Skip to content

Module Glossary

Page Maps

graph LR
  family["Reproducible Research"]
  program["Deep Dive DVC"]
  section["Recovery Scale Incident Survival"]
  page["Module Glossary"]
  capstone["Capstone evidence"]

  family --> program --> section --> page
  page -.applies in.-> capstone
flowchart LR
  orient["Orient on the page map"] --> read["Read the main claim and examples"]
  read --> inspect["Inspect the related code, proof, or capstone surface"]
  inspect --> verify["Run or review the verification path"]
  verify --> apply["Apply the idea back to the module and capstone"]

This glossary belongs to Module 08: Recovery, Scale, and Incident Survival in Deep Dive DVC.

Use it to keep the module language stable while you move between the core lessons, the worked example, the exercises, and capstone review.

How to use this glossary

Read the module index first. Return here when two nearby ideas begin to blur. The terms below are decision tools, not vocabulary trivia.

Recovery contract terms

Term Meaning in this module Evidence example
recovery unit The smallest complete state whose restoration satisfies a real use, audit, or rollback promise. dataset pointer, content identity, parameters, and release manifest
durability boundary The combination of identity, storage, access, and verification expected to survive local loss. Git tag, candidate remote, read role, cache-empty receipt
protected revision A concrete Git revision whose DVC-referenced state still carries an obligation. main or release-2025-q1
recovery point objective (RPO) The maximum acceptable recorded-state loss. no loss of either protected specimen revision
recovery time objective (RTO) The maximum acceptable duration from recovery start to usable restored state. complete protected restore within an agreed interval
content identity A claim that recognizes the exact restored bytes. SHA-256 or DVC object hash
semantic verification A check that restored state still has the promised meaning or shape. expected row count, schema, population, or release verifier
cache-empty restore A restore performed without a pre-existing local artifact cache that could hide an incomplete remote. separate reviewer clone with .dvc/cache removed
read-only verification A proof route that observes candidate storage without populating or reconfiguring it. checkout, pull, digest, and verifier commands without push
evidence independence The property that a check does not create the state it claims to observe. candidate pull performed by a role without write permission

Storage and authority terms

Term Meaning in this module Common confusion
workspace Materialized files used by commands now. file presence is mistaken for durable storage
local DVC cache Machine-local content-addressed objects. a complete author cache hides remote gaps
DVC remote A configured content store that DVC can push to or pull from. configuration is mistaken for object coverage
recovery authority The storage boundary the project currently promises will restore protected state. every configured remote is treated as equally authoritative
candidate remote Storage being populated and evaluated for possible authority. a successful current pull is treated as accepted cutover
legacy remote Prior authority retained during migration, rollback, or archive review. continued existence is mistaken for a tested rollback plan
cutover The governed decision that changes recovery authority. URL change is mistaken for authority acceptance
rollback interval A bounded period during which authority can return to the prior source. source is deleted immediately after candidate acceptance
source-exit gate Evidence and approval required before decommissioning the prior source. a migration audit is treated as deletion permission

Retention and cleanup terms

Term Meaning in this module
obligation The active operational, rollback, publication, legal, decision, or incident reason state must remain available.
retention class Protected, time-bounded, archive, or disposable treatment that leads to concrete actions.
end condition An observable event that permits a retention class to change.
hold A temporary override that prevents ordinary expiry because incident, legal, or review evidence is still needed.
disposition evidence The identity, authority, scope, and reason recorded when state expires, moves, or is deleted.
reference scope The workspace, branches, tags, commits, experiments, and other projects DVC reads to build the preservation set.
reachability Whether selected DVC metadata refers to a cache object that must be kept.
garbage collection Removal of cache or remote objects outside the selected preservation set.
--dry preview Current DVC option that prints garbage-collection candidates without removing them.
remote collection Garbage collection using --cloud, with a larger shared blast radius than local cache collection.
shared cache One cache used by more than one DVC project, requiring the union of their protected references.

Incident and handoff terms

Term Meaning in this module
recovery incident Any event that makes promised state unavailable, unverifiable, or too slow, even if bytes still exist.
containment Temporary control that prevents deletion, cutover, publication, or uncontrolled repair from widening the incident.
first failure The earliest preserved evidence of the broken promise before repair changes the state.
affected obligation A promise shown to fail by evidence.
unknown obligation A promise not yet tested; neither affected nor safe.
last known good contract The most recent accepted combination of revisions, identities, authority, executor, access role, time, and result.
bounded repair A mutation limited to the diagnosed boundary, authorized source, target, object set, and stop conditions.
closure proof New read-only recovery evidence produced after repair from an independent empty-cache reviewer.
incident packet Contract, receipts, environment, impact, timeline, repair, verification, and follow-up evidence.
maintainer handoff Independent demonstration that an incoming steward can find, run, diagnose, and govern the recovery route.
human cache Undocumented knowledge held by one maintainer about remotes, credentials, exceptions, and decisions.

Distinctions that matter

Do not collapse Why
revision resolves / object restores Git history and DVC content fail at different boundaries
object restores / identity matches available bytes may not be the promised bytes
identity matches / meaning is accepted a digest cannot decide scientific or downstream validity
current state / protected history current workspace coverage may omit releases
migration / verification write-enabled copy can repair candidate storage
candidate acceptance / source retirement rollback, holds, archive, and deletion authority remain
RPO / RTO state coverage and recovery duration are different objectives
failed / unknown untested state should not be declared affected or safe
repair receipt / closure proof a successful write does not prove later read-only recovery
result: PASS / decision: ACCEPT a correctly rejected teaching case passes its test by deciding BLOCK

Stable review questions

Use these when a claim feels broad:

  • Which artifact family and protected revisions are in scope?
  • Which obligation protects each revision?
  • What are the expected content identity and semantic checks?
  • What RPO and RTO were declared before the drill?
  • Did every reviewer begin without artifact cache?
  • Could the verification identity write to candidate storage?
  • Did Git resolve before DVC retrieval was attempted?
  • Does garbage-collection scope include protected branches, tags, experiments, and shared projects?
  • What would dvc gc ... --dry propose, and who may approve action?
  • Which evidence was preserved before repair?
  • Which new reviewer supplied closure proof?
  • Does the decision accept candidate authority, retire the source, or neither?