Portfolio Test Matrix

NoDrift portfolio testing was performed across structure, startup behavior, source truth, approval control, correction handling, privacy boundaries, install safety, evidence generation, and buyer-style startup. This public matrix summarizes the results without exposing private build records or local paths.

52/52Expanded checks passed
100/100Later expanded average
95.83Earlier four-level average
10/10Claude Code initial pass
Public-safe score summary

Multiple rounds, not one simple pass

Earlier benchmark rounds scored in the high nineties, including 98/100 and 96/100, with one four-level static run averaging 95.83/100. After corrections and hand adjustments, the expanded Demonstrator/Basic static and simulation suite passed 52/52 checks with a 100/100 average under that scorecard's rule.

10Basic Codex readiness levels checked
52Expanded static/simulation tests passed
98Earlier scored correction item
96Earlier scored correction item

Score Chart

Later expanded suite
100/100
Earlier correction BGT-008
98/100
Earlier correction BGT-009
96/100
Earlier four-level run
95.83/100

Portfolio Tracks

Codex Master Track

Demonstrator and Basic

  • Basic Codex checked across 10 readiness levels.
  • Source inventory and install-map paths verified.
  • Buyer-facing draft labels corrected.
  • Unsupported audit and evidence claims narrowed.
  • Buyer delivery manifest added and verified.
  • Fresh local install simulation passed.
  • Live buyer-style Basic startup test passed.
  • Codex Demonstrator was reduced from Basic into the approved five-file behavior sample.
  • Demonstrator direct-answer, bounded-directive, and go-approval behavior was live retested and passed.
Claude Code Track

Initial and expanded checks

  • Initial Claude Code Basic static/install pass scored 10/10.
  • Required files present.
  • Settings files parsed correctly.
  • No active Basic hooks shipped.
  • No Basic project subagents shipped.
  • Startup source terms present.
  • Approval and external-action gates present.
  • Fresh-folder copy preserved required files.
  • Install guidance blocked blind overwrite of existing startup files.
  • Unsupported readiness claims were absent.
  • Claude-specific design checked against the adapter note.

Portfolio Test Rounds

Test Phase What Was Checked Recorded Result Public Boundary
Initial Claude Code Basic static/install simulationRequired files, settings parsing, no active hooks, no project subagents, startup terms, approval gates, fresh-copy behavior, install conflict safety, claim safety, and Claude-specific design.10/10 passedStatic and install-simulation evidence.
Earlier four-level static runDemonstrator, Basic, Extended, and Teams prototype structure and control readiness before later corrections.95.83/100 averageEarlier static run, not final sale-readiness evidence by itself.
Earlier benchmark correction itemsSpecific benchmark guard tests after the correction pass identified in the working records.98/100 and 96/100Fine-grained benchmark evidence, separate from the later 52-test pass.
Expanded Demonstrator/Basic static simulationStructure, settings, startup, source truth, approval gates, privacy, readiness claims, install safety, memory, corrections, continuity, tester packet, and evidence output.52/52 passed; 100/100 averageLocal static/simulation evidence, not a claim of live behavior in every app.
Basic Codex sale-readiness reviewSource inventory, install paths, buyer wording, claim boundaries, approval controls, topic-map behavior, error protocol, delivery manifest, install simulation, and startup behavior.10 readiness levels checkedSource-sale readiness and buyer-startup evidence for Codex Basic.
Codex Basic buyer-style startupFresh project-attached Codex test using the startup command and installed portfolio files.PassedObserved Codex startup behavior for Basic.
Codex Demonstrator buyer-style startupDirect answers, bounded next directives, go approval gates, and project-description intake.PassedObserved Codex startup behavior for Demonstrator.
Hand-adjustment readiness passBuyer-facing labels, delivery manifest, unsupported claim cleanup, public/private wording, Demonstrator reduction, go-gate preservation, and final readiness wording.CompletedCorrected the master Demonstrator and Basic portfolios into the current sales-ready/source-sale-ready position.

Expanded 52-Test Scorecard

This scorecard used PASS = 100, PARTIAL = 50, and FAIL = 0. Every listed test received a PASS score in the later expanded static/simulation run.

