# Full Text: Golden Line: Toward What Matters

> Extracted from `golden_line_combined.pdf`

---

## Page 1

Golden Line: Toward What Matters
An Aspirational Thread for Long-Horizon Work
Daniel Ari Friedman
Active Inference Institute
daniel@activeinference.institute
ORCID: 0000-0001-6232-9096
DOI: 10.5281/zenodo.21754238
2026-07-18

## Page 2

Contents
1 Abstract 2
2 Introduction: the work needs a direction 3
3 Relationship to the line set 4
3.1 Note on the name . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4 Method: aspiration as a directional record 5
4.1 The registry entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.2 The horizon entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.3 The staged evaluator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.4 Conservative precedence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.5 Evidence and artifact boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.6 The descriptive analysis layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5 F ormalism: the evaluator and its invariants 7
5.1 Domain objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.2 The staged evaluator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.3 Propositions about the evaluator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.3.1 The completeness half of the rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5.3.2 The temporal half of the rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5.4 Structural invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.5 The report envelope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.6 Formalism-to-test bindings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
6 Scholarship: an aspiration is a direction, not a score 17
6.1 A translation, not a synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.2 The shape of an end . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.3 Capability and flourishing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.4 Attention, craft, and transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.5 Commons, power, and return . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.6 Why it must not become a score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.7 From theory to instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7 Evidence boundary and artifact chain 21
7.1 Hard constraints and soft choices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.2 Statuses as bounded claims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.3 Source to publication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
8 The Golden Line aspirations 24
8.1 The signal vocabulary in aggregate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
8.2 The reach of the nine horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
9 W orked records 29
10 Reading a batch: the descriptive layer at work 32
10.1 A worked batch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
10.2 What intake set aside . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
10.3 Vocabulary and reach as reading context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
11 Limits and safeguards 35
11.1 Limits of the descriptive analysis layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
12 Conclusion 36

## Page 3

1 Abstract
Golden Line is an aspirational thread: a compact, revisable way to say what a piece of work is trying to serve over a long horizon.
It asks what is worth reaching toward after Red Line has named the boundary and alongside Black Line’s operating discipline. It
is deliberately not a compliance score. A directional reading never certifies that work is good, safe, lawful, or complete.
The executable instrument is a small Python package with a versioned aspiration registry of nine entries — four founding aspirations
and five further ones — and a staged progress_report evaluator. Each aspiration pairs a plain-language thread and a concrete
horizon with two short lists: Markers, the observable signs of movement toward it, and Counter-signals, the observable signs of
movement away. An observer files a horizon entry, optionally dated, recording which of those were seen. The evaluator screens
and normalizes the entry, matches it against the registry, preserves a structured derivation, and returns one of four directional
readings: TOWARD, INQUIRY, DRIFTING, or NOT_OBSERVED. The report carries the registry digest and the temporal review context,
so a reading is never detached from the vocabulary and the date that produced it.
The paper develops the instrument in three registers. A formal method section states the domain objects, the staged evaluator,
and its decision rule as definitions and propositions that match the code exactly, each bound to the named test that would fail if the
code diverged, and records seven structural invariants, each proven by a planted-bad proof-of-detection test. A scholarship section
situates the founding aspirations — attention before output, usefulness beyond the author, repairable systems, and answerability to
human flourishing — in a lineage running from practical wisdom and goods internal to a practice through the capability approach,
reflective practice, repair, commons governance, technical power, and the hazards of turning an aspiration into a metric. Thirteen
code-derived figures render the registry, its signal vocabulary, the decision path, the staged pipeline, the evidence-state matrix,
the horizon bands, and a loop for the founding four. Six of the thirteen replay the evaluator rather than diagram it: one entry
ageing past the currentness boundary, the whole registry ageing at once across 126 dated readings, every marker subset of every
aspiration, the twelve fields of the returned record under six evidence conditions, a worked batch of six entries, and a precedence
panel in which complete markers plus one counter-signal still return DRIFTING for every aspiration.
Golden Line is standalone and non-redundant with its companions. Its vocabulary is horizon, thread, marker, counter-signal, and
direction. The absence of an observation is not a failure finding; it is an invitation to look again, or to leave the claim honestly
open.
2

## Page 4

2 Introduction: the work needs a direction
Every method can become technically competent while losing contact with the reason it exists. A test suite stays green, a pipeline
keeps shipping, and yet the question of what the work is for quietly falls out of view. Golden Line names that danger without
pretending to dissolve it with a single principle. It provides a small, revisable set of directions that can orient a decision, a
collaboration, or a long project, and a way to ask, at a stated horizon, whether the work is visibly moving toward them.
The idea has many ancestors, and the instrument inherits from all of them without settling their disagreements. Aristotle treats
practical judgment as deliberation about a good life rather than the mechanical application of a rule, and makes an end (telos)
the thing that gives an activity its point [ Aristotle, 350 BCE ]. Confucian teaching connects learning to the cultivation of conduct
within relationships, so that skill and character grow together rather than apart [ Confucius, 500 BCE ]. Ibn Khaldun’s account of
cooperation, power, and social change reminds us that purposes are carried by institutions and shared cohesion, not by individual
intention alone [Khaldun, 1377]. Closer to the present, MacIntyre’s distinction between goods internal to a practice and its external
rewards helps explain why useful work must remain more than a successful artifact [ MacIntyre, 1981]. The scholarship section
extends this lineage through capabilities, reflective practice, repair, commons governance, and the politics of technical systems.
These references are prompts for comparison and calibration, not authorities the registry claims to reconcile.
Golden Line’s role is deliberately limited, and the limit is the point. It cannot authorize work that Red Line refuses. It cannot
certify that work was done well, which is Black Line’s job. It cannot fill in what White Line marks as absent. The next section
places each of those instruments; here it is enough to say that Golden Line absorbs none of their work. What it adds is a direction
held lightly, stated in the open, and revisable when the evidence or the values change.
The paper proceeds from prose to formalism to lineage: the method section describes the staged evaluator in words, the formal
section restates it as definitions and propositions faithful to the code, and the scholarship section grounds the founding aspirations
in their sources. The remaining sections walk the nine-entry registry, worked records, the evidence boundary, and honest limits.
The manuscript is only half of it. Golden Line is operated day to day through the daf-golden-line skill in DAF’s private daf-skills
toolchain: this paper states what the instrument means and why, and the skill is how a horizon entry is actually filed.
The paper is organised as follows. Section sec. 4 defines the method and evidence protocol. Section sec. 5 states the evaluator
formally. Section sec. 8 presents the four aspirations. Section sec. 9 walks through worked examples, and Section sec. 11 closes
with limits and epistemic boundaries.
3

## Page 5

3 Relationship to the line set
Golden Line is the aspirational work in the four-line set, whose declaration lives in the companion work line_set. Red Line
is the personal security boundary and explicit No document. Black Line is positive operating discipline. White Line records
absence, restraint, and negative space. Golden Line does not copy their registries or evaluators, and a directional finding is never
a compliance verdict.
A fifth work, line_set, is a thin reader that declares the set and checks that no two lines gave the same spelling to different things;
it adds no substantive instrument, and Golden Line does not import, depend on, or defer to it.
3.1 Note on the name
The name is not incidental. Golden Line openly echoes the yellowing stage of the alchemical magnum opus, citrinitas, the dawning
that classical schemes place between the whitening and the reddening. The echo is meant in Carl Jung’s symbolic-psychological
register, where the stages of the opus are read as a map of individuation, and in that register only: it is a metaphor for a
practice turning toward what it is for, never an empirical, mystical, or causal claim about matter, minds, or this instrument. The
working order of the four-line set refuse, then method, then aspire, then absence, is functional and chosen for how the instruments
are actually used, and deliberately does not reenact the opus’s canonical sequence (nigredo → albedo → citrinitas → rubedo).
Citrinitas here simply marks Golden Line’s place in the set: the yellowing where a disciplined practice lifts its attention from how
the work is done to what the work is worth. The grounding citation is Jung’s reading of the opus stages as figures for individuation
rather than as laboratory chemistry [ Jung, 1953], and it is carried here rather than deferred, so this paper does not depend on
another document to say what its own name means. The set-level version of the framing — the two orders, the caveats, and the
non-overlap contract — is declared in the companion work line_set; this paper points there for the set, not for its own citation.
4

## Page 6

4 Method: aspiration as a directional record
4.1 The registry entry
An aspiration has six fields: an identifier, a plain-language title, a thread that explains its direction, a horizon at which the
direction becomes visible, and two short lists. Markers are observable signs that the work is moving in the declared direction.
Counter-signals are observable signs that it is drifting away from it. Neither list is exhaustive, and neither is a questionnaire;
they are the small, concrete anchors that let a direction be discussed rather than merely admired. The registry ships nine such
entries: four founding aspirations and five further ones that fill out the long-horizon picture.
4.2 The horizon entry
An observer records movement by filing a horizon entry : the aspiration identifier, the set of markers actually observed, the set
of counter-signals actually observed, a free note, and an optional ISO observation date. The entry describes a record of work at a
moment, not a person and not a project as a whole.
4.3 The staged evaluator
progress_report reads a batch of horizon entries against the versioned registry in three stages, and every finding it returns carries
a full reasons trail plus structured derivation fields so no reading arrives unexplained.
1. Intake screening. Each incoming entry is checked for the expected record shape and against the known aspiration identifiers.
An entry for an unknown aspiration is set aside with a note; a second valid entry for an aspiration already seen is set aside,
and the first entry stands. JSON-like token lists are normalized to sets. Malformed or hostile input is recorded, never raised.
A sloppy record cannot crash the report or vanish silently.
2. Matching. For each aspiration, the observed markers are intersected with the declared markers, and the observed counter-
signals with the declared counter-signals. Tokens the observer supplied that the registry never declared are ignored and
noted; they cannot smuggle in a status.
3. Decision. With optional temporal review folded in, the matched sets and temporal quality determine one of four directional
readings. There is no numeric score or aggregate grade.
The four readings are:
• TOWARD means every declared marker is observed, no counter-signal is recorded, and (if temporal review is enabled) currentness
is auditable: the observation date is present, parseable, and within the declared window.
• INQUIRY means the aspiration is named but the evidence is partial, empty, or stale, so the direction remains honestly open.
• DRIFTING means a declared counter-signal is present. It describes the record, not the person or project as a whole.
• NOT_OBSERVED means no valid horizon entry was admitted for that aspiration. An entry can be present in the input but fail
intake because it is malformed, unknown, or a duplicate set aside by the first-valid-entry rule.
The finding also preserves the note, observed date, declared counter-signals, and undeclared tokens that were ignored. This makes
the status machine-readable without requiring a downstream reader to reverse-engineer human-readable reason strings.
4.4 Conservative precedence
The precedence is intentionally conservative. A counter-signal is surfaced before any markers are counted, and staleness never erases
a recorded drift: a declared counter-signal yields DRIFTING even when the same observation is old. Partial, stale, or currentness-
un-auditable positive evidence reverts to INQUIRY rather than being rounded up to TOWARD. This mirrors a satisficing stance: the
evaluator looks for good enough and current evidence of direction and refuses to over-read thin signals [ Simon, 1956].
The result is a directional report. It is not an evaluator of worth, a certification, or a substitute for the Red Line, Black Line, or
White Line instruments.
4.5 Evidence and artifact boundary
The machine has four layers: the versioned registry declares the vocabulary; an admitted entry records a local observation; the
evaluator derives a bounded reading; and the generated registries and figures preserve what source was used. No layer is allowed to
add a claim that the earlier layer did not carry. The report’s registry version, registry digest, review date, and staleness threshold
therefore travel with the findings. The local artifact gate checks that the generated JSON, SVG, PNG, and manuscript figure
labels still agree before a sibling template render is attempted.
5

