Observer cycle
scoped71 scheduled steps reached
Step completion is reported separately from evaluability. A return code is process evidence, not a verdict that the underlying check could see.
Public evidence snapshot · 20 August 2026 · 08:00–08:20 KST cycle
Claims should decay.
Evidence should be dated.
This is a deliberately small, capital-redacted extract from one completed observer cycle. It is not live telemetry, a security certificate, or a performance report. Process completion and evaluability are different claims; omissions, blind checks and failed evaluations remain part of the result.
Observed state
What this public snapshot can support
Observed governance surface
observed47 gates · 11,617 gate evaluations
31 gates were observed enforcing, 2 advisory and 14 not exercised in this cycle; 89.1% of the recorded gate evaluations were vetoes. These are not order or fill counts.
Authority audit coverage
open18 of 56 surfaces
Coverage is partial. Machine-initiated and unattributed arming events remain open evidence, including a human-only class.
Validation closure
open36 days · 0 effect-verified
Nine remediations were adopted. Zero is deliberate non-credit: no change is called effective without a preregistered observable and a sufficient post-change window.
Independent challenge
scoped122 targets · 13 survivors
Four findings were emitted. The most demanding preregistered challenge level did not pass in this snapshot.
Recovery proof
openNot verified
No current observer receipt supports a full clean-box, end-to-end restore claim across the fleet.
Published extract
A fixed artifact, not a live badge
The JSON extract freezes the public fields behind this page. Its checksum identifies the exact published file; it does not independently reproduce the private observer run or turn this disclosure into an attestation.
- Extract ID
- NB-PUBLIC-2026-08-20-AM
- Cycle
- 08:00–08:20 KST cycle
Post-change verification · 20 August 2026 · 18:00 KST cycle
Four preregistered predictions, four observed passes
This receipt was produced by the first normal cycle after the observer changes landed. It appends evidence to the morning snapshot; it does not rewrite that earlier record or turn a scoped replay into a fleet-wide guarantee.
P1 · pass
Taxonomy gap closed
The preregistered authority-surface finding resolved after 28 days on the first normal post-change cycle.
P2 · pass
Replay was non-blind
The instrumented system produced a dated MATCH receipt: 1,481 of 1,481 meaningful decisions and 201 of 201 vetoes matched.
P3 · pass
The guardrail held
Unclassified authority write sites fell from three to zero while the machine-initiated count stayed at eight and the checked population grew from 659 to 663.
P4 · pass
Protected history still accepted normal work
Two ordinary fast-forward updates landed on the remote main branch with deletion and non-fast-forward protection left active.
- Receipt ID
- NB-PUBLIC-2026-08-20-PM
- Closure state
- OBSERVING · 0 effect-verified
Policy
What the architecture says must happen. A design target, not proof that it happened.
Observed
What a dated run actually saw on the covered surfaces.
Coverage
What was not reached, not exercised, failed to evaluate, or remains unattributed.
Evidence archive
History is appended, not overwritten.
20 August 2026 · evening verification
NB-PUBLIC-2026-08-20-PM · four preregistered checks passed; campaign remains observing.
20 August 2026 · morning cycle
NB-PUBLIC-2026-08-20-AM · the site's first machine-readable public extract.
Publication boundary
Public evidence is allow-listed: no balances, account identifiers, secrets, strategy parameters, raw hostnames, or private configuration are published. Checksums pin the public extract and follow-up receipt, but independent reproduction of the private run and a longer soak window remain work in progress; these are dated disclosures, not a live attestation.