{"id":16,"date":"2026-09-22T01:08:35","date_gmt":"2026-09-22T01:08:35","guid":{"rendered":"https:\/\/spectrcyde.com\/?p=16"},"modified":"2026-09-29T01:26:03","modified_gmt":"2026-09-29T01:26:03","slug":"the-run-that-is-not-a-capture-building-evidence-that-cannot-flatter-itself","status":"publish","type":"post","link":"https:\/\/spectrcyde.com\/?p=16","title":{"rendered":"The Run That Is Not a Capture: Building Evidence That Cannot Flatter Itself"},"content":{"rendered":"\n

Date:<\/strong> September 13, 2026<\/p>\n\n\n\n

Author:<\/strong> SCYTHE Team<\/p>\n\n\n\n

Category:<\/strong> Evidence-Centered Computing, RF, GraphOps, Promotion Discipline<\/p>\n\n\n\n

—<\/strong><\/p>\n\n\n\n

An SDR is plugged into this workstation. A `Realtek RTL2838` sits on<\/p>\n\n\n\n

`Bus 001 Device 026`, tuned to nothing, doing nothing. Over the last several<\/p>\n\n\n\n

weeks SCYTHE built the entire path that would let a live walk produce evidence<\/p>\n\n\n\n

good enough to promote into GraphOps \u2014 the durable ledger, the ownership lock,<\/p>\n\n\n\n

reconciliation, ceilings, the execution boundary, the derived-evidence reader<\/p>\n\n\n\n

and its producer.<\/p>\n\n\n\n

And then, at the moment the run was scoped, the answer came back: **do not open<\/p>\n\n\n\n

the radio.**<\/p>\n\n\n\n

Not because the radio is dangerous. Because a capture would have changed<\/p>\n\n\n\n

nothing, and running one anyway would have been authorization theater with a<\/p>\n\n\n\n

USB cable attached.<\/p>\n\n\n\n

That decision, and the six slices of code that followed from it, are the most<\/p>\n\n\n\n

interesting thing SCYTHE has built this quarter. The engineering story is not<\/p>\n\n\n\n

“we captured RF.” It is “we established exactly what our evidence is allowed to<\/p>\n\n\n\n

claim, and then wrote code that cannot claim more.”<\/p>\n\n\n\n

## The Finding That Made the Run Small<\/strong><\/p>\n\n\n\n

The question was simple: what does a walk verdict actually depend on?<\/p>\n\n\n\n

SCYTHE’s walk checker compares two position fixes against a kinematic budget.<\/p>\n\n\n\n

Did the operator move further than a person walking could have moved in the<\/p>\n\n\n\n

elapsed time? Two hashes travel with each fix:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

signal_chain_hash          sensor identity, sample type, rate, antenna,<\/p>\n\n\n\n

                           feedline, extension, gain  — all DECLARED<\/p>\n\n\n\n

receiver_state_chain_hash  a manifest of position, speed and orientation<\/p>\n\n\n\n

                           authorities<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

Neither touches a sample. `check_walk_step` never sees one either. A walk<\/p>\n\n\n\n

verdict is a statement about displacement, and displacement is computed from<\/p>\n\n\n\n

latitude, longitude and time.<\/p>\n\n\n\n

So we asked what a capture would buy. It would activate the volatile IQ buffer,<\/p>\n\n\n\n

`rtl_tcp`’s socket binding, the ring’s invalidation rules, and the recovery<\/p>\n\n\n\n

subsystem’s posture \u2014 the entire Q1 surface \u2014 and produce an artefact<\/p>\n\n\n\n

identical in every field the verdict depends on<\/strong>. A live observation does not<\/p>\n\n\n\n

become more meaningful because a receiver was consuming samples nobody looks at.<\/p>\n\n\n\n

The act was classified instead:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