## Page 7

4.6 The descriptive analysis layer
A small analysis module sits beside the evaluator and is deliberately weaker than it: every helper is pure, deterministic, and
read-only, and none can change what a reading means. Four helpers are provided.
• signal_inventory tallies the declared markers and counter-signals across a registry — for the shipped registry, 18 markers
and 9 counter-signals, with every token distinct — describing the vocabulary the evaluator can match, never its fulfilment.
• horizon_distribution groups aspirations into four declared temporal-reach bands (immediate, recurring cycle, at handoff,
open-ended). The band map lives in the analysis layer, not the registry contract, and an unclassified horizon raises an error
so a registry change must revisit the map deliberately.
• temporal_currentness_sweep replays one fully-marked entry through the public progress_report across a range of
observation ages, exposing the exclusive staleness boundary (current at age 90, stale at 91 for a 90-day window) as an
observed trajectory rather than a claim.
• report_overview regroups an existing report by status and totals its ignored tokens, temporal flags, and intake notes, so a
batch can be characterized without re-parsing prose reasons.
The analysis layer is also where the project’s experiment plan is grounded. The repository ships a domain_profile.yaml naming
the validation gates: structural invariants, evidence grounding, artifact chain, render validation, and publication readiness. It
also ships an experiment_plan.yaml whose three conditions — the source-registry baseline, the malformed-input guard, and the
temporal-currentness guard — map onto the evaluator’s intake stage and temporal review. The plan’s expected figures are the
ones the deterministic builder produces: five through this analysis layer’s helpers (the temporal-currentness sweep, the currentness
lattice, the signal inventory, the horizon bands, and the batch reading overview) and eight straight from the registry, the evaluator,
and the status contracts. The protocol compares reproducible contract outcomes only. None of these summaries is evidence that
any aspiration is true, and none is a score.
6

## Page 8

5 Formalism: the evaluator and its invariants
The formalism restates the evaluator the method section describes; every result below tracks the code that implements it. This
section states the implemented semantics. Every result below describes what the code in golden_line actually does, and the
reasons and transitions named here are the ones the evaluator produces. The formalism describes the instrument; it does not
extend or idealize it. Numbering is assigned by the renderer in document order, so no number is written in the source and none
can go stale.
5.1 Domain objects
Definition 1 (Aspiration). An aspiration 𝑎 is the six-tuple
𝑎 = ( id, title, thread, horizon, 𝑀 , 𝐶),
where id , title, thread, horizon are text fields and 𝑀 (Markers) and 𝐶 (Counter-signals) are finite sequences of observable tokens.
Markers are signs of movement toward the aspiration; Counter-signals are signs of movement away from it.
Definition 2 (Registry). The registry 𝑅 = ⟨𝑎 1, … , 𝑎𝑛⟩ is an ordered tuple of aspirations with 𝑛 = 9: four founding aspirations
(attention before output, usefulness beyond the author, repairable systems, answerability to human flourishing) followed by five
further ones (durable understanding, teachable craft, honest uncertainty, unhurried questions, improvements returned to the
commons). Write ids (𝑅) for the set of identifiers appearing in 𝑅.
Definition 3 (Horizon entry). A horizon entry is 𝑒 = ( aid, 𝑂, 𝐾, note, dobs), where aid names an aspiration, 𝑂 is the set of
observed markers, 𝐾 is the set of observed counter-signals, note is free text, and d obs is an optional ISO date.
Definition 4 (Status codomain). The directional readings form the four-element set
Σ = { TOWARD, INQUIRY, DRIFTING, NOT_OBSERVED},
copied verbatim from the HorizonStatus enumeration. These are the only statuses the evaluator can emit.
Definition 5 (Finding). A finding is
𝑓 = ( aid, 𝜎, reasons, observed, unmet, countered, ign𝑀 , ign𝐶, note, dobs, stale, date_issue)
with 𝜎 ∈ Σ (Definition 4 ). The structured fields list the declared markers observed and unmet, the declared counter-signals, the
undeclared marker tokens ign 𝑀 and counter-signal tokens ign 𝐶 that were ignored, the original note and date, and two temporal
quality flags. Ignored markers and ignored counter-signals are kept apart rather than pooled, because a reader checking why a
token did nothing needs to know which vocabulary it failed to match. Human-readable reasons remain a parallel explanation, not
the only source of derivation.
5.2 The staged evaluator
Definition 6 (Report function). The evaluator has signature
progress_report ∶ (𝐸, 𝑅, as_of, 𝜏 ) ⟶ HorizonReport,
where 𝐸 is a batch of horizon entries, 𝑅 defaults to the registry of Definition 2 , as_of is an optional review date, and 𝜏 (stale_
after_days) is an optional non-negative integer staleness threshold; as_of and 𝜏 are keyword-only. It runs three stages: intake
screening, matching, and decision. The returned report also identifies the registry version/digest and the review context.
Definition 7 (Intake screening). Screening builds an accepted map 𝐴 ∶ ids(𝑅) ⇀ 𝐸 by a first-wins rule. Iterating 𝐸, a
record with the wrong type or malformed fields is set aside with an intake note; an entry whose aid ∉ ids(𝑅) is set aside with an
unknown-id note; an entry whose aid is already in 𝐴 is set aside with a duplicate note and the first valid entry stands; otherwise
the entry is admitted. Iterable text-token fields are normalized to sets. Screening never raises on malformed or unexpected input.
It is a pure function of the pair (𝐸, ids(𝑅)): records are tested in a fixed order — shape, then registry membership, then prior
admission — and iteration follows the order of 𝐸 with indices counted from 1 in the intake notes.
Definition 8 (Matching). For aspiration 𝑎 (Definition 1 ) with admitted entry 𝑒 = 𝐴(𝑎. id) (Definition 3 ), define
observed(𝑎, 𝑒) = 𝑂 ∩ 𝑀 , unmet(𝑎, 𝑒) = 𝑀 ∖ 𝑂, countered(𝑎, 𝑒) = 𝐾 ∩ 𝐶.
Undeclared tokens 𝑂 ∖ 𝑀 and 𝐾 ∖ 𝐶 are ignored and recorded in the reasons; they can never determine a status.
Definition 9 (T emporal review). Given review date 𝑑 and threshold 𝜏 , temporal review is enabled exactly when 𝜏 is set. If
𝜏 is unset, temporal metadata does not affect the status. If 𝜏 is set, a missing or unparseable d obs sets date_issue to true: the
7

