Skip to content

support-methodology.md

Latest commit

 

History

History
67 lines (53 loc) · 2.96 KB

support-methodology.md

File metadata and controls

67 lines (53 loc) · 2.96 KB
Raw

Provider Support Methodology

Provider support owns the method, not the data-owner environment. For an external proof, provider support never operates commands, requests screen access, receives prohibited data, or interprets raw results. The data-owner technical owner controls the sample, execution, local analysis, evidence request, and approval.

Before execution

  1. Confirm the data owner has the runbook, the sampling plan, and the reviewed CodeProbe version.
  2. Confirm the sample was frozen before results: time window, selection method, anonymous task mix, exclusions, and the rule for attrition.
  3. Confirm the profile has exactly A and B, changes one declared dimension, uses the same task set, and requires three repeats.
  4. Confirm the data owner expects at least ten paired distinct scorable tasks after attrition.
  5. Confirm provider support will log every intervention using the fixed actor-role and event codes.

Do not approve a sample's semantic representativeness. That judgment belongs to the data-owner technical owner and is recorded as an attestation.

During execution

Permitted help is limited to the published runbook, generic guidance, asynchronous coordination, the bounded live session, sanitized diagnostics, and an optional security follow-up. Work only from data-owner-reported coded status. Do not ask for repository identifiers, paths, source, prompts, patches, traces, task-level results, raw results, logs, or diagnostics.

Stop the external proof if Provider Engineering participates or if any provider person directly accesses the environment, provides bespoke code, repairs evidence, receives prohibited data, or reinterprets raw results.

Evidence review

Accept only a directory containing the exact five artifacts listed in EVIDENCE_BUNDLE.md. Do not accept archives of .codeprobe/, screenshots, local aggregate reports, ad hoc redactions, or supplemental files.

Use bounded-findings.md to check:

  • the approval digest is consistent;
  • the sample contains at least ten paired distinct tasks and three repeats;
  • A and B used the same task set;
  • the data owner attested privacy, sample fidelity, result fidelity, and usefulness;
  • support was not disqualifying; and
  • the conclusion is exactly advance_a, advance_b, or insufficient_evidence.

The fixed exported findings.md is the report of record. The review worksheet must not add claims about ROI, productivity causality, production outcomes, procurement readiness, or customer-grade calibration.

Internal rehearsal

For the internal stage, two non-implementer operators independently run this same protocol on a shared authorized repository, then exercise it on a second authorized repository. Internal access does not relax the task floor, repeats, sampling, logging, or evidence gate. Internal results prepare operators but cannot replace the external proof.