Vendor management
Vendor SLA language that operations can actually enforce
Contracts that promise ‘best efforts’ produce best excuses. Trading and research operations need testable verbs: which percentile of freshness, over what window, measured how, with which escalation clocks, and what happens to the invoice when reality diverges. This note is how you align procurement, legal, finance, and the NOC around numbers you can graph—not adjectives you can litigate.
Ambiguous terms to burn
‘Commercially reasonable’, ‘promptly’, ‘material’ without numeric triggers—each is a lawyers’ picnic and an ops fog bank.
Push counterparts to define observability: who runs the stopwatch, from which vantage, with what tooling. If measurement is ‘vendor reports internally,’ you bought a black box.
Good SLA atoms
P95 freshness vs authoritative clock; max consecutive missing bars; time-to-ingest after vendor publish; error rate on API calls; RTO/RPO for historical replays.
Separate availability of API from correctness of data—vendors love conflating ‘we answered 200’ with ‘the series was right.’
Attach exclusion lists: planned maintenance, client-caused throttling, upstream exchange halt—define what is out of scope in writing.
Escalation ladders
L1 vendor portal, L2 named TAM with 30-minute callback, L3 exec bridge with breach notice—spell it out. If the contract lacks phone numbers, you bought SaaS without a spine.
Prenegotiate comms during Sev-1: who may speak externally, in which channels, with what latency—PR should not improvise with NOC guessing.
Credits vs remedies
Service credits are weak tea for a trading desk; still negotiate them—finance uses them to justify replacement RFPs. Pair with termination for persistent breach clauses you can measure.
Translate credits into hours of analyst time and delayed decisions when briefing executives—make the soft cost visible.
Synthetic monitoring
Probe the same endpoints production uses; store results beside lineage so ‘green vendor page / red reality’ is obvious in postmortems.
Alert on divergence between vendor-published status and your probes—silent divergence is how trust dies slowly.
Change management
Schema changes require semver or date-stamped migration guides; breaking changes deserve lead time in writing, not discovery at Sunday deploy.
Require vendor playbooks for silent restatements: notification cadence, client diff artifacts, and rollback expectations.
Multi-vendor resilience
Document which series have hot failover and which require manual swap—honesty prevents fantasy risk limits.
Run quarterly swap drills on cold-standby names—if drill never happens, assume manual path is fantasy.
Finance coupling
Give CFO staff a monthly one-pager: contractual SLO vs measured, credits accrued, open breaches, and projected renewal leverage.
Without that bridge, infra fights vendor battles alone—and renews at a disadvantage.
RFP questions that separate adults from vendors
Show last quarter’s measured P95 ingest lag for our symbol set; show postmortem for your worst outage; show how you notify clients before silent restatements.
Ask for example runbook pages from a real incident redacted—template fluff is a negative signal.
Public desk routes referenced above
Same URL discipline applies after you authenticate—bookmarkable paths, no tenant data on these pages.
- Service status Public narrative around uptime; align its claims with what engineering will defend in postmortems.
- Readiness JSON Coarse liveness contract—use it to rehearse what is safe to expose anonymously vs behind auth.
- Operating rhythm Template language for stages, owners, and completion rules in a mandatory morning loop.
- Guides Pair public guides with internal SOPs so training points at one narrative spine.
- Platform framing How tenancy and public chrome split when SEO and auth walls must coexist.
Operator toolkits
Checklists and scripts you can lift into runbooks, vendor RFPs, or board appendices.
Commercial and operational patterns only; have counsel review enforceability in your venue.