f3e0ff/llms-full.txt
## Source Files
- `members/use.md`
- `specs/g2p/spec.md`
- `specs/desp/spec.md`
- `architecture/failsafe-model.md`
- `specs/gcap/spec.md`
---
---
title: "Use Member"
description: "Per-class consume positions, commit-% listen, edge workload mapping, self-dispatch and telemetry MUSTS, ramp limits, and firm-limit fallback."
---
A **Use Member** is the Grid 2 Member Element for a flexible large load. The living behavioral contract is `members/use.md` (**Status:** Draft — seeking input). Each one-minute interval the Member broadcasts a signed per-class **request to consume** (`+`) on G2P, listens for its Allocator’s per-class **commit-%**, self-dispatches within that commit-% and host-utility ramp limits, and reports **verified self-dispatch**. A missing or stale Commitment forces revert to the Grid 1 **firm load limit** or manual curtailment.
`members/use.md`, G2P **26.0-draft**, DESP **26.0-draft**, and GCAP **26.0-draft** are living Draft files seeking input. RFC 2119 / RFC 8174 keywords in those files are normative. G2P does not yet specify wire format, encoding, per-element state machines, or conformance test vectors.
## Status and stack position
| Surface | Role for Use |
|---|---|
| `members/use.md` | Behavioral MUST/MAY contract for the load Member |
| G2P §3.1 / §3.2 / §3.4 | Service Descriptor, Commitment, Response / Telemetry |
| DESP | Classes **A / B / C**, dynamic reclassification, C→B→A curtailment |
| GCAP | Deterministic per-class **commit-%**; headroom, then Buffers, then Use curtailment |
| Failsafe model | Firm-limit / manual-curtailment fallback when context is stale |
Use is one of three Member kinds (Use, Buffer, Source). Membership is spec compliance at a point of interconnection covered by a provisioned Allocation Domain, under commercial terms with the host utility. The smallest domain is a host utility plus a single flexible load.
Every Use Member MUST hold a stable element identifier bound to that physical point of interconnection. Identifier format is open (G2P recommends a domain-scoped hierarchical ID). v26.0 positions are **energy within the one-minute interval** on the bulk power system.
## Associated physical asset
Flexible large load: capable datacenter sites (any mix of flex-capable load and/or on-site power resources) and electrified industry.
Internal facility assets — GPUs, cooling, behind-the-meter storage, process lines — stay behind the Member. The Allocator never sees or directs those workloads; it clears **class-level positions only**.
## Interval contract
Grid 2 allocations apply to whole one-minute intervals. Clearing MUST finish so the Member receives **commit-%** with time to self-dispatch **before the interval opens**. Exact descriptor / clearing / publication deadlines and clock-sync rules are open in G2P.
Where a Grid 1 dispatch instruction conflicts with a Grid 2 allocation, **the Grid 1 instruction prevails**. Sub-minute primary controls (governor, AGC, protection) stay out of scope for the orchestration plane.
```mermaid
sequenceDiagram
participant Use as Use Member
participant Alloc as Allocator
Note over Use,Alloc: One-minute interval (commit before open)
Use->>Alloc: Service Descriptor — per-class request to consume (+)
Note over Alloc: GCAP: headroom → Buffers → C then B then A
Alloc-->>Use: Commitment — per-class commit-%, interval, domain, inputs version
Use->>Use: Self-dispatch within commit-% and agreed ramps
Use->>Alloc: Response / Telemetry — actual per-class dispatch
alt Missing or stale Commitment
Use->>Use: Fallback — Grid 1 firm load limit or manual curtailment
end
```
### Broadcasts (G2P Service Descriptor)
One `+` consume position per service class over a forward time vector. Each position MUST be signed by the originating Member. Positions MUST be energy in the interval (v26.0).
DESP **Assured (A)**, **Preferred (B)**, **Best Efforts (C)**. G2P interaction-table names Firm / Semi-Firm / Flex map to A / B / C; naming reconciliation is an open item.
Descriptor covers a forward vector. Length, resolution, and revision rules are unspecified.
A Service Descriptor MAY be reissued each interval. Positions MAY be reclassified each interval (DESP dynamic classification). All Service Descriptors MUST be authenticated; integrity MUST be verifiable end-to-end.
### Listens (G2P Commitment)
Allocator output to each Use Member for the **coming interval only**. Allocations expire with the interval.
Interval the Commitment applies to. MUST be present.
Allocation Domain the Commitment was computed in. MUST be present.
Input digest / version the Allocator cleared against. MUST be present. Allocators SHOULD publish the digest in-domain.
A Member MUST treat a **missing or stale** Commitment as fallback-triggering. A Member that has not received a commit-% for the **current** interval MUST treat itself as in fallback. Explicit staleness timers remain open in G2P.
### Reports (G2P Response / Telemetry)
Each Member MUST report **actual per-class dispatch against its Commitment** so compliance is auditable and the overlay’s aggregate footprint is measurable. Verification granularity, metering source of truth, and attestation are open.
