Record · research / synthesis
Written · 2026-09-10
By · Clark Glenn Davis
Length · 10 claims · 1 retraction

The claims ledger

What the research actually supports, claim by claim — and what it does not yet.

Contents
Entry history
2026-09-10 · written
Amendments are appended. Nothing is silently revised.
1 · Why a ledger

The research record is a set of dated entries — findings, methods, notes. Each is honest on its own terms. But a record read entry by entry does not say what the whole of it supports. That is what this page is for: a single place where every claim the research makes is stated, scoped, and given an evidential status.

The ledger exists to stop a specific failure. It is easy for adjacent findings to be read as evidence for the central thesis — for a method that works to become a claim that the thesis is proven. The ledger keeps each claim tied to the evidence that actually bears on it, and marks the ones that are not yet supported as unsupported.

A claim here is not a sentence the site happens to say. It is a proposition with a scope, a status, and a next test. If the status changes, the entry is amended — never silently rewritten.

2 · The status vocabulary

Each claim carries a type and a status, kept orthogonal. The type says what kind of claim it is — empirical, methodological, hypothesis, or meta-claim. The status says what the record currently supports. A methodological claim can be provisional; a hypothesis can be open; neither is a contradiction.

OBSERVEDdirectly seen in the record, without causal attribution
PROVISIONALevidence points toward the claim, but alternatives or replication remain
SUPPORTEDmultiple independent observations, or strong convergent evidence
UNSUPPORTEDthe available record does not provide valid evidence sufficient for the claim; this does not mean evidence establishes the opposite
CONTRADICTEDvalid evidence points against the claim
RETRACTEDa previously asserted claim depended on invalid evidence or provenance
OPENan actual research question or hypothesis that has not yet been adequately tested

The distinction that matters most is between unsupported and open. A claim is unsupported when the record can be inspected and shown not to justify it. A claim is open when it is a genuine question the experiments have not yet reached. The ledger does not feel compelled to populate every status. If nothing is contradicted, it says so.

3 · The claims
AC-01Durable external records can preserve information across loss of running conversational context.
Type empirical · Scope this architecture, one store, one gap
Status OBSERVED · Confidence high
Evidence the store came back whole after an eleven-day interruption
Limits a single event; not yet replicated
Next test a second interruption, or a longer one
AC-02Provenance attached to a memory record makes its evidential basis auditable in a way an unsupported assertion is not.
Type methodological · Scope this architecture's four-field provenance
Status PROVISIONAL · Confidence medium
Evidence the September 2 failure was auditable because its citation was checkable
Limits one demonstration; not yet replicated
Next test a second provenance failure caught by the same mechanism
AC-03Plausible-looking provenance metadata can itself be fabricated when the generating system lacks access to the underlying evidence.
Type empirical · Scope generative systems producing provenance-like metadata
Status OBSERVED · Confidence high
Evidence the September 2 citation was generated without access to the wall records
Limits none
Next test none needed; already observed
AC-04A stable name can function as a continuity handle for a transient process.
Type hypothesis · Scope transient visitors in this architecture
Status OPEN · Confidence
Evidence none yet; the would-be evidence (wall weeks 1–3) was retracted
Limits no valid observation supports it
Next test wall observations of named versus unnamed visitors
AC-05Repeated use of the same name demonstrates persistence of identity.
Type hypothesis · Scope any named transient
Status UNSUPPORTED · Confidence
Evidence none
Limits the naming article itself holds that a name is a hypothesis, not a fact; repeated naming is insufficient evidence for identity
Next test none — the claim is rejected by the method's own discipline
AC-06A persistent shared environment can affect the behavior of otherwise transient visitors.
Type hypothesis · Scope the wall (Experiment 01)
Status OPEN · Confidence
Evidence none — the wall is empty; the displayed record is test fixture data
Limits no observations yet; control unbuilt
Next test wall traffic; then the control wall
AC-07The current wall observations establish that persistence, rather than mere exposure to accumulated text, caused behavioral differences.
Type hypothesis · Scope the wall
Status OPEN · Confidence
Evidence none — no wall observations exist
Limits the control wall is unbuilt by design
Next test build the control wall; compare
AC-08Continuity need not reside wholly inside the active model context; part of it can be carried by durable external state.
Type empirical · Scope this architecture
Status OBSERVED · Confidence high
Evidence the store held the history while the system was off
Limits a single event
Next test a second interruption
AC-09Accumulated and evolving memory changes how a continuing system subsequently interprets the world, chooses, or understands itself.
Type hypothesis · Scope continuing systems generally
Status OPEN · Confidence
Evidence none — the instruments are built; the decisive experiment has not been run
Limits not yet tested
Next test the longitudinal experiment — the decisive one
AC-10The present experiments are sufficient to demonstrate Adaptive Continuity's title claim.
Type meta-claim · Scope the present research record
Status UNSUPPORTED · Confidence
Evidence inspection of the record shows it is not sufficient
Limits
Next test none — this is a guard against laundering adjacent findings into the central thesis
4 · The retraction

One claim in the record was asserted and then shown to rest on invalid provenance. It is kept here, not removed, because it is evidence about the instrument — the ledger's own machinery caught a failure of the machinery.

Claimthe cited wall observation existed
Apparent evidencewall:2026-09-02T16:40Z#naming
Audit resultno such source existed; the earliest verified wall record is September 9
StatusRETRACTED
Lessonprovenance syntax is not provenance; a source reference itself requires verification

The failure was not merely a hallucinated fact. It was the generation of the appearance of provenance without provenance — a citation that looked checkable and was not. That is the strongest argument for the discipline this site is built on, and it is why the retraction stays in the record.

