PILOT OR RENEWAL / SUPPLIED EVALUATION EXPORTS
Pilot Evidence Preparation
A provider's “Pass” is one observation. Keep the cases you expected to test, every result you supplied, and the questions still waiting for an independent reviewer.
This browser method starts blank. Nothing is submitted. Your source text stays in this browser's memory until you reset or close it; downloads are files you choose to keep. No model or tenant is connected.
TRY THE METHOD
Use a fictional example, or prepare your own permitted export.
Supported input: the specified evaluation CSV headers for this adapter. Quoted fields and original line endings remain in the retained source. Up to 200 data records and 100 required cases. No evaluation is run here.
No example loaded.
Name the pilot boundary.
These are supplied facts for this file. No identity, permission or approval is inferred.
What did you agree to test?
Keep this register independent from the export. A missing result must not make a required case disappear. Case IDs and relationships are entered explicitly.
Retain the supplied export.
Use only a file you are allowed to share with this method. Every logical data row remains visible, including blank, repeated or malformed records.
Supported header contract and limits
Required exact columns: question, actualResponse, testMethodType_1, result_1. Optional: expectedResponse, and method, result, threshold and explanation fields for methods 1–3. Other headers stay visible and open. Supported provider results: Pass, Fail, Invalid, Error; other values stay unknown.
UTF-8 only. A CSV byte-order mark is retained. The 128 KiB source, 256 KiB packet and row limits reject oversized input as a whole. Original CSV rows are never executed as instructions or formulas.
No source records prepared.
Account for every supplied row.
Map each observation to an exact required case, record an owned exclusion, or keep it unresolved. A repeated question does not establish retry identity. This method never selects a latest or best attempt.
For mapped or excluded rows, separately check that you inspected the source record and this preparation mapping. That declaration never accepts the answer or approves the pilot.
No row inspection is recorded.
Record current preparation declarations.
Check only statements the named owner can actually declare for these exact inputs. A later source, register or mapping edit clears these declarations and row inspections.
These are declarations, not authentication of a person, source rights or actual testing. Independent institution reviews remain pending.
CURRENT SUPPLIED INPUTS
The file before the decision.
Every required case
Open work
Every original row and native result
Keep the complete preparation file beside the Scorecard draft. The draft indexes required cases; it does not replace the full source row register. Every independent review, acceptance and effort value in that draft stays pending or unknown.
Supplied-input fingerprints
Fingerprints compare supplied input. They do not verify identity, authority or whether a review occurred.
Reopen a preparation packet or use the exact JSON contract
Ordinary preparation uses the fields above. This optional editor keeps invalid drafts intact. Apply a JSON edit explicitly before preparing or downloading.
No unapplied JSON edit.
A PREPARATION FILE. A DECISION FOR YOUR INSTITUTION.
Keep the failed cases.
Keep the unknowns.
Keep the owner.
This method parses supplied evaluation text. It does not run evaluations, connect to a tenant or core, compile or execute institution changes, take a financial action, certify compliance, give a legal opinion or claim achieved savings.
Native provider results, completeness and preparation declarations do not establish independent acceptance. This release supports a specific supplied export contract; real customer exports and native model activation need separate checks.
No account, cloud history, monitoring or reminders. No production remote MCP endpoint. If you later share a source or result with an assistant, that provider's actual account terms apply.
LOCAL EVALUATION RELEASE / 0.1.0
Keep the method with the complete file.
These separate packages use the same supplied-input preparation engine. They do not extend Evidence Desk 0.6.0. Node.js 22 or later is required for local commands. No external dependencies or model key are required by this method.
- Local browser and CLI source 0.1.0
- Focused local skill 0.1.0
- Claude local assistant source 0.1.0
- Gemini CLI extension source 0.1.0
For the optional local assistant, follow its included instructions and verify a fictional tool call in your approved host. A download, local protocol test, authenticated model call and public directory listing are separate checks. Native activation and directory acceptance are unverified.
This site has no production remote MCP endpoint. The Grok web application and remote connectors are unavailable here. If you separately share a file with an assistant, that provider’s actual account terms apply.