## Edge workload mapping
The Use Member owns all operational decisions:
- How to map internal workloads onto DESP classes
- Which demand is deferrable, and when
- Anticipating its own forward load
This is the **scale without singular operator** rule: every asset runs the same algorithm against the same observable signals. The Allocator holds no internal intelligence or discretion.
DESP §3 mapping is **informative only** (Emerald AI Flex 0/1/2/3 and ERCOT CLR/PCLR constructs). Standard per-facility-type guidance is an open item on `members/use.md`.
| DESP class | Informative workload hint | Curtailment rank |
|---|---|---|
| Traditional Firm (Grid 1) | Outside Grid 2; never curtailed by GCAP | Not a Grid 2 class |
| **A** Assured | ≈ Flex 0 / LPC-aligned firm portion; latency-sensitive inference | Last among Grid 2 |
| **B** Preferred | ≈ Flex 1–2 / CLR-PCLR dispatchable; checkpointed training, deadline batch | After C, before A |
| **C** Best Efforts | ≈ Flex 3; opportunistic / spot compute | First |
A domain MAY constrain the mix of classes a Member can declare (for example a minimum Class C share). Those limits are **commercial-terms** in the domain agreement, not protocol fields.
When constraints bind, GCAP reduces Use allocations in DESP order **C → B → A**, never touching Grid 1 Firm. **Buffer committed dispatch clears before any Use class is curtailed.** Within a class, curtailment MUST be proportional (connection size and differential request size). The exact proportionality formula is open.
## Self-dispatch and telemetry MUSTS
Sign and send one consume position per class for the forward vector before the (still unspecified) descriptor deadline.
Accept only an authenticated Commitment that names this interval, this domain, and a clearing-inputs version. If it does not arrive, or is stale, enter fallback — do not invent a commit-%.
MUST self-dispatch **within** the per-class commit-% for the interval, within parameters agreed with the host utility. MUST respect agreed ramp rates on every transition (interval-to-interval, into fallback, out of fallback).
MUST report actual per-class dispatch against the Commitment (G2P §3.4). Grid 1 verifies the overlay by aggregate footprint, not by inspecting facility internals.
Do not treat unused requested-and-committed energy as automatically returned to the pool mid-interval. Whether undershoot recycles mid-interval is an explicit open item on `members/use.md`.
## Ramp limits
| Bound | Rule |
|---|---|
| Per-Member delta | GCAP: allocation deltas between consecutive intervals MUST respect that Member’s agreed ramp rates |
| All transitions | Failsafe model: into fallback, out of fallback, and between interval allocations MUST follow rates agreed with the host utility |
| Whole-pool fallback | Allocator MUST bound total cleared participation so simultaneous Member fallback stays inside manageable aggregate Grid 1 ramp limits |
Per-class snap-back / ramp envelopes (research cites 15-minute ramp discipline) are open in DESP. Quantitative domain-growth methodology is an open engineering question.
## Firm-limit fallback
On loss of Allocator, peers, or fresh context, or on a stale Commitment:
| Trigger | Required Use behavior |
|---|---|
| Missed / stale / invalid Commitment | MUST revert to Grid 1 baseline (**firm load limit**) **or** manual curtailment |
| Transition | MUST follow allowed ramp rates |
| Determinism | MUST be locally determinable from the Member’s own state and last-known configuration; MUST NOT depend on reaching a remote peer |
| Interval validity | Allocations are valid for their interval only |
Allocator-side complement: if constraint inputs are stale or the domain is partitioned, the Allocator MUST NOT clear; Members detecting a missed clearing fall back as above. Partition between peer Allocators MUST NOT trip Members in unaffected domains.
Open failsafe items that affect Use implementations: quantitative missed-interval thresholds, re-entry / hysteresis after fallback, coincidental-fallback stability, and audit/telemetry to verify fallback compliance.
v26.0 is energy consumed and provided inside the one-minute window. Transmission-level deployments are expected on private networks. Service Descriptors and Commitments MUST be authenticated. PKI, NERC CIP mapping, and replay protection are open (Discussion: Engineering).
Wire format · intra-interval timing budget · identifier format · forward-vector length · staleness timer values · minimum telemetry schema · unused-allocation recycling · standard facility-type class maps · class-count extensibility · per-class SLA envelopes.
## Open items (Use-owned)
Tracked on `members/use.md` and the linked protocol specs:
- Standard workload→class mapping guidance per facility type
- Minimum telemetry for commit-% compliance verification
- Behavior when actual load undershoots requested + committed amounts
- DESP naming (Assured/Preferred/Best-Efforts vs Firm/Semi-Firm/Flex)
- Enforcement of declared class behavior (ties to G2P §3.4)
## Next
Committed dispatch clears before any Use class is curtailed.
No-discretion clearing that produces per-class commit-%.
DESP A/B/C, C→B→A order, and per-interval reclassification.
Firm-limit fallback, interval-only commits, locally determinable ramps.
Descriptor, Commitment, and verified-self-dispatch message families.
Informative Flex 0–3 / CLR mappings; Allocator never sees internals.