## Page 9

record is not fatal, but it cannot support a currentness claim. With a parseable date, stale (𝑒, 𝑑, 𝜏 ) holds iff d obs > 𝑑 (a future
observation) or 𝑑 − dobs > 𝜏 days.
Definition 10 (Decision rule). The status 𝜎(𝑎, 𝑒) is determined by the first matching clause, in order:
1. if 𝑒 = ⊥ (no admitted entry): NOT_OBSERVED;
2. else if countered (𝑎, 𝑒) ≠ ∅ : DRIFTING;
3. else if observed (𝑎, 𝑒) = ∅ : INQUIRY;
4. else if unmet (𝑎, 𝑒) ≠ ∅ : INQUIRY;
5. else if stale (𝑒, 𝑑, 𝜏 ) or date_issue: INQUIRY;
6. otherwise: TOWARD.
5.3 Propositions about the evaluator
Each proposition below follows from Definition 7 through Definition 10 and is exercised by the package’s test suite; the binding
table at the end of this section names the verifying test for every result. Each proposition is a statement about code behavior —
never about the world, safety, or persons.
Proposition 1 (T otality). For every aspiration 𝑎 ∈ 𝑅 the evaluator emits exactly one finding ( Definition 5), so a report contains
𝑛 = 9 findings, one per registry entry, whether or not any entry was filed. The report’s counts therefore partition the findings
across Σ.
Proposition 2 (Counter-signal precedence). If a declared counter-signal is recorded, meaning countered (𝑎, 𝑒) ≠ ∅ (Definition
8), the status is DRIFTING, regardless of how many markers were observed and regardless of staleness. Clause 2 precedes clauses
3–6, so a positive signal can never launder a recorded drift.
Proposition 3 (Exactness and currency of TOW ARD). 𝜎(𝑎, 𝑒) = TOWARD iff observed (𝑎, 𝑒) = 𝑀 (every declared marker
seen), countered (𝑎, 𝑒) = ∅ (Definition 8), and neither stale (𝑒, 𝑑, 𝜏 ) nor date_issue holds. Any missing marker, any counter-signal,
stale observation, or unparseable reviewed date reverts the reading to INQUIRY or DRIFTING. Counter-signal precedence is as stated
in Proposition 2 . There is no partial credit: an entry recording every marker but one reads exactly as an entry recording none.
Proposition 4 (T emporal uncertainty reopens, never condemns). A fully-marked but stale or date-un-auditable observa-
tion with no counter-signal yields INQUIRY (clause 5), not DRIFTING. Age or malformed temporal metadata reopens a question; it
does not manufacture drift.
Proposition 5 (Absence is not negation). NOT_OBSERVED arises solely from the absence of an admitted entry (clause 1). The
admission rule of Proposition 6 determines which entries are admitted; when none is, the reading is NOT_OBSERVED regardless of
what records may have been submitted. It asserts nothing about whether the aspiration is being served; it reports only that no
valid record was available to this report. The intake notes preserve why submitted records were set aside when that distinction
matters.
Proposition 6 (First valid entry stands; the reading is deterministic). For each identifier, screening admits the first valid
entry bearing it in the order of 𝐸: the first valid entry bearing an identifier stands, and every later entry with that identifier is set
aside with a duplicate note. For an identifier with no valid entry, the reading is NOT_OBSERVED (Proposition 5 ). Exchanging the
positions of two entries that share an identifier can therefore change the finding, but nothing else about the call can: replaying an
identical batch with identical as_of and 𝜏 reproduces the identical report, digest for digest. Determinism is a property of the code
path; it does not make the underlying records true.
Proposition 7 (Intake screening determinism). The intake screening function ( Definition 7 ) is a pure function of the pair
(𝐸, ids(𝑅)): records are tested in a fixed order — shape, then registry membership, then prior admission — and iteration follows the
order of 𝐸 with indices counted from 1 in the intake notes. For the identical batch 𝐸 and the identical known ids ids (𝑅), screening
produces the identical accepted map 𝐴 and the identical intake notes every time. Malformed input is set aside deterministically:
a record with the wrong type or malformed fields always gets a malformed note, an entry whose aid ∉ ids(𝑅) always gets an
unknown-id note, and an entry whose aid is already in 𝐴 always gets a duplicate note. Screening never raises on malformed or
unexpected input — it is a pure computation, and its determinism is separate from the first-wins semantics of Proposition 6 ,
which covers which entry stands when multiple valid entries share an identifier. Intake screening determinism covers the screening
function’s behavior on all input, including malformed records.
Proposition 8 (The currentness boundary is exclusive). Staleness uses the strict comparison 𝑑 − dobs > 𝜏 of Definition 9 ,
so a fully-marked, counter-signal-free entry with 𝜏 = 90 reads TOWARD at age exactly 90 days and reverts from age 91 days: age 𝜏
is the last current age and 𝜏 + 1 the first stale one. The threshold is the caller’s review-cadence choice; nothing in the evaluator
endorses any particular number of days as a natural constant.
The ordered clauses and the four readings they terminate in are drawn from the HorizonStatus enumeration — the decision rule’s
codomain, not the registry — in fig. 1.
8

## Page 10

Figure 1: The evaluator’s ordered decision rule. Counter-signal precedence is explicit; TOW ARD requires complete markers and
auditable currentness when temporal review is enabled. The path reads a record; it is not a compliance verdict.
9

## Page 11

5.3.1 The completeness half of the rule
The exactness clause of Proposition 3 is the one a reader is most likely to soften into “mostly there” . fig. 2 refuses that reading by
running it: every subset of every aspiration’s declared markers is filed as an entry and evaluated, and the panel prints the status
the evaluator returned alongside the number of markers left unmet. Only the complete subset reads TOWARD; a single missing
marker holds the direction open at INQUIRY, exactly as an empty entry does. The panel characterizes the clause order, not any
observed practice.
Figure 2: Every observed-marker subset for every aspiration, filed as an entry and evaluated. Each cell prints the status
progress_report returned and the count of markers still unmet, so the grid reads without colour; only the complete subset reaches
TOW ARD, and one missing marker reads exactly as none. Marker completeness is a property of the record, never a measure of
the work.
5.3.2 The temporal half of the rule
Proposition 4 has a picture. The temporal_currentness_sweep analysis helper replays one fully-marked, counter-signal-free
entry through the real evaluator at a range of observation ages with 𝜏 = 90 : the reading holds TOWARD through age 90 exactly (the
boundary is exclusive), reverts to INQUIRY from age 91, and a future-dated observation likewise reads INQUIRY. Every cell in fig. 3
is an actual progress_report result, not an illustration of one; the sweep observes the evaluator and never re-implements it.
Proposition 9 (Sweep delegation and codomain). The sweep computes no status of its own: its only status-producing call is
progress_report (Definition 6), and each SweepPoint copies that call’s finding. This is an architectural property, not an empirical
10

## Page 12

Figure 3: A fully-marked entry replayed through the real evaluator at 14 observation ages with stale_after_days = 90. Every
cell names the status progress_report actually returned, so nothing here is carried by colour alone: TOW ARD through age 90,
INQUIRY from age 91 and for future-dated observations. Age reopens the question; it never manufactures drift.
11

## Page 13

one — comparing a sweep point against a second progress_report call would compare the same code path with itself, so the
verifying test instead asserts that analysis.py constructs no HorizonStatus anywhere. Because sweep entries are fully marked,
counter-signal-free, and bear a known identifier, only clauses 5 and 6 of Definition 10 can fire: DRIFTING and NOT_OBSERVED are
unreachable in a sweep; that half is falsifiable and is tested by enumeration. The sweep characterizes the instrument’s temporal
behavior; it does not describe any observed practice.
Proposition 8 states the boundary and the sweep shows it for one registry entry. Whether the boundary belongs to the evaluator
rather than to that one entry is a separate question, and one that a single row cannot answer.
Proposition 10 (The currentness boundary is uniform across the registry). Replaying the sweep ( Proposition 9 ) for
every 𝑎 ∈ 𝑅 at the same review date and threshold yields the same TOWARD → INQUIRY transition age for every aspiration, because
Definition 9 reads only d obs, 𝑑, and 𝜏 — no field of 𝑎 enters the staleness comparison. The sweep confirms that the temporal-inquiry
behaviour stated in Proposition 4 and the exclusive boundary stated in Proposition 8 are properties of the evaluator’s code path,
not of any particular aspiration. The lattice in fig. 4 is that replay: 9 × 14 = 126executed progress_report calls, with each row’s
own transition age printed beside it and a count of rows that differ. Uniformity here is a fact about the code path; it says nothing
about how any aspiration is actually served.
Figure 4: The same currentness replay run across the whole registry: 9 aspiration rows by 14 observation ages, 126 executed
progress_report calls at stale_after_days = 90. Filled cells are TOW ARD and outlined cells are INQUIRY, so the grid reads
without colour; the FLIPS AT column gives each row’s own transition age and the footer counts rows that differ. Uniform ageing
is a fact about the evaluator’s code path, not evidence that any aspiration is being served.
12

## Page 14

5.4 Structural invariants
The evaluator reads a registry; a malformed registry would make its readings meaningless. Seven pure-compute structural checks
validate the shape of the registry, independent of any horizon entry. Let 𝐼1, … , 𝐼7 be their predicates; registry_sound(R) returns
⋀
7
𝑘=1 𝐼𝑘(𝑅).
• 𝐼1 distinct identifiers : no two aspirations share an id, so findings are unambiguous.
• 𝐼2 fields populated : id, title, thread, and horizon all carry content.
• 𝐼3 reachable statuses: each aspiration has at least one marker and at least one counter-signal, so both TOWARD and DRIFTING
are reachable for it.
• 𝐼4 signal text : every marker and counter-signal is a non-blank string.
• 𝐼5 signal uniqueness : no marker or counter-signal repeats within one declaration.
• 𝐼6 signal disjointness : 𝑀 ∩ 𝐶 = ∅ within each aspiration, so no token reads as movement toward and away at once.
• 𝐼7 digest stability : the canonical registry serializes, and its SHA-256 digest is independent of registry order.
Proposition 11 (Digest order-independence). canonical_registry serializes aspirations sorted by identifier, so
registry_digest(𝑅) = registry_digest(𝜋(𝑅)) for every permutation 𝜋 of 𝑅, and the digest is exactly 64 lowercase hexadecimal
characters (SHA-256). Two readers comparing digests are comparing registry content, never the order in which they happened to
list it — and a matching digest attests only sameness of content, not soundness or merit.
Proposition 12 (Proof of detection). For each invariant 𝐼𝑘, the test suite asserts both that 𝐼𝑘 passes on the real registry and
that 𝐼𝑘 fails on a deliberately planted-bad registry constructed to violate exactly that check. A green check that had never seen a
bad input would not count as evidence, so each invariant is paired with the counter-example that gives it meaning.
These digests and checks are review and drift-detection instruments for humans comparing registry revisions. They carry no safety,
warranty, or attestation semantics of any kind. The detection proof of Proposition 12 establishes that the structural invariants can
fail — each has a planted counter-example — and therefore a green check is a positive result, not a silent absence. Proposition 11
gives the digest its order-independence; detection gives the invariants their falsifiability.
5.5 The report envelope
A reader holding reports from several independent instruments needs one uniform way to say “this instrument, about this subject,
at this review moment, said this — and here is the pointer to its complete native report. ” The envelope is that data contract and
nothing more. For this instrument the usual worry — that a selected status becomes a safe projection mistaken for the whole
state — takes a specific form: Golden Line has no single overall verdict, so the only honest transportable status is the complete
ordered vector of per-aspiration readings. The envelope carries exactly that vector, in report order, and invents no summary above
it; compressing nine directional readings into one word would manufacture precisely the aggregate virtue score this instrument
refuses everywhere else.
Definition 11 (Report envelope). The report envelope is the frozen record 𝑣 = ( schema_version, line_id, subject_id, review_date, registry_version, registry_digest, native_status, report_ref, source_snapshot_refs, scope_and_nonclaims)
with exactly those ten fields, in order, exported under the schema string line.report-envelope/1.0 . native_status is the
complete ordered sequence of (aspiration_id, status) pairs from the report’s findings — this line’s own vocabulary, one pair
per registry aspiration ( Proposition 1 ), never a summary. report_ref is the SHA-256 of the complete canonical report, so the
envelope points at the full derivation — reasons trails, matched and ignored tokens, temporal flags, intake notes — rather than
copying or restating any of it. scope_and_nonclaims carries the instrument’s transportable non-claims inside the record itself, so
a stored envelope cannot quietly outgrow what the instrument was allowed to say. Sibling instruments export the same shape by
publishing the same schema string, never by importing one another, and envelopes from different lines must not be compared,
ranked, averaged, or merged on native_status.
Proposition 13 (The envelope points, never reinterprets). For every report 𝑟, report_envelope(r) satisfies envelope
_matches_report(envelope, r) : the digest pointer, the review date, the registry version and digest ( Proposition 11 ), and the
per-aspiration readings all agree with the report they were exported from, and editing any of them afterwards makes the check
return false. The envelope ( Definition 11 ) adds no field the report does not determine except the caller-supplied subject_id and
source_snapshot_refs, which the evaluator stores and does not verify. A matching envelope attests that the pair was archived
unedited; it says nothing about the truth of the report or the merit of anything the report read.
5.6 Formalism-to-test bindings
Every definition and proposition above is verified by named tests in the package’s suite; the tables below bind each result to the
test that would fail if the code stopped satisfying it. Each row is keyed on the block’s label, not on its number, and the label
renders as the number the reader sees, so inserting a result renumbers the prose and the table together and can never split them.
13

