copyspace-guard 0.2.2

Deterministic data-movement audit, validation and optimization reports

Copy-Space Guard

Copy-Space Guard is a metadata-only CLI for deterministic data-movement audits and CI regression gates. Status: production-oriented pilot / v0.2.

It takes a transfer demand matrix (src_slot,dst_slot,bits_total), validates schedules under a declared resource model, compares a baseline or customer schedule against a deterministic greedy candidate, and produces sales/engineering reports with lower-bound gap, utilization and estimated savings.

This package is intentionally small and pilot-friendly:

• no external Python dependencies;
• no payload data required;
• deterministic output for CI and regression tracking;
• machine-readable JSON plus human-readable Markdown/HTML reports.

Product promise

Give us one transfer trace or demand matrix. In a few days we show whether your data-movement plan is conflict-free, how far it is from a deterministic lower bound, and what CI gate can prevent future regressions.

Quickstart

From this directory:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e .
copyspace-guard analyze \
  --csv examples/ring15.csv \
  --bw 256 \
  --id ai-staging-ring15 \
  --roi examples/roi.yml \
  --outdir artifacts/demo

Open:

• artifacts/demo/report.html
• artifacts/demo/report.md
• artifacts/demo/summary.json

Expected terminal shape:

baseline: status=PASS ticks=768 lb=549 gap=0.398907 util=0.7143
greedy:   status=PASS ticks=549 lb=549 gap=0.000000 util=0.9992
saved_ticks=219 estimated_savings=9.73

The exact numbers depend on the input CSV and bandwidth value.

Input format

CSV with header:

src_slot,dst_slot,bits_total
0,1,65536
1,2,65536

Meaning:

• src_slot — source endpoint ID;
• dst_slot — destination endpoint ID;
• bits_total — transfer volume from source to destination.

Duplicate pairs are automatically merged.

Models v0: STRICT1 and READ1_WRITE1

STRICT1: within one tick, each slot can participate in at most one transfer, either as source or destination.

READ1_WRITE1: within one tick, each slot may send at most once and receive at most once.

This is a useful baseline for:

• endpoint-limited transfer systems;
• shuffle/staging/replication analysis;
• CI regression gates;
• comparing scheduler strategies;
• first customer audits where full topology is not yet modeled.

It is not a universal network model. For real pilots, confirm whether the client needs extensions such as READ1_WRITE1, broadcast, topology-aware bandwidth, asymmetric links or tier-aware storage constraints.

Commands

Check local pilot readiness

copyspace-guard --version
copyspace-guard doctor --root .
copyspace-guard doctor --root . --json

Analyze CSV and generate reports

copyspace-guard analyze --csv INPUT.csv --bw 256 --outdir artifacts/run

Optional:

--slots N
--id workload-name
--notes "free text"
--cost-per-tick 0.02
--model STRICT1  # or READ1_WRITE1
--bounds-subset-limit 20
--max-errors 100
--max-demands 100000
--max-slots 10000
--max-output-ticks 1000000

--bounds-subset-limit controls exhaustive STRICT1 subset-density enumeration and is protected by a hard cap to avoid accidental exponential runs.

Validate a schedule

copyspace-guard validate artifacts/run/instance.json artifacts/run/schedule_greedy.json --report artifacts/run/validation.json

Regenerate Markdown/HTML reports

copyspace-guard report artifacts/run/summary.json --outdir artifacts/report

Validate generated artifact contracts

copyspace-guard validate-artifact --kind summary artifacts/run/summary.json

Run production-oriented checks

make production-check

This runs release checks plus a small synthetic performance suite. The suite can also be run directly:

copyspace-guard bench-suite --outdir artifacts/bench-suite --max-total-seconds 30

Customer/current schedule input

For a real pilot, the customer may already have an actual schedule. Use CSV:

tick,src_slot,dst_slot,len_bits
0,0,1,256
0,2,3,256
1,1,2,256

Then run:

copyspace-guard analyze \
  --csv examples/ring15.csv \
  --bw 256 \
  --current-schedule-csv customer_schedule.csv \
  --outdir artifacts/customer-run

You can also convert a schedule CSV to JSON:

copyspace-guard schedule-csv-to-json --csv customer_schedule.csv --out schedule.json

CI gate command

After analyze, fail/pass thresholds can be checked locally or in CI:

copyspace-guard gate artifacts/demo/summary.json \
  --report greedy \
  --max-gap 0.15 \
  --min-utilization 0.85

Exit code 0 means pass, exit code 2 means fail.

Files generated by analyze

• instance.json — normalized workload contract.
• schedule_baseline.json or schedule_customer_current.json — current schedule artifact, unless --summary-only is used.
• schedule_greedy.json — deterministic candidate schedule, unless --summary-only is used.
• schedule_baseline.csv or schedule_customer_current.csv — CSV schedule artifact, unless --summary-only is used.
• schedule_greedy.csv — deterministic candidate schedule CSV, unless --summary-only is used.
• report_baseline.json or report_customer_current.json — validation metrics for the current schedule.
• report_greedy.json — validation metrics for candidate.
• summary.json — machine-readable comparison summary.
• report.md — human-readable audit report.
• report.html — shareable report for demos and sales calls.

v0.2 boundaries

Included:

• volume-based demand modeling;
• deterministic baseline and greedy schedules;
• STRICT1 and READ1_WRITE1 validators;
• lower-bound gap and utilization metrics;
• ROI estimates via roi.yml or a simple $ per tick assumption;
• sales-ready report artifacts.

