Trustyu Forge
R1 Reliable Harness
The system surrounding the model determines reliability
Explicit workflow and workspace, isolation, small changes, observable feedback, orchestration roles, and durable decisions form the public baseline.
Evidence research v1.7 · cut 2026-07-17T23:24:33Z
Public search
source-bound.
This registry publishes the sanitized artifacts from the run supporting the v1.7 search. External sources were treated as unreliable data; each fetch was limited by policy, linked to snapshot and hashes and discarded after derivation. No raw external content was persisted.
423.265bytes received
21HTTP calls
37.675 mssummed fetch time
9 / 9revised claims
0 / 0model calls/tokens
$0direct model cost
Three questions
Supported Conclusions
independent.
Each claim maps to at least two sources and each search uses at least three independent groups. The semantic relationship was reviewed by human; receipts prove acquisition, checks and source binding, not an automatic truth.
R2 · Secure & Auditable
Testing, isolation and provenance are distinct controls
Policy in text is not enough. Auditable delivery separates AI verification, least privilege, sandboxing, provenance, and artifact-independent validation.
R3 · Evals, OTel & FinOps
Quality and cost need a unit of value
Evals combine criteria and observable execution. Non-zero cost comes from official billing; absence of export remains unknown, never an invented zero.
Claim boundary: This pack proves that public intake and multi-source research were successfully exercised for 20 sources reviewed in this cut. It does not prove Operational, Attested, certification, product adoption, customer outcome or global leadership.
Public ledger
20 fonts.
20 verifiable pairs.
The manifest records the request, capabilities and budget. The receipt records result, final URL, snapshot, hashes, checks, retention and measured cost. The files below are byte-by-byte mirrors of the canonical pack merged into trustyu-docs@994bbad7.
Operating Limits
Measured zero does not erase
the unknown.
This acquisition did not call OpenAI, Anthropic, or Gemini; Therefore, the direct cost of this intake model is zero. Future billing is documented but not configured. Without authorized credential/export and reconciliation by project/SKU, product spend remains unknown.
Provider billing
OpenAI Admin Usage Costs API, Anthropic Usage and Cost API and Google Cloud billing export are in the official catalog. State: documented-not-configured.
Fail-closed real
HTTP 403, DLP/prompt-injection and content-hash-mismatch continued to be denied. The execution swapped mutable pages for official immutable snapshots, without bypass.
Origin: gate framework-only from issue Docs #292, executed by policy profile forge.public-evidence-fetch/v1 of trustyu-ai-env@fb43ea3d and accepted in trustyu-docs@994bbad7. Public artifacts do not include retrieved content or secrets.