This is a real evidence pack, generated by the RailCall dev station from its own 1,005 sealed receipts — published unedited, including the 3 receipts that failed re-audit (failures are reported, never hidden). Generate yours in Studio → Receipts → Evidence pack. railcall.ai

RailCall Compliance Evidence Pack (sample)

Period: 2026-07-17T14:01:31 → 2026-08-16T14:01:31 (UTC) · Station: unregistered · Generated: 2026-08-16T14:01:36Z

Generated locally by the station that performed these actions. Free during early access — evidence packs may become a paid feature later; your receipts and raw export remain free forever.

1 · Executive summary

1005 governed actions in the period, 1002 of them Ed25519-signed. Total external spend: $100.00.

96% of workflow execution units in this period ran as local compute without consulting a model (27 of 28 units, from the sealed routing verdicts in the receipts themselves).

Outcomes

COMPLETED524
REFUSED114
ROLLED_BACK60
SENT10
approved_not_executed46
blocked_by_policy1
credential_present_untested3
executed84
failed_safely8
failed_with_receipt6
not_configured1
pending_approval109
unknown39

By channel

cli1
mcp1
mcp_tools_call1
scheduler665
session6
studio23
studio_mcp_gate6
terminal_confirm26
ui_click66
unknown205
vscode_chat3
vscode_hud2

Refusals by gate

capability_scope1
disk_space1
live_policy2
plan_pin_changed1
policy_block1
require_human_unattended108

2 · Verification

The station re-audited its ENTIRE receipt corpus from canonical disk bytes at generation time:

UNSIGNED0
UNTRUSTED_KEY0
VERIFIED265
VERIFIED_MOCK5

3 receipt(s) FAILED re-audit — named in manifest.json under verification.failed; a failure is reported, never hidden.

An auditor can independently re-verify any receipt at railcall.ai/verify (runs entirely in the browser — no RailCall server involved) or with the offline CLI verifier. The station's public key and fingerprint are in manifest.json.

3 · Execution policy in force

{
 "apply": {
  "live_workflows_enabled": false
 },
 "dag": {
  "live_workflow_ids": [
   "singleops_backlog_to_discord",
   "singleops_backlog_to_google_sheet",
   "single_ops_to_sheet"
  ],
  "live_workflows_enabled": true
 },
 "mcp": {
  "live_workflow_ids": [],
  "live_workflows_enabled": true,
  "module_reads_enabled": true,
  "lazy_tools_enabled": true
 },
 "mcp_expose": {
  "workflow_ids": []
 },
 "routing": {
  "local_only": false,
  "sensitive_stays_local": false
 },
 "guards": {
  "min_free_disk_mb": 500
 }
}

Every change to this policy is a signed audit event; the full change history is in the station's hash-chained audit log (Studio -> Settings -> Audit chain -> Raw JSON).

4 · Control mapping

HIPAA §164.312(a)(1) Access controlStudio session tokens (0600, loopback-only), per-channel authentication (session / cli / scheduler / mcp), RBAC roles on team stations.
HIPAA §164.312(b) Audit controlsEvery action seals an Ed25519-signed receipt; refusals are receipts too; hash-chained audit log with off-box witness anchoring.
HIPAA §164.312(c)(1) IntegrityReceipt integrity hashes recompute from canonical bytes; tampering with any sealed field breaks the signature (verified in this pack).
HIPAA §164.312(d) Person or entity authenticationApprovals bind the exact payload hash; station identity is issuer-signed; team approvals are member-key co-signed.
HIPAA §164.312(e)(1) Transmission securityCredentials never leave the local vault; egress passes policy + redaction gates; MCP responses are attestation-only by default.
SOC 2 CC6/CC7 (change & ops monitoring)Execution-policy changes are signed events; plan pins stop drifted workflows; spend caps and freeze provide preventive controls.

5 · Appendix

The raw receipts for this period are included under receipts/ — the same data available through the free export on every tier. manifest.json lists each file's sha256 and carries the pack's own integrity hash + signature.