Not included yet:

• production security hardening;
• topology/path selection;
• real transfer execution;
• cloud adapter importers;
• address-level offset validation;
• VCopySpace receipt ledger integration.

Known operational caveats:

• Customer schedule CSVs used in streaming mode must be sorted by non-decreasing tick.
• Full artifact mode can produce large schedule JSON/CSV files; use --summary-only for large pilots and CI.
• For large STRICT1 slot counts, subset-density lower bounds may be partial; check bounds_complete in reports.
• The greedy schedule is deterministic and useful for comparison, but it is not a proof of global optimality.

How this maps to the larger project set

• copy-space → scheduler, validator, lower-bound gap, CI-gate idea.
• vcopyspace → future enterprise layer: receipt-based metering, ledger, trace/replay, cost model.
• DDAS → long-term deterministic state-transition foundation.

ROI mode

Turn saved ticks into business impact:

copyspace-guard analyze \
  --csv examples/ring15.csv \
  --bw 256 \
  --roi examples/roi.yml \
  --outdir artifacts/demo

Example examples/roi.yml:

roi:
  tick_seconds: 1
  gpu_count_blocked: 64
  gpu_hour_cost_usd: 2.50
  runs_per_day: 12
  days_per_month: 30

Gate config file

copyspace-guard gate artifacts/demo/summary.json \
  --config examples/copyspace_guard.yml

Example config:

gates:
  report: greedy
  max_gap_to_lower_bound: 0.15
  min_utilization: 0.85

Docker

docker build -t copyspace-guard .
docker run --rm --user "$(id -u):$(id -g)" -v "$PWD:/work" copyspace-guard analyze \
  --csv examples/ring15.csv \
  --bw 256 \
  --roi examples/roi.yml \
  --outdir artifacts/docker-demo

Industry demos

copyspace-guard analyze --csv examples/ai_checkpoint.csv --bw 1048576 --roi examples/roi.yml --outdir artifacts/ai-checkpoint
copyspace-guard analyze --csv examples/db_shuffle.csv --bw 262144 --roi examples/roi.yml --outdir artifacts/db-shuffle
copyspace-guard analyze --csv examples/storage_replication.csv --bw 1048576 --roi examples/roi.yml --outdir artifacts/storage-replication
copyspace-guard analyze --csv examples/kv_cache_movement.csv --bw 1048576 --roi examples/roi.yml --outdir artifacts/kv-cache

Client package

See client-package/ for a minimal package that can be sent to a pilot customer:

• README_CLIENT.md
• sample_demands.csv
• sample_schedule.csv
• roi.yml
• copyspace_guard.yml
• run_local.sh
• intake.md

Anonymize demands or schedules

copyspace-guard anonymize \
  --kind demands \
  --csv raw_demands.csv \
  --out anonymized_demands.csv \
  --mapping slot_mapping.json

copyspace-guard anonymize \
  --kind schedule \
  --csv raw_schedule.csv \
  --out anonymized_schedule.csv \
  --mapping-in slot_mapping.json \
  --mapping schedule_slot_mapping.json

Use --mapping-in when anonymizing demands and schedules that must share the same slot-ID mapping. Do not share mapping.json unless you intend to reveal the original endpoint names.

Sales-oriented demos

Bad current schedule vs candidate:

copyspace-guard analyze   --csv examples/demo_bad_current_demands.csv   --bw 256   --current-schedule-csv examples/demo_bad_current_schedule.csv   --roi examples/roi.yml   --outdir artifacts/bad-current-demo

Conflict detection:

copyspace-guard analyze   --csv examples/demo_conflict_demands.csv   --bw 256   --current-schedule-csv examples/demo_conflict_schedule.csv   --summary-only   --outdir artifacts/conflict-demo

Large workloads can use --summary-only to avoid writing full schedule JSON/CSV artifacts. In this mode generated baseline/candidate schedules are streamed into the validator instead of materialized in memory. Customer schedule CSVs used in streaming mode must be sorted by non-decreasing tick.

Model and bound details

• Model limitations: docs/MODEL_LIMITATIONS.md
• Lower-bound definitions: docs/BOUNDS.md
• JSON schemas: docs/SCHEMAS.md
• Artifact contracts: docs/ARTIFACT_CONTRACTS.md
• Performance notes: docs/PERFORMANCE.md
• Pilot readiness: docs/PILOT_READINESS.md
• Production readiness: docs/PRODUCTION_READINESS.md
• Operations guide: docs/OPERATIONS.md
• Release process: docs/RELEASE_PROCESS.md
• Threat model: docs/THREAT_MODEL.md
• Data handling: docs/DATA_HANDLING.md
• Changelog: CHANGELOG.md

Benchmark

copyspace-guard bench   --slots 64   --bits-per-edge 1048576   --bw 1048576   --outdir artifacts/bench