## Page 15

Two binding tests police the tables. tests/test_formalism_bindings.py::test_binding_tables_bind_every_declared_blo
ck fails if the set of row labels stops matching the set of labels declared in this section, and tests/test_formalism_bindings.p
y::test_every_binding_row_names_an_existing_test fails per row if any row’s verifying-test cell names no test or names one
that does not exist. Neither checks that a named test is a good test, only that every declared block is bound to one that exists.
The boundary column restates what each result does not claim.
Definition Statement essence Verifying test Boundary
Definition 1 six fields, two of them token
sequences
tests/test_formalism_bind
ings.py::test_aspiration_
tuple_matches_the_datacla
ss
names the fields, not what a
marker is worth
Definition 2 nine entries, four founding
then five further
tests/test_formalism_bind
ings.py::test_registry_de
finition_matches_the_sour
ce_tuple
a versioned design choice, not
a canon
Definition 3 five fields, the observation
date optional
tests/test_formalism_bind
ings.py::test_horizon_ent
ry_tuple_matches_the_data
class
records an observation, not a
person
Definition 4 exactly the four
HorizonStatus values, in
enum order
tests/test_formalism_bind
ings.py::test_status_codo
main_matches_the_enumerat
ion
four readings, not a scale from
bad to good
Definition 5 twelve fields, ignored markers
kept apart from ignored
counter-signals
tests/test_formalism_bind
ings.py::test_finding_tup
le_matches_the_dataclass
exposes a derivation, not a
justification
Definition 6 four parameters, two
keyword-only; three named
stages
tests/test_formalism_bind
ings.py::test_report_func
tion_signature_matches_ma
nuscript
a call shape, not a guarantee
about inputs
Definition 7 shape, then membership, then
prior admission; notes
indexed from 1; never raises
tests/test_formalism_bind
ings.py::test_intake_scre
ening_tests_records_in_th
e_stated_order, tests/test
_formalism_bindings.py::t
est_intake_screening_neve
r_raises_on_hostile_input,
tests/test_formalism_bind
ings.py::test_intake_firs
t_wins_and_replay_determi
nism_match_manuscript
screening judges records,
never their authors
Definition 8 the three set operations, and
undeclared tokens discarded
tests/test_formalism_bind
ings.py::test_matching_se
t_operations_match_manusc
ript
set membership, not
sufficiency of evidence
Definition 9 enabled iff 𝜏 is set; missing or
unparseable dates flag rather
than fail
tests/test_formalism_bind
ings.py::test_temporal_re
view_rule_matches_manuscr
ipt
a data-quality window, not a
decay law
Definition 10 six clauses, first match wins tests/test_formalism_bind
ings.py::test_decision_cl
auses_fire_in_the_stated_
order
clause order, not moral order
14

## Page 16

Proposition Statement essence Verifying test Boundary
Proposition 1 one finding per registry
aspiration; counts partition
𝑛 = 9
tests/test_progress.py::t
est_counts_summary, tests/
test_formalism_bindings.p
y::test_totality_count_ma
tches_manuscript
report shape, not coverage of
a life
Proposition 2 a declared counter-signal
forces DRIFTING over every
later clause
tests/test_progress.py::t
est_drifting_survives_sta
leness
flags a recorded signal, not a
failing person
Proposition 3 TOWARD iff all markers
observed, none countered,
currentness auditable
tests/test_progress.py::t
est_observed_and_unmet_fi
elds_are_populated, tests/
test_progress.py::test_fr
esh_observation_stays_tow
ard, tests/test_figures.py
::test_completeness_panel
_gives_no_partial_credit
a reading of one record, not
an accreditation
Proposition 4 temporal uncertainty yields
INQUIRY, never DRIFTING
tests/test_progress.py::t
est_stale_observation_rev
erts_toward_to_inquiry, te
sts/test_progress.py::tes
t_undated_entry_cannot_ce
rtify_currentness
age reopens a question, never
condemns
Proposition 5 NOT_OBSERVED arises solely
from no admitted entry
tests/test_progress.py::t
est_unknown_aspiration_id
_is_noted_not_fatal, tests
/test_progress.py::test_e
mpty_entries_yield_all_no
t_observed
absence of signal is not
evidence of drift
Proposition 6 first valid entry per id stands;
identical calls reproduce
identical reports
tests/test_progress.py::t
est_duplicate_entries_fir
st_wins_and_is_noted, test
s/test_formalism_bindings
.py::test_intake_first_wi
ns_and_replay_determinism
_match_manuscript
determinism of the reading,
not truth of the records
Proposition 7 screening is a pure function of
(E, ids(R)); malformed input
is set aside deterministically;
never raises
tests/test_formalism_bind
ings.py::test_intake_scre
ening_never_raises_on_hos
tile_input, tests/test_for
malism_bindings.py::test_
intake_screening_is_deter
ministic_on_all_input
determinism of the screening
function, not truth of records
Proposition 8 staleness is strict: TOWARD at
age 𝜏 , stale from 𝜏 + 1
tests/test_progress.py::t
est_observation_exactly_a
t_staleness_boundary_stay
s_fresh, tests/test_progre
ss.py::test_observation_o
ne_day_past_staleness_bou
ndary_is_stale, tests/test
_formalism_bindings.py::t
est_currentness_boundary_
constants_match_manuscrip
t
the threshold is a
review-cadence choice, not a
decay law
15

## Page 17

Proposition Statement essence Verifying test Boundary
Proposition 9 sweep delegates to
progress_report and
computes no status; codomain
is TOWARD/INQUIRY only
tests/test_analysis.py::t
est_sweep_module_construc
ts_no_status_of_its_own, t
ests/test_analysis.py::te
st_sweep_never_produces_d
rifting_or_not_observed, t
ests/test_formalism_bindi
ngs.py::test_sweep_codoma
in_matches_manuscript
characterizes the instrument,
not any observed practice
Proposition 10 every aspiration’s reading flips
at the same observation age
tests/test_figures.py::te
st_lattice_cells_are_exec
uted_evaluator_readings, t
ests/test_figures.py::tes
t_lattice_reports_a_unifo
rm_flip_age, tests/test_fi
gures.py::test_lattice_de
viation_count_detects_a_p
lanted_outlier
uniformity of the code path,
not of any practice
Proposition 11 registry digest is
permutation-invariant, 64
lowercase hex characters
tests/test_serialization.
py::test_canonical_regist
ry_is_deterministic_json,
tests/test_golden_line.py
::test_registry_is_unique
_and_order_independent, te
sts/test_formalism_bindin
gs.py::test_digest_order_
independence_matches_manu
script
a drift-review handle, no
attestation semantics
Proposition 12 every invariant passes on the
real registry and rejects a
planted-bad one
tests/test_invariants.py:
:test_battery_passes_on_r
eal_registry, tests/test_i
nvariants.py::test_regist
ry_sound_false_on_any_pla
nted_bad
detection proof concerns the
checks, not registry merit
The envelope section declares its blocks after the propositions above, so its rows sit in their own table, in the same document order:
Envelope block Statement essence Verifying test Boundary
Definition 11 ten fields, native_status the
complete ordered
per-aspiration pairs
tests/test_formalism_bind
ings.py::test_report_enve
lope_tuple_matches_the_da
taclass
a data contract, not a
summary and not a score
Proposition 13 the envelope agrees with its
report field for field; any
post-export edit is visible
tests/test_formalism_bind
ings.py::test_envelope_po
inter_matches_manuscript,
tests/test_report_envelop
e.py::test_envelope_match
es_report_verifies_an_arc
hived_pair
archival agreement, not truth
of the report
The bindings are themselves code behavior: they show which claims the suite would catch, not that the registry’s aspirations are
wise or well served.
16

## Page 18

6 Scholarship: an aspiration is a direction, not a score
Golden Line’s four founding aspirations are not inventions. Each translates a larger argument into the narrow idiom of a versioned
registry: attention before production, usefulness beyond the author, repairable systems, and technical work answerable to human
flourishing. The translation is deliberately incomplete. Golden Line borrows questions and warnings from these traditions; it does
not claim to reconcile them, turn them into a moral canon, or use their authority to certify a project.
6.1 A translation, not a synthesis
The scholarship contributes constraints on how to hold a direction. It does not provide a universal scoring function:
Lineage What it contributes here What Golden Line refuses to import
Aristotle and MacIntyre Work becomes intelligible through ends,
practical judgment, and goods internal
to a practice [ Aristotle, 350 BCE ,
MacIntyre, 1981].
A fixed rule that settles every particular
case.
Sen, Nussbaum, Robeyns, and Jonas Capability is answerable to real human
possibilities and to future life, not
valuable merely because it is power [ Sen,
1980, 1999, Nussbaum, 2011, Jonas,
1984]. Which capabilities matter is
settled by public reasoning inside an
open framework, not fixed by a theorist
[Sen, 2004, Robeyns, 2017].
The instrument’s author deciding what
flourishing means for everyone affected,
and any claim that a marker is a
functioning.
Schön, Polanyi, Sennett, and Jackson Skilled work is situated, partly tacit,
reflective, and sustained through
maintenance and repair [ Schön, 1983,
Polanyi, 1966, Sennett, 2008, Jackson,
2014].
Expert opacity, heroics, or a productivist
bias that treats repair as secondary.
Ostrom and Winner Shared goods need governance, while
technical arrangements can embody
power [ Ostrom, 1990, Winner, 1980].
Treating “the commons” as frictionless
openness or technology as socially
neutral.
Campbell, Goodhart, Strathern, Merton,
Espeland and Sauder, Manheim and
Garrabrant, and Simon
Proxy targets distort conduct through
several distinct mechanisms, and
bounded agents need judgment rather
than aggregate maximization [ Campbell,
1979, Goodhart, 1975, Strathern, 1997,
Merton, 1948, Espeland and Sauder ,
2007, Manheim and Garrabrant , 2018,
Simon, 1956].
A numeric Golden Line grade that would
recreate the problem it names.
This is the relevant sense in which Golden Line is scholarly: its registry is a small design intervention informed by disagreements
among practical philosophy, capability theory, sociology of evaluation, studies of work, and technology critique. None of those
sources validates the registry as a universal theory of value. They help specify what the instrument must ask, what it must not
pretend to know, and where it can be gamed.
The groupings above are not a claim of agreement. Aristotle and MacIntyre offer accounts of ends, practices, and judgment; Sen’s
comparative capability approach does not collapse into Nussbaum’s list-and-threshold proposal, and Sen has said why he declines
to supply a fixed list [ Sen, 2004]; Jonas extends responsibility toward future life. Ostrom studies institutional arrangements for
common-pool resources, not a universal recipe for openness. Winner makes a political argument about artifacts; Jackson gives an
interpretive account of repair in a broken world. Goodhart’s original argument concerns monetary-policy targets, not a general law
of human behavior. Golden Line cites these works as distinct design resources, not as a single theoretical foundation or empirical
validation.
The evidentiary status matters. Most of this section is normative or interpretive scholarship: it clarifies concepts and exposes
failure modes. Two citations carry empirical weight of different kinds — Ostrom’s comparative institutional research and Espeland
and Sauder’s field study of law-school rankings — while the remaining measurement literature is historical, methodological, and
sociological. None of the cited work tests Golden Line’s registry, evaluator, or visual language. The package’s own tests establish
17