LIVE_POSITION_ATTESTATION      a real bounded observation over real<\/p>\n\n\n\n

                               attested positions<\/p>\n\n\n\n

INSTRUMENT_CONFIGURED_IDLE     the instrument is declared, and idle<\/p>\n\n\n\n

RF_MEASUREMENT_NOT_PERFORMED   no sample was taken, and none was needed<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The RTL2838’s presence may be declared<\/em>. The run must not open, tune, reset,<\/p>\n\n\n\n

claim, or otherwise communicate with it.<\/p>\n\n\n\n

## The Schema Could Not Express That, So It Changed<\/strong><\/p>\n\n\n\n

Here is where a contract stops being prose. The artefact provenance in<\/p>\n\n\n\n

`scythe_derived_evidence` is a closed<\/strong> field set \u2014 a name that is not in the<\/p>\n\n\n\n

set is refused rather than ignored \u2014 and it carried no measurement status at<\/p>\n\n\n\n

all. An artefact could describe a sample rate and a gain, and nothing in the<\/p>\n\n\n\n

record distinguished configured<\/em> from exercised<\/em>.<\/p>\n\n\n\n

Adding a required field to a closed set changes what the old schema name means,<\/p>\n\n\n\n

so the name changed with it:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

scythe.derived-evidence-artefact.v1  ->  .v2<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

A v1 artefact is now refused as foreign \u2014 carrying the new fields or not,<\/p>\n\n\n\n

because the version is not a hint to be overridden by whatever happens to be<\/p>\n\n\n\n

present. There is no migration and no default<\/strong>, and none is needed: no v1<\/p>\n\n\n\n

artefact was ever produced. Writing a compatibility path for a population of<\/p>\n\n\n\n

zero would mean inventing exactly the defaults the amendment forbids, for a<\/p>\n\n\n\n

reader that would never meet one.<\/p>\n\n\n\n

The two new fields are closed to a single value each:<\/p>\n\n\n\n

“`python<\/p>\n\n\n\n

MEASUREMENT_STATUSES = (RF_MEASUREMENT_NOT_PERFORMED,)<\/p>\n\n\n\n

INSTRUMENT_STATES    = (INSTRUMENT_CONFIGURED_IDLE,)<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

One member is the honest size. This tree can produce exactly one kind of<\/p>\n\n\n\n

artefact, and a second value would name a capability that does not exist, read<\/p>\n\n\n\n

by a reader that has never seen one produce anything. When a measurement path<\/p>\n\n\n\n

is contracted, its amendment adds its value.<\/p>\n\n\n\n

## Configuration Is Not Measurement<\/strong><\/p>\n\n\n\n

The subtler rule is the one a well-meaning classifier would get wrong.<\/p>\n\n\n\n

An artefact may legitimately carry the configuration it did not exercise \u2014<\/p>\n\n\n\n

`sample_rate_hz = 2_400_000`, `gain_db = 40.2`, `device_id = “rtl2838-0bda:2838″`<\/p>\n\n\n\n

\u2014 and it is still a run where nothing was measured. Every populated setting<\/p>\n\n\n\n

travels beside its own label:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

sample_rate_hz             2400000<\/p>\n\n\n\n

sample_rate_hz_exercise    CONFIGURED_NOT_EXERCISED<\/p>\n\n\n\n

gain_db                    40.2<\/p>\n\n\n\n

gain_db_exercise           CONFIGURED_NOT_EXERCISED<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

A setting without its label, a label without its setting, and any other label<\/p>\n\n\n\n

are each refused \u2014 and neither the reader nor the producer supplies the missing<\/p>\n\n\n\n

claim, because labelling on the caller’s behalf would make the label unable to<\/p>\n\n\n\n

be wrong.<\/p>\n\n\n\n

The reader’s status function reads one field and looks at nothing else:<\/p>\n\n\n\n

“`python<\/p>\n\n\n\n

def declared_measurement_status(provenance):<\/p>\n\n\n\n

    “””What the artefact says, never what the reader would guess.”””<\/p>\n\n\n\n

    return provenance[“measurement_status”]<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

