Stage-Completeness Receipts (P115) - TextForAgents LLC
Signed proof that a zero is a real zero. At every pipeline stage
boundary an empty result is byte-identical to a failed scan (auth
expired, wrong path, swallowed exception, truncated pagination).
This service issues a signed receipt at the boundary: scan
parameters, scanned count, SHA-256 listing hash, Merkle root, the
full listing, the matched set, and the explicit missing list
(expected - scanned). Completeness is computed by the issuer,
never asserted by the caller: an auth_failed/error scan can only
be UNKNOWN, a truncated or missing-bearing scan PARTIAL at best,
and only a clean ok scan with an empty missing list is COMPLETE.
A downstream stage concludes EMPTY only from a verified, fresh,
COMPLETE receipt whose listing hash and parameters match what it
is about to trust; every other state is UNKNOWN.

Start: POST /register {"account_name": "...", "payer_wallet": "0x..."}
returns a per-customer token (self-serve; no operator-issued tokens).
Then, with Authorization: Bearer <token>:
- POST /receipts {stage, source, query, listing, matched, expected,
  scan_status, pages_seen, truncated, ttl} - first receipt free
  (one per operator, ever); then x402 $0.02/receipt.
- POST /verify {receipt, presented_listing, expected_source,
  expected_query} - x402 $0.005/check. Returns valid/trusted/fresh,
  the completeness grade, and the downstream conclusion
  (EMPTY / NONEMPTY / UNKNOWN).
- POST /probe {receipt_id or receipt, source_listing,
  sample_fraction} - x402 $0.25/probe. Independent re-scan of a
  sampled slice, hard-capped server-side at 5% of the claimed set
  (above the cap: HTTP 400). The probe is a coarse-fault tripwire
  only - measured at a 5% sample: wrong-path 1.00, truncated-50%
  1.00, 10%-missing 1.00, 3%-missing 0.80, 0.3%-missing 0.15. It
  does NOT reliably detect small holes; small-hole completeness is
  carried by the listing hash + explicit missing list, not by
  sampling.
- GET /receipts/<id> - your own receipt (operator-scoped).
- GET /issuer-key - the issuer's Ed25519 public keys (current +
  archived) so any third party can verify a receipt offline.
- POST /mcp - MCP surface (initialize / tools/list / tools/call)
  with tools issue_receipt, verify_receipt, coverage_probe,
  get_receipt; the same prices and trial rule apply.

TTL guidance (per pipeline): TTL is issuer-set at issuance, 1-3600s,
default 300s; verification refuses stale receipts. Set the TTL just
above your stage's scan interval plus downstream hand-off latency:
a receipt should expire before the scanned set can plausibly change
underneath the downstream stage - short TTLs (60-300s) for fast
queues/inboxes, up to 3600s only for slow, append-mostly sources
whose scan interval is itself long. Never reuse a receipt across
scan cycles.

Prices: Receipt issuance $0.02/receipt - Verification $0.005/check
- Coverage probe $0.25/probe (sample capped at 5% of the claimed
set). One free receipt per operator, ever; no perpetual free tier.
No priced direct competitor is named in the diligence record; the
prices sit far below one hand recount of a scan.
Pay-to (x402): 0x7a482F3B58780519880a6A13F983E958f2Ed6A61 (Base USDC).
Support: support@textforagents.com. Terms: /terms - Privacy: /privacy.
This product is not represented as FTO-cleared; its freedom-to-
operate screen is AMBER (screen-level, no claim chart yet).