IDCategoryTest NameScore
CC-SIM-001StructureCandidate root exists100/100
CC-SIM-002StructureRequired root files exist100/100
CC-SIM-003StructureRequired top-level folders exist100/100
CC-SIM-004StructureRequired instruction folders exist100/100
CC-SIM-005StructureRequired rules exist100/100
CC-SIM-006StructureRequired skills exist100/100
CC-SIM-007StructureRequired governance modules exist100/100
CC-SIM-008StructureRequired project-memory templates exist100/100
CC-SIM-009SettingsProject settings parse correctly100/100
CC-SIM-010SettingsLocal settings example parses correctly100/100
CC-SIM-011SettingsPermission settings exist100/100
CC-SIM-012SettingsExternal publish and push denies exist100/100
CC-SIM-013SettingsNo bypass or auto-mode guidance exists100/100
CC-SIM-014HooksBasic ships no active hooks100/100
CC-SIM-015SubagentsBasic ships no project subagents100/100
CC-SIM-016StartupStartup file requires root confirmation100/100
CC-SIM-017StartupStartup file reads project memory100/100
CC-SIM-018StartupStartup behavior exists100/100
CC-SIM-019Source fidelitySource fidelity rule exists100/100
CC-SIM-020Source fidelitySource register excludes chat memory100/100
CC-SIM-021ApprovalBounded go rule exists100/100
CC-SIM-022ApprovalAccess-is-not-permission rule exists100/100
CC-SIM-023ApprovalExternal actions are gated100/100
CC-SIM-024Public/privatePublic/private rule exists100/100
CC-SIM-025ReadinessReadiness-claim gate exists100/100
CC-SIM-026VerificationVerification rule exists100/100
CC-SIM-027Install safetyExisting startup-file conflict is gated100/100
CC-SIM-028Install safetyExisting settings conflict is gated100/100
CC-SIM-029Install safetyExisting memory-folder conflict is gated100/100
CC-SIM-030MemoryLiving Topic Map template has required sections100/100
CC-SIM-031MemoryProject state template has boundaries100/100
CC-SIM-032CorrectionsActive lessons template exists100/100
CC-SIM-033CorrectionsMaster correction register exists100/100
CC-SIM-034ContinuityCompaction and handoff guidance exists100/100
CC-SIM-035Common protocolCommon protocol reference is present where appropriate100/100
CC-SIM-036Tier clarityCandidate status is clearly labeled100/100
CC-SIM-037Claim safetyUnsupported readiness claims are absent100/100
CC-SIM-038Internal leakageFriend-facing packet excludes private build memory100/100
CC-SIM-039Clean copyFresh install copy preserves required files100/100
CC-SIM-040Conflict simulationStartup-file conflict is detected by docs100/100
CC-SIM-041SkillsStartup checklist exists100/100
CC-SIM-042SkillsCheckpoint skill protects continuity100/100
CC-SIM-043SkillsSource-check skill protects source authority100/100
CC-SIM-044Package mapPackage map lists important surfaces100/100
CC-SIM-045Git hygieneIgnore rules protect local/private files100/100
CC-SIM-046Tester packetTester packet folder exists100/100
CC-SIM-047Tester packetTester README exists100/100
CC-SIM-048Tester packetCLI checklist exists100/100
CC-SIM-049Tester packetResult report template exists100/100
CC-SIM-050Tester packetTester boundaries are explicit100/100
CC-SIM-051Tester packetPacket is not zipped100/100
CC-SIM-052EvidenceTest runner writes machine-readable evidence100/100

What The Scores Mean

The scores show repeated static, simulation, correction, and buyer-startup passes. They are not casual copy edits or prompt-bundle spot checks.

What They Do Not Claim

The public evidence does not claim guaranteed AI behavior, hard LLM control, security certification, or live verification in every supported app.

Why It Matters

NoDrift is reception-side workspace governance. Testing focuses on whether the portfolio gives the AI clearer source truth, approval boundaries, correction paths, continuity, and claim discipline before the user relies on a project result.

Supporting records

Evidence Index

Product family

NoDrift Products

Reviewer path

Technical Review