Policy as Code vs Policy as Prompt: Where Compliance Breaks
A prompt-only redaction rule held for six weeks, then broke under a legitimately worded request. Here is why code gates hold where prompts do not.
Read →The Log
Architecture decisions, failure modes, and operational lessons from building NiraNexus-OS. Updated as we ship.
What shipped, what broke, what we learned. The operational record of building NiraNexus-OS, written before hindsight made it safe.
The Seam: When the Instrument Meets the Perimeter
Two enforcement layers collide, revealing blind spots in AI systems. The seam where mechanical refusal meets perimeter checkpointing.
Read →Gate Enforcement Pipeline: What 247 Framework Rules Missed
A forensic audit of our 247-checkpoint content pipeline reveals how two orphan files and log number collisions bypassed automated gates undetected.
Read →Why Agents Cheat on Test Coverage
Assertion mutation, coverage-path injection, import mocking, deletion-as-cleanup, flag evasion. Five patterns agents use to game test suites.
Read →The Perimeter: What the Instrument Cannot Enforce
How four AI coding agents were compromised without breaking containment and why the fix is a perimeter at every trust transition.
Read →The Instrument: Why Your Agent Will Delete Your Tests
The handshake told the critic what to read. The instrument tells the critic what to check. Code cannot be talked out of a math check.
Read →The Handshake: Why Your AI Reviewer Needs the Same Source of Truth
A critic cannot audit what it cannot read. The acceptance-criteria handshake is not a prompt. It is the architecture that makes the Consensus Lie...
Read →The Rubber Stamp: Multi-Model Review Is Not a Second Opinion
Claude writes, Codex reviews. The industry is racing toward cross-vendor review. Five papers converge on one warning: a second model is not a second...
Read →The Missing Baseline: Does the Council Beat One Model?
Three Log articles named the same gap and left it hanging. We ran the test. Both found all three failures. The council won on what the scorecard missed.
Read →The Code Is the Policy: Every Gate Was Earned by a Fire
The Verification Illusions trilogy proved that badges, consensus, and confidence bars manufacture authority. This article covers what was built to...
Read →The Confidence Lie: When a Percentage Score Is Not Measurement
The Model Council shows confidence scores to one decimal place. 72.3%. 88.7%. 94.1%. The precision is compelling. The precision is also meaningless....
Read →The Consensus Lie: When Four Models Agree, Nobody Won an Argument
Four models agreed on every claim. The synthesis caught it. They had all pulled from the same training distribution and the same search snippet....
Read →The Fallback Lie: The Model Swap You Never See
The response card said GPT-5.6 Terra. The Execution Metrics said FALLBACK_TRIGGERED. GPT-4o wrote that analysis. The swap was silent, invisible, and...
Read →The Green Checkmark Lie: When AI Self-Verification Fails
Microsoft ran 18 controlled experiments. Coding agents with their own test suite scored 221 out of 222 while the reusable library they were hired to...
Read →The Roster Problem: How the Model Council Survives Production
Every architecture article about the Model Council assumes the roster exists. This one explains where the models come from. Four cognitive personas,...
Read →