Exercise Answers¶

Page Maps¶

graph LR
  family["Reproducible Research"]
  program["Deep Dive DVC"]
  section["Data Identity Content Addressing"]
  page["Data identity answer reasoning"]
  capstone["Data identity audit"]

  family --> program --> section --> page
  page -.interprets.-> capstone

flowchart LR
  attempt["complete artifact"] --> compare["compare evidence route"]
  compare --> challenge["challenge supplier and identity claims"]
  challenge --> revise["repair overclaims"]
  revise --> verify["repeat recovery proof"]

These answers model reasoning. Compare your claim boundaries, evidence, and limits rather than copying wording.

Answer: Bound the data identity claim¶

A suitable claim:

The pointer records baseline content ID be7d…a717, size 40, and workspace projection observations.csv. The named cache object and workspace independently share SHA-256 9771…2fb4. Paired contrasts show identity follows bytes rather than path. Recovery claims additionally require an accessible supplier and verified materialization.

Model distinctions:

Question	Evidence	Limit
which bytes recorded?	pointer ID and size	does not prove availability
where projected?	pointer path	does not identify bytes
locally available?	named cache object plus digest	not shared durability
remotely recoverable?	cache-cold pull after local loss	time/object bounded
scientifically suitable?	schema, provenance, domain checks	not proved by digest

Unsupported claims include universal remote completeness, future retention, scientific validity, and identical runtime interpretation.

Answer: Trace a pointer to matching bytes¶

Model trace:

Surface	Value	Evidence role
pointer path	`observations.csv`	projection location
hash type	`md5`	DVC addressing scheme in specimen
content ID	`be7d1d8e728323219fd8315dcbfea717`	managed identity
size	`40`	recorded size check
cache path	`.dvc/cache/files/md5/be/7d1d8e728323219fd8315dcbfea717`	local object address
workspace SHA-256	`977190…12fb4`	independent byte evidence
cache SHA-256	`977190…12fb4`	independent equality evidence

The content ID splits after two characters to form the cache path. DVC's MD5 locates its managed object; the audit's SHA-256 independently compares workspace and cache bytes.

The proof is not simply “hashes exist.” It joins pointer ID, derived address, object existence, size, and byte equality.

Answer: Disprove path-based identity¶

The same-content contrast changes location while holding bytes fixed:

data/observations.csv -> be7d…a717
archive/renamed.csv   -> be7d…a717

Inference: different paths do not require different content identity.

The changed-content contrast holds location fixed while changing bytes:

data/observations.csv: be7d…a717 -> dae5…00ae

Inference: stable path does not preserve content identity after byte change.

Together:

flowchart TD
  same["same bytes"] --> p1["path A"]
  same --> p2["path B"]
  p1 --> id1["same identity"]
  p2 --> id1
  changed["changed bytes"] --> p1b["path A"]
  p1b --> id2["new identity"]

Neither contrast proves semantic equivalence across different dataset roles. Equal bytes can be used under different claims.

Answer: Build a state-layer ledger¶

Model ledger:

Moment	Pointer	Workspace	Cache	Remote	Supplier	Decision
baseline after add	present	present	present	not configured	cache	local identity recorded
cache case before checkout	present	absent	present	not needed	cache	recoverable
cache case after checkout	present	present	present	not needed	cache	local recovery accepted
remote case after push	present	present	present	present	cache/remote	remote published
remote case after loss	present	absent	absent	present	remote	recoverable
remote case after pull	present	present	present	present	remote	remote recovery accepted
pointer-only before checkout	present	absent	absent	not configured	none	unrecoverable
pointer-only after checkout	present	absent	absent	not configured	none	rejection confirmed

The ledger prevents pointer presence from being mistaken for supplier availability.

Answer: Interpret commands as state moves¶

Model:

Command	Before	After	Required verification
add	unmanaged workspace bytes	pointer plus local cache object	pointer/cache/workspace agreement
push	cache object present, remote not yet evidenced	selected object remotely published	receipt plus later clean read
checkout	pointer/cache present, workspace absent	workspace projected	restored digest and identity
pull	pointer present, local cache/workspace absent, remote supplied	cache and workspace restored	supplier isolation and digest

Command choice:

unmanaged workspace: add;
workspace absent, cache present: checkout;
workspace/cache absent, remote present: pull;
pointer only, no supplier: no command can recreate the bytes.

Exit zero is command evidence. Recovery requires after-state and identity comparison.

