Most knowledge surfaces only show you what made the cut. A governed record is different: it shows what was claimed, what was refused, and why. Nothing quietly disappears. When something was left out, the record says so, and tells you who left it out and on what authority.
This page shows that idea at two distinct layers, drawn from two different example sources. The first is a publication layer, where a human editor explains, in plain prose, why something is not being treated as a signal. The second is an engine layer, where a machine refused an unsupported claim and left a mechanical trail you can check. The two layers are kept apart on purpose: one is editorial judgment, the other is engine output with proof attached.
What is this claim, and what is it built on?
The admitted example is a synthetic Counterpose-shaped signal. It is rendered here with the same GARPedia record components used at /records/SYN-CP-SIG-0001: the header carries its identity and status, the canonical body carries the position, and the provenance rail on the right shows what backs it — sources, citation spans, entities, and typed cross-references.
record SYN-CP-SIG-0001slug synthetic-counterpose-shaped-signal
Synthetic Demo: Counterpose-Shaped Signal With Edition Amendment
Counterpose-shaped record that exercises status badges, the edition selector, amendment narration, entity and source rails, and typed cross-references.
lifecycleAmended in placeanchorsAll anchors resolved
eligibilitypublication-eligible
computed at 2026-05-07T11:00:00Z
rendered at 2026-05-07T11:00:00Z
Position
This synthetic signal records, in renderer-bootstrap form, what a Counterpose-shaped publication looks like when GARPedia renders it: status badge, edition selector, amendment narration, entity and source rails, and typed cross-references with a canonical relationship-type union.1
Status badges and edition history
The status badge surfaces an Active, Amended, Superseded, Confirmed, Disputed, or Withdrawn state; the edition selector lists every edition with its release timestamp; and the amendment narration explains, edition by edition, what changed and why.2
Reader-facing narration is required for editions after the first. The first edition does not require narration because there is no prior state for the reader to compare against.3
Typed cross-references
Cross-references carry a relationship type drawn from a canonical union: responds_to, extends, contests, synthesizes, derives_from, supersedes, excerpted_from, and related. Open-text relationship strings produced by upstream profiles are normalized into that union by relationshipTypeNormalizer; unknown values fall back to related and preserve the original label so a reader still sees the upstream wording.4
Method transparency
The signal renders its supporting machinery in plain view: source identity and capture hash for every cited span, the rubric instruments referenced by the entity rail, and the discovery posture and control discharge notes recorded at render time. None of those surfaces is interactive in renderer-bootstrap mode.5
What was considered and left out, and why?
“Left out” means two different things, and a reader deserves to know which one they are looking at. Below, the same question is answered twice: once by a human editor, and once by a machine. They are not the same kind of answer.
publication layer
Editorial judgment, narrated
Human editorial judgment. No engine receipt is claimed for this layer.
Why This Is Not a Signal
A reader could reasonably expect this record to also flag a related, widely repeated assertion: that long-run growth in this domain settles at one fixed, universal number. The editor considered it and chose not to treat it as a signal. The reason is editorial, not mechanical: the assertion is stated confidently in the wider conversation, but the editor judged that it generalizes a single modeling convention into a rule it was never meant to be.
That judgment is a person’s reading of the material. It is offered as negative space — a deliberate note of what was left out and why — so that the absence is visible rather than silent. It is not a measurement, and it does not claim to be one.
Amendment log
Reader-facing narration of every edition after the first. Narration is required for editions after the first because the reader has a prior state to compare against; the first edition is always present but is not narrated.
2026-04-15T10:12:00Z — SYN-CP-ED-0001
Edition 1minor amendment
First edition; no prior state to compare against.
2026-04-22T16:30:00Z — SYN-CP-ED-0002
Edition 2substantive amendment
Substantive amendment: revised the discussion of cross-reference relationship types after a reviewer raised that the previous edition treated extends and synthesizes as interchangeable.
2026-05-05T09:45:00Z — SYN-CP-ED-0003
Edition 3minor amendmentcurrent
Minor amendment: tightened the amendment-narration section to specify that narration is required for editions after the first.
engine layer
Mechanical refusal, with proof attached
Here the “left out” is not a person’s reading. The Amnesiac spine routed a candidate claim, found it unsupported by the material in front of it, and refused it. The refusal is recorded with a reason, and — unlike the editorial note above — it is backed by the receipt in the next section.
engine layerWhat the engine considered and refused
The spine routed each candidate claim, declined to admit the ones it could not support, and recorded the reason. These refusals are mechanical: the engine emitted them, and they are carried into the receipt below.
claim_unsupportedrejected
recorded refusal reason
candidate refused under ContextPacket must_refuse_if_unsupported: The Gordon growth model requires a perpetual growth rate of exactly 2.3% for all NPV calculations.
How do we know no machine made it up?
This question belongs to the engine layer only. The editorial note above is a human judgment and is not backed by any of the artifacts below; only the engine’s refusal is. The receipt names the operation the spine ran and the artifact hashes it produced, so the refusal can be checked rather than trusted.
engine layerSovereignty receipt
The receipt is what lets a reader know no machine quietly made up the answer: it names the operation, the artifact hashes it produced, and the class of output it was allowed to emit.
A fictional publishing surface used in the GARPedia renderer demo to exercise Counterpose-style publisher rendering without referencing any real outlet.
publisher-one@synthetic.invalid
The editorial layer’s authority is the publisher’s own. What it left out is a disclosed editorial decision, made under the publisher’s custody. No engine proof stands behind it, and none is claimed.
engine layer — Amnesiac posture
amnesiac-spine-walkthrough.v0_1
scope
query_npv
egress class
render
The engine layer’s authority is the recorded operation itself. Its refusal posture is mechanical and bounded to the scope above, and the receipt in the previous section is the proof of what it did.
The honest limit
These are two layers illustrated on two examples today, not a single claim shown end to end. The editorial layer and the engine layer come from different sources, and this page deliberately keeps them apart rather than splicing them into one record. Full single-claim unification — taking one signal’s source material all the way through the Amnesiac spine so the editorial negative space and the mechanical refusal describe the same claim — is out of scope for this page.
That single-signal proof does now exist, but upstream rather than on this page. The arcs-amnesiac project added a real CP-33 signal walkthrough that carries one signal end to end, and it is rendered in arcs-amnesiac’s own source-owned combined viewer — arcs-amnesiac / walkthroughs/viewer/index.html, shown there alongside the synthetic NPV spine and the CP-33 signal datasets — where the canonical artifacts live. GARPedia is not the source of truth for that proof: this page only points to it and does not copy those artifacts in.