Meridian Field Services: instructional example, not project evidence
The compiled view of the live register. Every decision citation elsewhere in the workspace resolves to a row here, and to nowhere else.
Scope: full register, as of 2026-09-15 · live register at workspace registers[1]. Tier mapping (agreed E. Sandoval with M. Chen, July 7): architectural signs by the engineering lead, product owner consulted; business signs by the dispatch operations owner with the sponsor; operational signs by the delivery lead. Contested tiers escalate to the steering group. Reversibility sets ceremony: one-way doors get slow, broadly consulted review; two-way doors get fast small-group decisions.
| Id | Title | Tier | Door | Status | Date | Decider | Chain |
|---|---|---|---|---|---|---|---|
| ADR-2 | Poll-based position ingestion | architectural | two-way | superseded | Jul 8 | E. Sandoval | → ADR-6 |
| ADR-5 | Constrained suggestion output | architectural | one-way | accepted | Jul 11 | E. Sandoval | none |
| ADR-6 | Vendor webhook push for positions | architectural | two-way | accepted | Jul 12 | E. Sandoval | supersedes ADR-2 |
| ADR-7 | Access-rights model: RBAC with hub scope | architectural | one-way | accepted | Jul 15 | E. Sandoval, A. Reyes | none |
| BD-1 | Pilot at the Toronto hub first | business | one-way | accepted | Jul 9 | M. Chen | none |
| BD-2 | Hub-six go/no-go criteria | business | one-way | proposed | decide by Oct 1 | M. Chen (owner) | none |
| OD-1 | Single-owner rollback authority | operational | two-way | accepted | Jul 13 | Delivery lead | none |
MADR-shaped frozen entries carrying the evaluation law (what, why considered, who, how evaluated, findings) plus the confirmation rule. Entries are written before or during the decision; thin alternatives betray a retrospective entry.
Context: position freshness drives board liveness (quality scenarios[2]); polling at 60-second intervals (ADR-2) cannot meet the 5-second visibility target and burns vendor rate limits. Drivers: liveness target, vendor rate limits, adapter-seam isolation. Options: tighter polling (fails rate limits); vendor webhook push (meets target, needs signature verification); hybrid poll-plus-push (complexity without added freshness). Who & how: E. Sandoval with A. Reyes, walked against the INT-1 contract July 12. Outcome: webhook push; signature verification becomes a security requirement (SR-021[3]).
Consequences: positive, the liveness target is reachable. Negative, stated: a single vendor push path is now an availability dependency, monitored as RSK-04[4]; the spoofing surface moves to the webhook endpoint, analyzed as TM-01[5]. Confirmation: the PS-1 board-liveness run meets its p95 target (PS-1[6]).
Kept in full, never deleted: its context (early uncertainty about vendor webhook maturity) explains why polling was ever chosen. The chain links run both directions.
ADR-5 and ADR-7 render as compressed cards; their consequences already live where they act: TM-08 eliminated[5] and the permission model[7].
Context: five hubs, one pilot budget (objectives[8]). Drivers: fastest observable value, union-briefing sequencing, data-migration risk containment. Options: Toronto (largest volume, union-represented, richest data); Calgary (smallest, fastest, least representative); parallel two-hub (splits attention). Who & how: M. Chen with the sponsor, scored against the drivers July 9. Outcome: Toronto first. Consequences: the union monitoring-notice question becomes pilot-critical (tracked once at security open questions[3]); volume assumptions rest on Toronto figures (ASM-01[9]). Confirmation: pilot value review at week 6 against the BRD's success measures.
Drivers and options listed, outcome empty; logging it before it is made is what keeps the hub-six alternatives honest.
Context: rollback debates during incidents waste the window (rollback runbook[10]). Outcome: the release owner pulls rollback alone; no meeting; fix-in-place is time-boxed first. Decider: delivery lead, July 13. Confirmation: the next rollback drill's decision-to-pull time stays under its target. The runbook owns the procedure; this entry owns only the authority decision.
| Question | Owner | Decide by | Blocks |
|---|---|---|---|
| Should BD-2's go/no-go criteria include the production-sizing breakpoint re-run (performance open questions[6])? | M. Chen | Oct 1, 2026 | BD-2's acceptance only |