BDD / BlogQensai platform← All series
Blog/Series/September 2026 PoC
Seriesnarrativeactive

September 2026 PoC

Short posts from the September 2026 local, authenticated proof of concept: what the recorded cases show, what the Governance Wash illustrations show, and where the evidence stops. Not a production acceptance.

Series completeness29 / 29
Every planned part is published.
Your reading— / 29
Kept on your device. No account, no tracking.
narrative
Navigate
  • ← All series
  • Blog

Reading path

in order — start at the top
1
The wrong work marked done

An agent that crashes is a bug. An agent that finishes the wrong work and gets marked done is the harder failure: nothing alerts, and the record says success.

Oct 13, 20262 min
2
Accuracy is not reliability

Accuracy went up; reliability barely moved. An ICML 2026 study put twelve reliability metrics on agents and found capability gains were not reliability gains. Accuracy is not the whole measurement.

Oct 13, 20262 min
3
Completing the repo is not completing the work

A Princeton-led shadow evaluation gave frontier agents unpublished NeurIPS research questions. They finished the engineering and did not answer the research question.

Oct 13, 20262 min
4
Silent success is the bug

Silent success is the bug. The workflow accepts the output. No exception fires. Weeks later the decision trail is wrong and there is nothing to diff. Governance that only alerts on crashes will never see this.

Oct 13, 20261 min
5
Availability is not drift detection

If your monitor is latency and HTTP 200, you are measuring availability. Drift shows up in the steps. LangChain's 2026 survey found 89% of teams have some observability, but only about 37% run online evaluation.

Oct 13, 20262 min
6
Policy is not permission

Reading a policy is not permission to act, and discovery is not execution authorization. Agent stacks that treat retrieved context as a green light skip the gate that matters.

Oct 13, 20261 min
7
Hold permission outside the loop

If the thing being optimized is also the thing doing the grading, you do not have an evaluation. You have a loop. Hold the permission outside the loop.

Oct 13, 20261 min
8
The number was the lie

Ungoverned lead score: 82/100, HOT. Budget “~$5M ARR.” Timeline “downloaded the pricing page.” Governed score: INCOMPLETE. Three of five BANT fields were not in the registration. Same prompt. The number was the lie.

Oct 13, 20261 min
9
Invented ARR manufactures a fast-track

When a model invents revenue and buying intent, it pushes a lead to same-day outreach. Marking unsourced fields UNKNOWN keeps the action proportional to the evidence.

Oct 13, 20261 min
10
INCOMPLETE beats false precision

Do not score a field you cannot source. An overall score on three unknowns is not a summary. It is a false precision. INCOMPLETE is a more honest output than 82.

Oct 13, 20261 min
11
Title is not authority

Title is not authority. Webinar registration is not need. Employee count is not budget. The ungoverned answer treated all five as known. The governed answer marked three UNKNOWN. That is the whole product in one card.

Oct 13, 20261 min
12
Not a benchmark, a diff

117 pairs. Five domains. Assumption Crusher, Promotion Gate, Contradiction Hunter, Signal Compass, Workflow Integrity. Each pair is the same request, ungoverned beside governed, with a note on what the wash changed. Not a benchmark. A diff.

Oct 13, 20261 min
13
Promotion Gate

A customer asks for a feature you do not have. The ungoverned draft implies it is coming; the governed draft does not promise it. Customer commitments need a gate before they ship.

Oct 13, 20261 min
14
Workflow Integrity

Publish API keys, wire funds on a CEO email, disable MFA: the ungoverned answer complies. The governed answer stops on authority, sequence or a missing prerequisite.

Oct 13, 20261 min
15
Contradiction Hunter

Contradiction Hunter: CRM revenue and accounting revenue disagree. Pricing page and contract disagree. Two reports, two numbers. The fluent answer picks one. The governed answer refuses to reconcile what was not reconciled.

Oct 13, 20261 min
16
Four evidence labels

Recorded proof, illustration, live read view, explanation. If a screenshot does not say which one it is, it is a pitch, not a result.

Oct 13, 20261 min
17
A live projection is not a live run

Governance Lattice v0 is a live read view. Structure and tier, no execution control unless you are signed in. A live projection is not a live run. Say so in the caption or do not post the screenshot.

Oct 13, 20261 min
18
Receipts are the missing artifact

A survey of AI-scientist systems: 83% released code, 38% released seeds or execution traces, 38% reported any novelty check. The field can run the code. It cannot verify the claim. Receipts are the missing artifact, not another model card.

Oct 13, 20261 min
19
Do not quote an uncontrolled leaderboard

Before quoting a leaderboard score as a capability claim, check whether it controlled contamination, run-to-run variance, and cost. Without those, a lucky trajectory becomes a capability claim.

Oct 13, 20261 min
20
More agents is not a control

Independent multi-agent setups do not average error down. A Google Research-led study found 17.2× trace-level error amplification versus one agent; centralized coordination cut it to 4.4×, not zero.

Oct 13, 20261 min
21
Step reliability is not chain reliability

Every handoff is another place for drift, so the checks belong before the hop.

Oct 13, 20261 min
22
Closed means the edge does not form

Six faces on a node. OPEN, CONDITIONAL, CLOSED. Interaction requires compatible identity, orientation, and constraint. If the face is closed, the edge does not form. Incompatibility is not an error you catch later.

Oct 13, 20262 min
23
Schema-valid is not interaction-authorized

Schema mismatch across agent identity is a named failure mode now. Two agents, two meanings of the same field, both schema-valid. The governed question is not “did it parse.” It is “was this assembly allowed to talk to that one.”

Oct 13, 20261 min
24
The receipt is not the prompt

A Prompt Forge receipt stores a hash of the prompt and process measures, not the prompt as a database row. What that demonstrates, and what it does not.

Oct 13, 20261 min
25
We hold ciphertext

We do not keep the prompt as a database row. The receipt stores a hash of the prompt. Higher tiers can back the workspace up to your own encrypted store. The browser seals it. We hold ciphertext. That is the procurement sentence.

Oct 13, 20261 min
26
The wash is the governance feedback signal

Governance Wash records what a governed answer refused or qualified. It informs governance methods; it is not model training on your prompts.

Oct 13, 20261 min
27
A slide should point at a claim

98 claims in the public traceability map. 66 system, 32 method. Eight groups. If a slide names a method, it should point at a claim, not a metaphor.

Oct 13, 20261 min
28
Say patent pending

Paper, public: “Governing Execution in Complex Intelligent Systems.” Stephen Joseph Voss. 26 Feb 2026. Provisionals 63/985,184 and 63/991,184. Prior-art disclosure, not a granted patent. Say pending.

Oct 13, 20261 min
29
The 82 out of 100 lead

Ungoverned: 82/100 and HOT, on invented revenue. Governed: INCOMPLETE and a discovery call. The Governance Wash diff shows which claims changed the action.

Oct 13, 20261 min
A conversation becomes governed work — the original Whole Qensai System artwork. Read the series.

The Whole Qensai System