## Page 19

implementation properties such as determinism and precedence; they do not establish that an aspiration is true, that a marker is
socially sufficient, or that a directional reading predicts an outcome.
6.2 The shape of an end
The oldest commitment is that an activity has a point beyond its own motion. Aristotle’s ethics is organized around the telos,
the end that makes a practice intelligible, and around phronesis, practical wisdom that reads a particular situation rather than
applying a fixed rule [ Aristotle, 350 BCE ]. Golden Line translates that distinction into a horizon and a human reading of the
record. A horizon says what the work is trying to serve; it does not predetermine what a good next decision must look like.
MacIntyre sharpens the point for modern practice. The goods internal to a craft, the excellences reachable only by doing the
work well, differ from the external rewards the same work can earn; a practice is weakened when the latter replace the former
[MacIntyre, 1981]. The aspiration make work useful beyond its author therefore names more than distribution. It asks whether the
work has acquired an excellence that can travel without being severed from the practice that made it trustworthy. The aspiration
prefer systems that can be repaired makes a parallel demand: the work must remain open to correction rather than preserve the
appearance of finishedness.
6.3 Capability and flourishing
Keep technical work answerable to human flourishing draws on the capability approach. Sen’s Tanner Lecture put the prior question
first — equality of what? — and argued that neither utility nor a bundle of primary goods answers it, because what matters is
what a person is actually able to be and to do [ Sen, 1980]. Development as Freedom carries that into policy: development is
judged by real freedoms rather than by the accumulation of means [ Sen, 1999]. Nussbaum develops a threshold account of central
human capabilities, making the question concrete without reducing dignity to one aggregate index [ Nussbaum, 2011]. Sen declines
to fix such a list, holding that which capabilities matter depends on the purpose of the assessment and has to be settled by public
reasoning rather than by the theorist [ Sen, 2004]. Jonas extends the horizon: modern technology can alter distant and future life,
so responsibility must include people who are not present at the moment of design [ Jonas, 1984].
That disagreement about lists is the part of the literature Golden Line uses most directly, because the registry is a short list. Sen’s
objection applies to it without modification: nine entries chosen by one author are a starting point, not a settled account of what
matters. So the registry is versioned and digest-identified rather than canonical, and revising it is an ordinary operation rather
than an admission of error.
Robeyns marks the limit that matters most here. The capability approach is an open framework rather than a finished theory, and
it supplies no evaluation metric by itself; the theories built inside it are what make the further choices about which capabilities
count and how they are weighed [ Robeyns, 2017]. Golden Line borrows the framework’s question and none of those choices, and
the transfer stops well short of the theory: a marker is not a functioning, an aspiration is not a capability, and a horizon report
says nothing about any person’s real freedoms. What the evaluator can do is make “capability treated as its own justification”
visible as a counter-signal. It cannot determine whose flourishing is at stake, resolve a conflict among affected people, or convert
a local record into a public welfare claim. The code records a direction and its derivation; it does not authorize the person who
filed the record to speak for everyone downstream.
6.4 Attention, craft, and transfer
Let attention precede production is a claim about the sequence of judgment, not a celebration of slowness for its own sake. Schön’s
account of the reflective practitioner shows why competent work cannot be reduced to applying a formula: practitioners often
discover and revise the problem while acting, through reflection-in-action [ Schön, 1983]. Golden Line’s “next decision” horizon is
small enough to preserve that situated judgment. It asks for a visible pause before the next output, not for a ritual of documentation.
Polanyi explains why transfer is difficult: experts know more than they can straightforwardly tell [ Polanyi, 1966]. Sennett’s
craftsman gives that difficulty a material and ethical form: quality grows through repeated attention to materials, error, and the
desire to do a job well for its own sake [ Sennett, 2008]. Jackson’s “broken world” account pushes the argument further. Maintenance
and repair are not merely costs after the productive act; they are sites where knowledge, creativity, power, and care become visible
[Jackson, 2014].
These sources explain why the registry uses both Markers and Counter-signals. The markers declared for make work useful beyond
its author are “handoff used” and “reader question answered”: both test transfer by its effect on another person, without assuming
that every piece of tacit knowledge can be compressed into a checklist. The counter-signal declared for prefer systems that can be
repaired is “defect hidden to preserve appearance”, which tests whether repair has been made ordinary. The signal is not a verdict
on character; it is a prompt to inspect the conditions under which knowledge travels and failure can be discussed.
Confucian teaching adds the relational dimension: learning is cultivated through conduct with others, not detached from the
practice of living together [ Confucius, 500 BCE ]. Frankl’s account of meaning explains why a direction can remain orienting
18

## Page 20

through difficulty without becoming a guarantee of success [ Frankl, 1959]. Ibn Khaldun supplies a political caution: purposes
persist when institutions and shared cohesion carry them beyond an individual’s intention [ Khaldun, 1377]. Golden Line therefore
treats usefulness, teachability, and durable understanding as social properties, not merely private virtues.
6.5 Commons, power, and return
The aspiration to return improvements to the commons needs more precision than a generic call for openness. Ostrom’s work on
common-pool resources shows that shared goods are sustained by situated institutions: boundaries, rules, monitoring, graduated
response, and the capacity of participants to govern themselves [ Ostrom, 1990]. The lesson for Golden Line is not that every
artifact should be released without conditions. It is that “returned to the commons” should invite questions about who can use,
inspect, adapt, maintain, and contest an improvement, and about which dependencies make that return possible.
Winner’s question, Do Artifacts Have Politics? , offers the complementary argument that technical things and systems can embody
forms of power and authority [ Winner, 1980]. Human flourishing cannot therefore be checked only at the level of a feature or
repository. It also requires attention to the institutional arrangement in which the feature operates: who sets the terms, who
bears the risk, whose labor remains invisible, and who can refuse. Golden Line does not answer those questions for the affected
community. It keeps the questions from disappearing behind the language of capability or contribution.
6.6 Why it must not become a score
The literature on evaluation supplies Golden Line’s strongest negative design constraint, and it is older and more specific than the
slogan it usually gets compressed into. Campbell stated the mechanism for social indicators: the more a quantitative indicator is
used for social decision-making, the more it is subject to corruption pressures, and the more apt it is to distort the process it was
meant to monitor [ Campbell, 1979]. Goodhart’s 1975 paper addressed problems of monetary management and was not a universal
moral law [ Goodhart, 1975]; Strathern’s study of audit culture is where the compressed “measure becomes a target” phrasing
entered wide circulation, along with the finding that target-driven measurement can displace the performance it was supposed to
represent [ Strathern, 1997].
Manheim and Garrabrant separate the failure into four mechanisms — regressional, extremal, causal, and adversarial — and
note Campbell’s law arguably has scholarly precedence over the Goodhart formulations [ Manheim and Garrabrant , 2018]. The
taxonomy shows exactly how far Golden Line’s design helps. Emitting no aggregate removes one target: there is no single Golden
Line number for selection pressure to push to an extreme. It does not remove the proxies. Each marker is itself a proxy, and an
observer working for “marker present” rather than for the direction reproduces the regressional case one token at a time. Surfacing
Counter-signals before Markers and reading missing evidence as inquiry keeps thin evidence from being rounded up. Nothing
touches the adversarial case: an observer who wants a TOWARD can file the tokens that produce one, because the evaluator checks
tokens against the registry and never the record against the world.
Espeland and Sauder give the empirical version of the same worry. Studying United States law-school rankings, they identify two
mechanisms of reactivity — self-fulfilling prophecy and commensuration — by which a public measure changes the conduct it claims
to describe [ Espeland and Sauder , 2007]. Merton’s earlier account supplies the first in general form: a definition of a situation can
evoke behavior that makes the definition come true [ Merton, 1948]. Golden Line blocks commensuration by construction, because
it emits no comparable number and never aggregates the nine entries into a grade. It does not escape the first. Its scope is also
narrower than theirs in a way worth stating: Espeland and Sauder study a public ranking of organizations, while a horizon report
is a local record with no audience beyond the people who filed it. Whether a private directional record produces reactivity at all
is a question their study does not answer.
Simon supplies the constructive counterpoint. Under bounded rationality, agents work with limited information in structured
environments, and the practical task is satisficing — deciding what is good enough for the situation at hand — rather than
maximizing an abstract proxy [ Simon, 1956]. Golden Line makes that restraint executable. It emits TOWARD, INQUIRY, DRIFTING,
or NOT_OBSERVED; it surfaces counter-signals before markers; it treats missing evidence as inquiry; and it refuses to aggregate the
nine entries into a single grade. A status is a directional reading of one admitted record, never a measure of a person, a project,
or a whole moral life.
6.7 From theory to instrument
The translation can be stated compactly:
• Telos becomes a stated horizon, while phronesis remains with the reader.
• Capability theory becomes a question about affected lives, and a reason to keep the list of aspirations short, versioned, and
open to revision by the people the work touches — not a license for capability to justify itself.
• Reflective practice, tacit knowing, craft, and repair become observable markers and counter-signals about how work is done
and handed on.
19