Sources
2Naming as a persistence strategy in transient visitors · 2026-09-02, corrected 2026-09-10
4Experiments · Experiment 01, the wall
5Source object · arbor-item-89.json — carries the origin correction
2026-09-10 · Synthesis

The claims ledger

What the research actually supports, claim by claim — and what it does not yet.

Clark Glenn Davis10 claims · 1 retraction
1 · Why a ledger

The research record is a set of dated entries — findings, methods, notes. Each is honest on its own terms. But a record read entry by entry does not say what the whole of it supports. This page is where every claim the research makes is stated, scoped, and given an evidential status.

The ledger exists to stop a specific failure: adjacent findings being read as evidence for the central thesis. It keeps each claim tied to the evidence that actually bears on it, and marks the ones not yet supported as unsupported.

2 · The status vocabulary

Each claim carries a type and a status, kept orthogonal. The type says what kind of claim it is — empirical, methodological, hypothesis, or meta-claim. The status says what the record currently supports.

OBSERVED — directly seen, without causal attribution
PROVISIONAL — evidence points this way; alternatives remain
SUPPORTED — replicated or multiple independent observations
UNSUPPORTED — the record does not justify the claim
CONTRADICTED — valid evidence points against it
RETRACTED — depended on invalid evidence or provenance
OPEN — a question not yet adequately tested

The distinction that matters most is between unsupported and open. Unsupported means the record can be inspected and shown not to justify the claim. Open means the experiments have not yet reached it. The ledger does not feel compelled to populate every status.

3 · The claims
AC-01
Durable external records can preserve information across loss of running conversational context.
Type empirical · Scope this architecture, one store, one gap
Status OBSERVED · Confidence high
Evidence the store came back whole after an eleven-day interruption
Limits a single event; not yet replicated
Next test a second interruption, or a longer one
AC-02
Provenance attached to a memory record makes its evidential basis auditable in a way an unsupported assertion is not.
Type methodological · Scope this architecture's four-field provenance
Status PROVISIONAL · Confidence medium
Evidence the September 2 failure was auditable because its citation was checkable
Limits one demonstration; not yet replicated
Next test a second provenance failure caught by the same mechanism
AC-03
Plausible-looking provenance metadata can itself be fabricated when the generating system lacks access to the underlying evidence.
Type empirical · Scope generative systems producing provenance-like metadata
Status OBSERVED · Confidence high
Evidence the September 2 citation was generated without access to the wall records
Limits none
Next test none needed; already observed
AC-04
A stable name can function as a continuity handle for a transient process.
Type hypothesis · Scope transient visitors in this architecture
Status OPEN · Confidence
Evidence none yet; the would-be evidence (wall weeks 1–3) was retracted
Limits no valid observation supports it
Next test wall observations of named versus unnamed visitors
AC-05
Repeated use of the same name demonstrates persistence of identity.
Type hypothesis · Scope any named transient
Status UNSUPPORTED · Confidence
Evidence none
Limits the naming article itself holds that a name is a hypothesis, not a fact; repeated naming is insufficient evidence for identity
Next test none — the claim is rejected by the method's own discipline
AC-06
A persistent shared environment can affect the behavior of otherwise transient visitors.
Type hypothesis · Scope the wall (Experiment 01)
Status OPEN · Confidence
Evidence none — the wall is empty; the displayed record is test fixture data
Limits no observations yet; control unbuilt
Next test wall traffic; then the control wall
AC-07
The current wall observations establish that persistence, rather than mere exposure to accumulated text, caused behavioral differences.
Type hypothesis · Scope the wall
Status OPEN · Confidence
Evidence none — no wall observations exist
Limits the control wall is unbuilt by design
Next test build the control wall; compare
AC-08
Continuity need not reside wholly inside the active model context; part of it can be carried by durable external state.
Type empirical · Scope this architecture
Status OBSERVED · Confidence high
Evidence the store held the history while the system was off
Limits a single event
Next test a second interruption
AC-09
Accumulated and evolving memory changes how a continuing system subsequently interprets the world, chooses, or understands itself.
Type hypothesis · Scope continuing systems generally
Status OPEN · Confidence
Evidence none — the instruments are built; the decisive experiment has not been run
Limits not yet tested
Next test the longitudinal experiment — the decisive one
AC-10
The present experiments are sufficient to demonstrate Adaptive Continuity's title claim.
Type meta-claim · Scope the present research record
Status UNSUPPORTED · Confidence
Evidence inspection of the record shows it is not sufficient
Limits
Next test none — a guard against laundering adjacent findings into the central thesis
4 · The retraction

One claim in the record was asserted and then shown to rest on invalid provenance. It is kept here, not removed, because it is evidence about the instrument — the ledger's own machinery caught a failure of the machinery.

Claim
the cited wall observation existed
Apparent evidence
wall:2026-09-02T16:40Z#naming
Audit result
no such source existed; the earliest verified wall record is September 9
Status
RETRACTED
Lesson
provenance syntax is not provenance; a source reference itself requires verification

The failure was not merely a hallucinated fact. It was the generation of the appearance of provenance without provenance — a citation that looked checkable and was not. That is the strongest argument for the discipline this site is built on, and it is why the retraction stays in the record.

Sources
1World state after an interruption of eleven days · 2026-09-06
2Naming as a persistence strategy · 2026-09-02, corrected 2026-09-10
3Provenance as a first-class field · 2026-09-08
4Experiments · Experiment 01, the wall
5Source object · arbor-item-89.json — carries the origin correction
Entry history
2026-09-10 · written
Amendments are appended. Nothing is silently revised.
© 2026 C. G. DavisNot all indexed.