Sizing the Local Journal for a Disconnected Biofoundry
A federated biofoundry must retain execution authority and assay evidence locally when its cloud connection disappears. The sizing constraint is the volume of unacknowledged telemetry accumulated before connectivity returns, together with the storage bandwidth needed to preserve that evidence while instruments continue operating.
Put the execution boundary beside the instrument
Consider a network partition that isolates a biofoundry site from its federation while an accepted assay continues running. The local controller must already hold the executable procedure, material identities, calibration references and permitted stopping boundaries, because a remote scheduler cannot provide timely decisions across the partition. Instrument interoperability standards such as OPC UA can carry typed observations and equipment state, but application logic must still establish which observations belong to each accepted physical action. Persist an action identifier and its intent before dispatch, then record the observed outcome without treating a missing cloud acknowledgement as permission to repeat dispensing. This adds journal traffic and requires explicit recovery states, but it prevents reconnection from converting transport retries into duplicate physical operations.
Carry the storage calculation through
Assume an aggregate journal ingress rate r of an estimated 8 MB/s, measured after encoding and including instrument records, assay attachments and execution metadata. For an isolation allowance T of an estimated 6 hours, the payload requirement is B = r × T, giving an estimated 173 GB in decimal storage units after time conversion and rounding. This calculation assumes sustained ingress throughout the outage, an initially cleared acknowledgement backlog and no deletion of records that lack durable remote acknowledgement. It sizes accumulated payload rather than the installed device, so filesystem reserve, storage amplification and any preexisting backlog must enter the capacity model separately. Keeping those terms separate makes the estimate auditable instead of hiding uncertain overhead inside an unexplained margin.
Make the error bars operational
Express the required usable allocation as C = (Q + rT)/η, where Q is the initial unacknowledged payload and η is the measured fraction of allocated capacity available for retained payload after overhead and reserve. Bound the result with C_low = (Q_low + r_low T_low)/η_high and C_high = (Q_high + r_high T_high)/η_low, using workload measurements and a declared connectivity restoration envelope. These are engineering bounds, not statistical confidence limits, and their endpoints must include attachment bursts, compaction behavior and the actual retention policy. The numerical payload estimate remains conditional until those bounds are measured, particularly because assay output and journal amplification can rise together. Apply the admission rule that free allocated capacity must exceed C_high for the remaining committed work, otherwise defer new assays and bring active procedures to their validated local stopping boundaries.
Size reconnection as part of continuity
Approve continued execution only when measured durable write throughput covers peak journal ingress and restored export throughput exceeds ongoing ingress, because backlog clearance takes t_clear = B_backlog/(u_export − r_live), making replay bandwidth and the biological acceptability of a local pause the next hardware and protocol decisions.