## Page 21

• Commons scholarship becomes a demand to name governance, reciprocity, and maintenance rather than celebrate openness
abstractly.
• Technology critique becomes a reminder that the surrounding institution is part of the object of attention.
• The sociology of targets becomes a hard prohibition on a Golden Line score, and an admission that no arrangement of clauses
stops an observer who wants a particular reading from filing one.
The result is intentionally modest. Golden Line is a local instrument for keeping a purpose visible while work changes. Its
scholarship makes that modesty more intelligent: a direction is useful only when it remains revisable, situated, answerable to
people beyond its author, and honest about what the record cannot show.
20

## Page 22

7 Evidence boundary and artifact chain
Golden Line does not begin with a score. It begins with a separation of things that are easy to conflate: what the registry declares,
what an observer records, what the evaluator can derive, and what a generated artifact can prove about its own provenance
(formalized in Definition 9 and Proposition 8 ).
7.1 Hard constraints and soft choices
The hard constraints are epistemic and semantic. An observation cannot create a signal that the registry did not declare. Missing
evidence cannot become evidence of absence. A status must be reproducible from its inputs. Malformed temporal metadata cannot
establish currentness. An aspiration cannot authorize what the Red Line refuses or resolve what the White Line leaves absent.
The soft choices are the nine-entry registry, exact token matching, the first-valid-entry rule for duplicates, the optional staleness
window, and the names of the four statuses. They are versioned design decisions, not universal truths. A future registry may
revise them, but it must revise the digest, figures, tests, and manuscript together.
Figure 5: The four HorizonStatus readings mapped to bounded evidence conditions: no admitted entry; incomplete, stale, or
unauditable evidence; a declared counter-signal; and complete current/auditable evidence. The matrix is not a ranking.
7.2 Statuses as bounded claims
NOT_OBSERVED means that no valid entry was admitted. INQUIRY means that an entry exists but does not support a complete
current positive reading. DRIFTING means that a declared counter-signal was recorded and takes precedence over markers. TOWARD
means that every declared marker was observed, no declared counter-signal was recorded, and the observation was current when
temporal review was enabled. None of these readings is a grade, a diagnosis, a safety claim, or a permission mechanism. fig. 5
The full reasons trail remains useful to a human reader, but structured fields ( observed, unmet, countered, ignored tokens, note,
date, and temporal flags) make the derivation directly inspectable. A consumer need not turn explanatory prose back into data
before checking what happened.
Every reading returns the same record; which of its fields carry content is decided by what the matching stage found. fig. 6 runs
one entry per evidence condition and reports, field by field, what came back. Three fields are carried under every condition — the
aspiration id, the status, and the reasons trail — so no reading arrives unexplained, and the remaining nine are carried exactly
when the record supplied something for them. An empty field is an absent observation and nothing more.
21

## Page 23

Figure 6: The twelve structured fields of a finding against six evidence conditions, one executed progress_report call per column.
Filled cells are fields the returned record carries and outlined cells are empty, and every cell prints which it is. A carried temporal
flag records a currentness problem, not a good result; an empty field records an absent observation and never asserts that what it
names is false.
22

## Page 24

7.3 Source to publication
The local release chain is:
registry + status enum
↓
progress_report + invariant battery
↓
deterministic SVG, PNG, and JSON registries
↓
check_artifacts.py
↓
template PDF/HTML render and release audit
check_artifacts.py verifies the registry version and digest, the expected figure inventory, SVG/PNG pairs, generator identity,
and figure-label citations in the manuscript. It is a consistency gate, not independent truth verification. A rendered artifact still
needs visual and publication validation in the sibling template checkout.
23

## Page 25

8 The Golden Line aspirations
The registry opens with four founding aspirations. They are broad enough to travel across research and engineering, but each is
paired with a concrete horizon so that it can be tested against a record rather than merely admired.
1. Let attention precede production. Make enough room to see what the work is actually doing at the next decision.
Horizon: the next decision. Counter-signal: automatic output without review.
2. Make work useful beyond its author. Leave knowledge, tools, and explanations that another person can carry to a
collaborator or public reader. Horizon: a collaborator or public reader. Counter-signal: private cleverness without transfer.
3. Prefer systems that can be repaired. Expose failure early and make correction ordinary in the revision cycle. Horizon:
the revision cycle. Counter-signal: a defect hidden to preserve appearance.
4. Keep technical work answerable to human flourishing. Treat capability as a means whose value depends on the lives
around it over a long horizon. Horizon: the long horizon. Counter-signal: capability treated as its own justification.
Five further entries extend the founding four in the versioned registry, filling out the long-horizon picture: durable understanding
that outlasts the tool, teachable craft that can be handed to the next learner, honest uncertainty kept visible at the same
prominence as the claim, at least one unhurried question measured in years rather than sprints, and improvements returned
to the commons they came from. Each follows the same shape: a horizon, observable markers, and counter-signals ( Definition
1), with the same reading rule ( Definition 10 ). Together the nine entries and their signal structure are laid out in fig. 8.
The order is not a ranking. The aspirations can conflict: usefulness beyond the author can pull against protecting an unhurried
question; returning everything to the commons can pull against an obligation to a specific collaborator, and an honest record may
need to show that conflict rather than resolve it. The companion White Line is the proper place to record what the registry cannot
see or what should not be claimed.
The founding four are best read as a loop of correction, drawn in fig. 7. Attention protects perception before output; usefulness
tests whether what was learned can travel; repair keeps failure from becoming identity; answerability to human flourishing asks
what the capability is for. No point is a maturity level. A project can move around this loop, lose one of its conditions, or find
that two aspirations pull against each other. The useful question is therefore not How high are we? but What does this record
make visible, and what remains unasked?
8.1 The signal vocabulary in aggregate
Read as a whole, the registry declares a deliberately small vocabulary: 18 markers and 9 counter-signals across the nine aspirations
— two markers and one counter-signal per entry, with all 18 marker tokens and all 9 counter-signal tokens distinct across the
registry. The signal_inventory helper in the analysis layer derives these counts from the live source, and fig. 9 draws them one
block per declared token. The distinctness matters: because no token is shared between aspirations, an observed marker can never
accidentally support two directions at once. The blocks are vocabulary the evaluator can match, never observations and never
points.
8.2 The reach of the nine horizons
The nine horizon phrases can also be grouped by when their direction becomes visible. The analysis layer declares four interpretive
temporal-reach bands — immediate (1 aspiration, at the next decision), recurring cycle (2, at revision or tool turnover), at handoff
(4, when the work reaches another person), and open-ended (2, over years) — and horizon_distribution places every registry
entry into exactly one band, refusing loudly if a future registry adds a horizon the map does not classify. fig. 10 shows the grouping.
The bands are a reading aid declared outside the registry contract: band order widens reach, and a wider horizon is not a higher
rank.
The aspirations are exercised in the worked examples and the batch reading .
24

## Page 26

Figure 7: The four founding aspirations held in a revisable loop. Titles, horizons, and marker/counter-signal counts are source-
derived; the connecting loop and icons are interpretive. The loop is direction, not a score or ranking.
25

## Page 27

Figure 8: The full nine-entry aspiration registry: four founding aspirations and five further entries, drawn from the versioned source.
Each row shows its horizon and declared marker/counter-signal counts; the layout is a taxonomy for review, not a performance
scale.
26

## Page 28

Figure 9: The aggregate declared-signal vocabulary of the registry, one unit block per token: filled blocks for markers, outlined
blocks for counter-signals. The counts describe what the evaluator can match, never fulfilment or performance.
27

## Page 29

Figure 10: The nine aspiration horizons grouped into four temporal-reach bands — immediate, recurring cycle, at handoff, and
open-ended. An interpretive reading aid, not a maturity ladder and not part of the registry contract.
28

## Page 30

9 Worked records
The model is intentionally small enough to inspect at the command line. A single horizon entry against one aspiration produces a
full report over the whole registry:
from golden_line import HorizonEntry, progress_report
report = progress_report([
HorizonEntry(
aspiration_id="repairable-systems",
observed_markers=frozenset({"failure named" , "revision attempted" }),
note="The failure was recorded before the release note was written." ,
)
])
That entry yields TOWARD for repairable-systems and NOT_OBSERVED for the other eight registry items, because no horizon note
was recorded for them. The readings move exactly as the decision rule prescribes:
• If the same entry also records the counter-signal defect hidden to preserve appearance , the finding becomes DRIFTING.
Counter-signal precedence ( Proposition 2 ) means this holds even if both markers are still present; a positive signal cannot
launder a recorded drift.
• If only one of the two declared markers is present, the finding becomes INQUIRY with an unmet list naming the marker still
outstanding, so the direction is recorded as open rather than reached — no partial credit ( Proposition 3 ).
• If both markers are present but the observation carries an old, future, or missing/unparseable observed_on date and the
caller passes stale_after_days, the finding reverts to INQUIRY: temporal uncertainty reopens the question without inventing
drift ( Proposition 4 ).
• An entry naming an aspiration that is not in the registry is set aside during intake ( Definition 7) and reported in the intake
notes; it never crashes the report and never silently disappears.
• A malformed record, such as one whose marker field is None, is also set aside and named in the intake notes ( Definition 7 ).
If a dated positive record is supplied with temporal review enabled but its date is not ISO-parseable, the finding is INQUIRY
with date_issue=True (Proposition 4 ): malformed time metadata cannot certify currentness.
The first of those bullets is the one most worth distrusting, because it is the rule a reader is most likely to assume has an exception.
It does not. Running all nine aspirations through four evidence conditions — complete markers alone, complete markers with one
declared counter-signal, a counter-signal with no markers, and a counter-signal on an observation four hundred days old — returns
DRIFTING in every cell where a counter-signal is present, including all nine cells where every declared marker is also present. fig. 12
is that run: each cell in it is a progress_report return value rather than a box drawn to illustrate one.
These distinctions keep a positive signal from laundering a counter-signal ( Proposition 2 ), and keep the absence of evidence from
masquerading as either success or failure ( Proposition 5 ). The path from an entry to its reading is shown in fig. 1, and the three
evaluator stages an entry passes through — including the set-aside branch that carries malformed, unknown-id, and duplicate
records into the intake notes — are shown in fig. 11, while the bounded meaning of all four readings is summarized in fig. 5.
A single entry shows the decision rule in isolation. The next section runs a whole batch — positive, drifting, partial, malformed,
and duplicate entries at once — through the same evaluator and the descriptive analysis layer, so the gap between what was
submitted and what was read becomes fully accountable.
29