Answer: Prove cache-only recovery¶

Model argument:

The pointer retained baseline ID be7d…a717. Before checkout, the workspace projection was absent and the named cache object remained. Checkout succeeded, recreated the workspace, preserved pointer identity, and produced baseline SHA-256. Accept local cache recovery for this object.

Why it is causal:

workspace supplier absent;
cache supplier present;
remote not needed;
restored identity matches.

Limit:

This does not prove remote availability, cross-machine recovery, future cache retention, or data validity.

Answer: Prove remote recovery after local loss¶

Model argument:

Push supplied baseline object be7d…a717 to the isolated remote. The audit then removed both workspace and named local cache object. Pull repopulated the cache and workspace. Recovered content ID equals baseline, and independent SHA-256 equals 9771…2fb4. Accept bounded remote recovery for this object and audit time.

Removing local cache rules out it as supplier. That makes the remote necessary in the tested route.

Not proved:

other objects exist;
other users are authorized;
future retention;
remote disaster recovery;
scientific suitability.

Answer: Defend negative recovery evidence¶

Model:

Survives

pointer path, content ID, size, and hash type.

Absent

workspace projection;
matching local cache object;
configured remote supplier.

Intervention

checkout returns 255;
stderr identifies unavailable cache-backed content.

After

workspace remains absent;
cache remains absent.

The audit result is PASS because every expected negative check succeeded. The recovery decision is REJECT because metadata alone cannot materialize content.

Useful searches include trusted teammate caches, governed remotes, archives, and published bundles by exact identity. If no trustworthy matching copy exists, record original identity as unrecoverable. Recollected data is a new identity.

Answer: Design a recovery manifest¶

Model shape:

{
  "schema": "data-recovery-manifest/v1",
  "recovery_id": "example-recovery-identity",
  "git_commit": "example-git-sha",
  "required": {
    "pointer": "data/observations.csv.dvc",
    "content_id": "be7d1d8e728323219fd8315dcbfea717",
    "hash_type": "md5",
    "size": 40,
    "baseline_sha256": "977190e4a0791b539f5eb50abc726b8d00195c0f4a0c2b39acfdd92d41512fb4"
  },
  "loss": {
    "workspace": "absent",
    "local_cache": "absent"
  },
  "supplier": {
    "type": "dvc_remote",
    "role": "recovery"
  },
  "receipts": {
    "state_move": "example-pull-receipt",
    "cache_identity": "example-cache-check",
    "workspace_identity": "example-workspace-check",
    "semantic_validation": "example-schema-check"
  },
  "decision": "ACCEPT_BOUNDED_RECOVERY",
  "limits": [
    "future retention not established",
    "scientific validity reviewed separately"
  ],
  "started_at": "2030-01-01T12:00:00Z",
  "completed_at": "2030-01-01T12:02:00Z",
  "verifier": "recovery-policy-identity"
}

The recovery ID joins one loss, supplier, move, and verification. The manifest records a remote role rather than credentials.

Answer: Assemble an independent identity review¶

A strong opening:

This packet distinguishes workspace location, recorded content identity, and supplier availability. Paired contrasts show identity follows bytes. Cache and remote cases isolate two recovery suppliers. Pointer-only failure proves metadata cannot replace missing content.

Artifact purposes:

Artifact	Purpose
`identity-claim.md`	bounds identity and meaning
`pointer-trace.tsv`	connects metadata to bytes
`path-contrasts.md`	disproves locator identity
`state-ledger.tsv`	records surviving layers
`command-receipts.md`	explains bounded moves
`cache-recovery.md`	proves local supplier
`remote-recovery.md`	proves remote supplier
`negative-recovery.md`	preserves rejected claim
`recovery-manifest.json`	joins recovery evidence

Unresolved risks:

remote retention and access policy;
integrity for unaudited objects;
schema/provenance coverage;
dependent result currency;
published bundle completeness;
runtime interpretation.

Verification:

make PROGRAM=reproducible-research/deep-dive-dvc capstone-data-identity-audit
make PROGRAM=reproducible-research/deep-dive-dvc capstone-data-identity-selftest
jq empty artifacts/learning/deep-dive-dvc/data-identity/recovery-manifest.json

Self-review standard¶

Your packet should distinguish:

locator from identity;
pointer metadata from content;
workspace projection from cache object;
local cache from shared remote;
command success from verified state move;
identity recovery from semantic validity;
current remote availability from retention;
audit PASS from recovery ACCEPT.

Those distinctions are the practical skill of Module 02.