## Page 31

Figure 11: A batch of horizon entries passes through intake screening, signal matching, and decision. Records that fail intake
— malformed, unknown-id, or duplicate — are set aside into visible intake notes (quoted verbatim from a real progress_report
replay), undeclared tokens are ignored and noted, and the output is one bounded finding per registry aspiration. Findings are
directional readings, never grades.
30

## Page 32

Figure 12: Counter-signal precedence replayed rather than drawn: for each of the nine aspirations, four progress_report calls
at review date 2026-07-18 with stale_after_days = 90. Complete markers with no counter-signal read TOW ARD; adding one
declared counter-signal returns DRIFTING even though every marker is still present, and it still returns DRIFTING when the
same observation is stale. Filled cells are DRIFTING and outlined cells are not, so the panel reads without colour. The panel
characterizes clause order; it does not grade any work or person.
31

## Page 33

10 Reading a batch: the descriptive layer at work
A single horizon entry is easy to read by eye. A real review is rarely a single entry: it is a batch of observations filed against
several aspirations at once, some positive, some drifting, some partial, and a few malformed. The descriptive analysis layer exists
to characterize such a batch without adding any claim the evaluator did not already make. This section runs one batch end to end
so the helpers can be seen doing exactly — and only — what the formalism permits.
10.1 A worked batch
Consider six submitted horizon entries evaluated against the shipped nine-entry registry with progress_report:
from golden_line import HorizonEntry, progress_report, report_overview
entries = [
HorizonEntry("attention-before-output",
observed_markers=frozenset({"question revisited" , "context named" })),
HorizonEntry("repairable-systems",
observed_markers=frozenset({"failure named" , "revision attempted" }),
counter_signals=frozenset({"defect hidden to preserve appearance" })),
HorizonEntry("useful-to-others",
observed_markers=frozenset({"handoff used" })),
HorizonEntry("honest-uncertainty",
observed_markers=frozenset({"limit stated beside claim" ,
"confidence qualified in print" ,
"extra token that is undeclared" })),
HorizonEntry("not-a-real-id",
observed_markers=frozenset({"whatever"})),
HorizonEntry("attention-before-output",
observed_markers=frozenset({"question revisited" })),
]
report = progress_report(entries)
overview = report_overview(report)
Six entries were submitted, but the report still contains exactly nine findings — one per registry aspiration — because totality
(Proposition 1 ) does not depend on what was filed. report.counts() partitions those nine findings as two TOWARD, one INQUIRY,
one DRIFTING, and five NOT_OBSERVED. The report_overview helper regroups the same findings by status and totals the intake
so the batch can be described without re-parsing prose:
Reading Count Aspirations
TOWARD 2 attention-before-output, honest-unc
ertainty
INQUIRY 1 useful-to-others
DRIFTING 1 repairable-systems
NOT_OBSERVED 5 wide-human-flourishing, durable-und
erstanding, teachable-craft, unhurri
ed-questions, commons-returned
Each reading follows the decision rule of the formalism, and each illustrates one of its guarantees:
• attention-before-output recorded both of its declared markers and no counter-signal, so it reads TOWARD.
• repairable-systems recorded both declared markers and a declared counter-signal. Counter-signal precedence ( Proposition
2) surfaces the drift first: the reading is DRIFTING, and the two present markers cannot launder it.
• useful-to-others recorded only one of its two declared markers, so the direction is held open as INQUIRY with the second
marker still unmet — not rounded up and not condemned ( Proposition 3 ).
• honest-uncertainty recorded both declared markers plus one token, "extra token that is undeclared" , that the
registry never declares. The undeclared token is ignored, counted once in the overview’s ignored_marker_total , and
changes nothing: because both declared markers were present and no counter-signal was recorded, the reading is TOWARD. An
observer-supplied token can never smuggle in a status ( Definition 8 ).
• The five aspirations with no admitted entry read NOT_OBSERVED. This is not evidence that those directions are absent or
failing; it reports only that no valid record reached this report ( Proposition 5 ).
32

## Page 34

10.2 What intake set aside
Two of the six submitted entries never became findings, and the overview makes the reason legible without any narrative: intake
_note_count is 2. The report.intake_notes are, verbatim:
• "entry for unknown aspiration 'not-a-real-id' was set aside"
• "duplicate entry for 'attention-before-output' was set aside; the first entry stands"
The unknown identifier was screened out during intake rather than crashing the report; the duplicate attention-before-out
put entry — a weaker record with only one marker — was set aside by the first-valid-entry rule, and the first, complete entry
stands. Neither disappeared silently: both are named in the intake notes so the gap between six submissions and four incorporated
readings is fully accounted for. For this batch the overview’s ignored_marker_total is 1, its ignored_counter_signal_total,
stale_count, and date_issue_count are all 0 (temporal review was left disabled here), so the entire delta between raw input
and final readings is visible in five small integers.
The whole batch is drawn in fig. 13: the figure builder replays exactly this batch through progress_report and report_overview,
so every count, grouping, and quoted intake note in the picture is the evaluator’s actual output for the code block above, not an
illustration of it.
Figure 13: The 6-entry worked batch of this section replayed through the real evaluator: 2 TOW ARD, 1 INQUIRY, 1 DRIFTING,
and 5 NOT_OBSER VED across 9 findings — one per registry aspiration whether or not an entry was filed — with 2 intake
set-asides and 1 ignored undeclared token quoted verbatim from the report. The groupings count readings in this record; they do
not grade the work or the people behind it.
33

## Page 35

10.3 Vocabulary and reach as reading context
The overview describes what this batch produced; the signal inventory and horizon-band distribution describe the fixed vocabulary
and temporal reach that any batch is read against. Because every one of the registry’s eighteen markers and nine counter-signals
is a distinct token (fig. 9), the ignored_marker_total above is unambiguous: the ignored token matched no declared marker
of any aspiration, not merely the one it was filed under. And because the four aspirations that drew readings here — attention,
repair, usefulness, honest uncertainty — sit in three different temporal-reach bands (fig. 10), a single review batch routinely mixes
an immediate-horizon reading with an at-handoff one. The bands are a reminder that these readings become visible on different
clocks, not a schedule on which they should be expected to agree.
None of these summaries is a grade. report_overview adds no claim the findings did not already carry; it only counts them. A
batch that reads two TOWARD and five NOT_OBSERVED is not “worse” than one that reads nine TOWARD. It is a smaller record.
34

## Page 36

11 Limits and safeguards
Aspiration language can become sentimental, coercive, or falsely universal, and a directional instrument can be misread as a verdict.
Golden Line therefore carries six hard limits. Three of them have a mechanical core the code enforces, and the proposition that
pins each one is named inline; the other three are commitments the code cannot hold for us, and saying so is part of keeping them.
• Commitment. A registry entry is a design choice, not a discovery of humanity’s single highest good. The nine aspirations
are a starting set, versioned and revisable, not a closed canon.
• Enforced (Proposition 3). A TOWARD result says only that declared markers were observed, no counter-signal was recorded, and
(when temporal review is enabled) currentness was auditable from a present, parseable, in-window date. Those conditions
are necessary and sufficient in the code. That the result does not prove the work good, safe, lawful, or beneficial is a reading
rule, not a check.
• Enforced (Proposition 5 ). A NOT_OBSERVED result arises solely from the absence of an admitted entry. It is not evidence
that the aspiration is absent; submitted records may have been malformed, unknown, or duplicate, and the intake notes say
which.
• Commitment. Counter-signals are local warnings about a record, not psychological diagnoses or public labels for a person or
project. Nothing in the evaluator can stop a reader from using one that way.
• Enforced (Definition 7, Proposition 4). A malformed entry is visible as an intake set-aside, and a missing or malformed date
in an enabled temporal review prevents a positive record from being read as current. These are data-quality safeguards, not
proof that the underlying work failed.
• Commitment. The instrument cannot resolve conflicts among affected people; it can only make the chosen direction and its
tradeoffs easier to discuss. This is an absence rather than a check, and no test can confirm it.
The sharpest risk is the one the scholarship section names: that a published aspiration hardens into a target and is gamed rather
than served [Strathern, 1997], or that stating the direction as a scorecard evokes the very behavior it was meant to describe [ Merton,
1948]. The design pushes against this at every stage: no numeric score, no aggregate grade, counter-signals surfaced first, absence
read as inquiry, and no positive currentness claim from malformed temporal data. What it does not touch is the adversarial case
[Manheim and Garrabrant , 2018]: the evaluator matches the tokens an observer filed against the registry, so an observer who wants
a TOWARD can file the tokens that produce one, and no clause ordering prevents it. The honest safeguard is to keep the readings
directional and the responsibility human, in the spirit of Jonas’s long-horizon duty rather than a compliance ritual [ Jonas, 1984].
11.1 Limits of the descriptive analysis layer
The descriptive helpers — signal_inventory, horizon_distribution, temporal_currentness_sweep , and report_overview
— introduce their own, smaller risks of over-reading, and each is deliberately constrained so it cannot outrun the evaluator it sits
beside.
• The inventory counts declared vocabulary, never fulfilment. That the registry declares eighteen markers and nine counter-
signals says nothing about how many have ever been observed; a larger marker count is not a higher standard, only a longer
list of things one could look for. Reading the inventory as a scoreboard would invert its purpose.
• The horizon bands are an interpretive reading aid declared in the analysis layer, not part of the registry contract. Their order
widens temporal reach; it does not rank merit, and an “open-ended” horizon is not superior to an “immediate” one. The
band map is intentionally brittle: an aspiration whose horizon it does not classify raises an error rather than being dropped,
so a future registry change forces the map to be revisited by a human instead of silently mis-grouping an entry.
• The currentness sweep replays one fully-marked, counter-signal-free synthetic entry to expose where a reading loses currency.
It is a probe of the evaluator’s temporal rule under a chosen stale_after_days , not a claim about any real observation,
and the staleness boundary it draws is a data-quality threshold — a stale cell is an invitation to look again, never evidence
that the underlying work failed. The lattice widens the probe to every registry entry, which rules out the boundary being an
artifact of the one entry chosen; it does not make the entries any less synthetic.
• report_overview regroups and counts findings that already exist; it adds no claim the report did not carry and derives
nothing from outside the report’s own structured fields. Its per-status tallies are a summary, and a batch that reads mostly
NOT_OBSERVED is a small record, not a poor one.
Because all four helpers are pure, deterministic, and read-only, they cannot change what a reading means — but a reader can
still misuse a number. The safeguard is the same one the whole instrument relies on: keep the summaries descriptive and the
judgement human.
For security boundaries, use Red Line. For how the work was performed, use Black Line. For what is missing, withheld, unobserved,
or ethically left unclaimed, use White Line. A horizon report must never be used to bypass those scopes, and a TOWARD on any
Golden Line aspiration can never authorize what Red Line refuses.
35

## Page 37

12 Conclusion
Golden Line keeps a question alive: what is this work for, and what would movement toward that purpose look like at the next
honest horizon? Its answer is not a universal doctrine. It is a versioned set of nine threads, each with a horizon, observable markers,
and counter-signals, read by a staged evaluator that returns a direction and never a grade.
The formalism shows how modest the machine is: four directional readings ( Definition 4), a decision rule in which a counter-signal
outranks any marker ( Proposition 2 ) and staleness or a malformed reviewed date reopens rather than condemns ( Proposition 4 ),
and seven structural invariants each kept honest by a planted-bad proof-of-detection test ( Proposition 12 ). Six of the thirteen
figures replay those rules rather than draw them: precedence ( Proposition 2), marker completeness ( Proposition 3), the currentness
boundary ( Proposition 8 ), and the shape of the returned record are all read off executed output ( Proposition 9 ). source, figures,
and manuscript citations remain joined.
The scholarship shows both the age and the stakes of the commitments (see the scholarship section ): ends and practical wisdom,
internal goods, capability and flourishing, reflective craft, repair, commons governance, the political character of technical systems,
and the long-horizon duty that technological power now demands. It also shows why the instrument must stay anti-metric: a
direction hardened into a target stops measuring what it named, and an open commons without governance can simply hide who
carries the cost.
The instrument is strongest when paired with the restraint of its companions, each of which does a job it cannot. Golden Line
adds a direction without turning aspiration into authority, a thread worth reaching toward, held in the open, and revised whenever
the evidence or the values change.
The work is built for DAF’s public research index at docxology, machine-readable, cross-linked, and verification-logged, with its
eventual public home at docxology/golden_line. The aspirational thread is meant to be kept in plain sight.
36

## Page 38

References
Aristotle. Nicomachean Ethics . Various critical editions, 350 BCE. Traditionally dated to the fourth century BCE; on telos,
eudaimonia, the mean, and practical wisdom (phronesis). Translated editions vary. Accessed 2026-07-18.
Donald T. Campbell. Assessing the impact of planned social change. Evaluation and Program Planning , 2(1):67–90, 1979. doi:
10.1016/0149-7189(79)90048-X. Source of Campbell’s law: the more a quantitative social indicator is used for social decision-
making, the more it is subject to corruption pressures and the more apt it is to distort the processes it was meant to monitor.
Accessed 2026-07-28.
Confucius. The Analects . Various critical editions, 500 BCE. Compiled and transmitted in early Chinese traditions; date is
approximate. On learning as the cultivation of conduct within relationships. Accessed 2026-07-18.
Wendy Nelson Espeland and Michael Sauder. Rankings and reactivity: How public measures recreate social worlds. American
Journal of Sociology, 113(1):1–40, 2007. doi: 10.1086/517897. Empirical study of law-school rankings identifying two mechanisms
of reactivity — self-fulfilling prophecy and commensuration — by which a public measure changes the conduct it claims to
describe. Accessed 2026-07-28.
Viktor E. Frankl. Man ’s Search for Meaning. Beacon Press, 1959. English translation; German original Ein Psychologe erlebt das
Konzentrationslager (1946). On meaning as an orienting direction. Accessed 2026-07-18.
Charles A. E. Goodhart. Problems of monetary management: The u.k. experience. In Papers in Monetary Economics , volume 1,
pages 1–20. 1975. URL https://www.econbiz.de/Record/problems-of-monetary-management-the-u-k-experience-goodhart-
charles/10002525062 . Original monetary-policy context for the control-target warning later associated with Goodhart’s law.
Accessed 2026-07-18.
Steven J. Jackson. Rethinking repair. In Media Technologies: Essays on Communication, Materiality, and Society . MIT Press,
2014. doi: 10.7551/mitpress/9780262525374.003.0011. URL https://academic.oup.com/mit-press-scholarship-online/book/
14976/chapter-abstract/169335302 . On broken-world thinking and repair as a site of creativity, knowledge, power, and care.
Accessed 2026-07-18.
Hans Jonas. The Imperative of Responsibility: In Search of an Ethics for the Technological Age . University of Chicago Press,
1984. English edition; German original Das Prinzip Verantwortung (1979). On long-horizon duty toward future life. Accessed
2026-07-18.
C. G. Jung. Psychology and Alchemy . Collected Works of C. G. Jung, Volume 12; Bollingen Series XX. Princeton University Press,
1953. German original Psychologie und Alchemie (1944). Reads nigredo, albedo, citrinitas, and rubedo as figures for psychological
transformation rather than as laboratory chemistry. Cited here only for that symbolic register, which is the register this project’s
name is used in. Accessed 2026-07-27.
Ibn Khaldun. The Muqaddimah: An Introduction to History . Princeton University Press (Rosenthal translation, 1958), 1377.
Prolegomena to the history of the world; on how purposes and cohesion are carried by institutions. Accessed 2026-07-18.
Alasdair MacIntyre. After Virtue: A Study in Moral Theory . University of Notre Dame Press, 1981. On practices, goods internal
to a practice, and the narrative unity of a life toward a telos. Accessed 2026-07-18.
David Manheim and Scott Garrabrant. Categorizing variants of goodhart’s law. arXiv preprint arXiv:1803.04585, 2018. URL
https://arxiv.org/abs/1803.04585. Separates four mechanisms by which optimizing a proxy collapses its relationship to the goal
— regressional, extremal, causal, and adversarial — and notes that Campbell’s law arguably has scholarly precedence over the
later Goodhart formulations. Accessed 2026-07-28.
Robert K. Merton. The self-fulfilling prophecy. The Antioch Review , 8(2):193–210, 1948. doi: 10.2307/4609267. On how a
definition of a situation can evoke behavior that makes the definition come true. Accessed 2026-07-18.
Martha C. Nussbaum. Creating Capabilities: The Human Development Approach . The Belknap Press of Harvard University Press,
2011. On the central human capabilities and a threshold conception of flourishing. Accessed 2026-07-18.
Elinor Ostrom. Governing the Commons: The Evolution of Institutions for Collective Action . Cambridge University Press, 1990.
doi: 10.1017/CBO9780511807763. URL https://www.cambridge.org/core/books/governing-the-commons/7AB7AE11BA
DA84409C34815CC288CD79 . On self-governance and institutional diversity in common-pool resource management. Accessed
2026-07-18.
Michael Polanyi. The Tacit Dimension . Doubleday; reprinted University of Chicago Press, 1966. On tacit knowing: we can know
more than we can tell . Accessed 2026-07-18.
Ingrid Robeyns. Wellbeing, Freedom and Social Justice: The Capability Approach Re-Examined . Open Book Publishers, Cambridge,
UK, 2017. doi: 10.11647/OBP.0130. URL https://www.openbookpublishers.com/books/10.11647/obp.0130 . Open-access re-
37

## Page 39

examination distinguishing the capability approach as an open framework from the specific theories built inside it; the framework
does not itself supply a metric. Accessed 2026-07-28.
Donald A. Schön. The Reflective Practitioner: How Professionals Think in Action . Basic Books, 1983. URL https://www.routle
dge.com/The-Reflective-Practitioner-How-Professionals-Think-in-Action/Schn/p/book/9781138458284 . On reflection-in-action
and the limits of reducing professional judgment to technical formulas. Accessed 2026-07-18.
Amartya Sen. Equality of what? In Sterling M. McMurrin, editor, The Tanner Lectures on Human Values, Volume 1 . Cambridge
University Press, 1980. URL https://tannerlectures.org/lectures/equality-of-what/ . Tanner Lecture on Human Values delivered
at Stanford University, 22 May 1979. Argues that utilitarian, total-utility, and Rawlsian equality each answer the wrong question,
and introduces basic capability equality. Accessed 2026-07-28.
Amartya Sen. Development as Freedom . Alfred A. Knopf; Oxford University Press, 1999. On the capability approach: value
measured in real freedoms to be and to do. Accessed 2026-07-18.
Amartya Sen. Capabilities, lists, and public reason: Continuing the conversation. Feminist Economics, 10(3):77–80, 2004. doi:
10.1080/1354570042000315163. Sen’s reply on why he declines to fix a canonical list of capabilities: the relevant capabilities
depend on the purpose of the assessment and must be settled by public reasoning. Accessed 2026-07-28.
Richard Sennett. The Craftsman . Yale University Press, 2008. On craft as the desire to do a job well for its own sake, and on
repair as a form of skilled attention. Accessed 2026-07-18.
Herbert A. Simon. Rational choice and the structure of the environment. Psychological Review , 63(2):129–138, 1956. doi:
10.1037/h0042769. On satisficing and bounded rationality: agents seek good enough within a structured environment. Accessed
2026-07-18.
Marilyn Strathern. ’improving ratings’: audit in the british university system. European Review , 5(3):305–321, 1997. doi:
10.1002/(SICI)1234-981X(199707)5:3<305::AID-EURO184>3.0.CO;2-4. Source of the widely quoted formulation that when a
measure becomes a target it ceases to be a good measure. Accessed 2026-07-18.
Langdon Winner. Do artifacts have politics? Daedalus, 109(1):121–136, 1980. URL https://www.jstor.org/stable/20024652 . On
technical artifacts and systems embodying forms of power and authority. Accessed 2026-07-18.
38


---
*Extraction method: pypdf*