Not `device_id`: an RTL2838 in the manifest is an instrument that was named<\/em>,<\/p>\n\n\n\n

and naming one is not using one. Not the settings: a populated rate and gain<\/p>\n\n\n\n

are a configuration, and the whole rule is that a declared configuration must<\/p>\n\n\n\n

not imply it was exercised.<\/p>\n\n\n\n

The important design choice is what happens to that configured-idle artefact:<\/p>\n\n\n\n

it is accepted, with its status preserved<\/strong>, not refused. Refusing it would<\/p>\n\n\n\n

teach a producer to omit the configuration entirely, after which the reader<\/p>\n\n\n\n

knows less<\/em> about what was attached than it does now. That is the refusal that<\/p>\n\n\n\n

makes the record worse.<\/p>\n\n\n\n

## The Observer Cannot Say It Either<\/strong><\/p>\n\n\n\n

The artefact is only half the record. The observation record \u2014 the thing that<\/p>\n\n\n\n

says what this run was<\/em> \u2014 also has to carry the declaration, and here the<\/p>\n\n\n\n

property is an absence rather than a feature.<\/p>\n\n\n\n

The observer has:<\/p>\n\n\n\n

– no constructor field for a measurement status<\/p>\n\n\n\n

– no `run` parameter for one<\/p>\n\n\n\n

– and, enforced by an AST test over the entire module, **no occurrence of<\/p>\n\n\n\n

  `RF_MEASUREMENT_NOT_PERFORMED` or `INSTRUMENT_CONFIGURED_IDLE` as a name or<\/p>\n\n\n\n

  inside any string**<\/p>\n\n\n\n

Containment, not equality \u2014 a token spliced into a longer literal would reach a<\/p>\n\n\n\n

record just as surely as one standing alone.<\/p>\n\n\n\n

The only route is an `Artefact`, taken nominally:<\/p>\n\n\n\n

“`python<\/p>\n\n\n\n

InstrumentDeclaration.from_artefact(artefact)   # type(artefact) is Artefact<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The observer never met an instrument and neither did the producer. The only<\/p>\n\n\n\n

party with a claim to make is the artefact, and the artefact makes it in<\/p>\n\n\n\n

writing. So a run carries a declaration or says it has none:<\/p>\n\n\n\n

| run | authority | status |<\/p>\n\n\n\n

| — | — | — |<\/p>\n\n\n\n

| derived artefact | `CARRIED_FROM_ARTEFACT` | the artefact’s own values |<\/p>\n\n\n\n

| constructed | `NO_INSTRUMENT_DECLARATION` | `NO_INSTRUMENT_DECLARATION` |<\/p>\n\n\n\n

A constructed run has no instrument<\/strong> \u2014 not an idle one, not a configured<\/p>\n\n\n\n

one, none. A constructed run handed a carried declaration, and a live run<\/p>\n\n\n\n

handed one from nowhere, both publish no record at all<\/em>: the mismatch is<\/p>\n\n\n\n

refused before the first verdict rather than recorded beside the numbers it<\/p>\n\n\n\n

would undermine.<\/p>\n\n\n\n

## One Read, Or The Record Describes A Different File<\/strong><\/p>\n\n\n\n

Verdicts and the declaration must come from the same bytes. Reading the<\/p>\n\n\n\n

artefact path once for the verdicts and again for the declaration would attest<\/p>\n\n\n\n

an instrument belonging to whatever the second open found.<\/p>\n\n\n\n

So the verdict source became a type rather than two arguments:<\/p>\n\n\n\n

“`python<\/p>\n\n\n\n

VerdictSource(declaration=…, verdicts=…)   # one artefact, one read<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

A declaration passed *beside*<\/em> an iterator is a declaration about whatever the<\/p>\n\n\n\n

caller says. One test counts the opens; one negative control performs the<\/p>\n\n\n\n

second one and fails that test.<\/p>\n\n\n\n

The same rule surfaced again a directory down. The record reports whether the<\/p>\n\n\n\n

promotion lineage changed during the observation, by digesting every published<\/p>\n\n\n\n

generation before and after. An empty digest map turned out to have **three**<\/strong><\/p>\n\n\n\n

causes:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

{}   the namespace does not exist<\/p>\n\n\n\n

{}   it exists and holds no generation<\/p>\n\n\n\n

{}   the listing failed<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

A record publishing only the map flattens all three into one \u2014 and the<\/p>\n\n\n\n

flattened version reads as the most reassuring of them: *a valid empty<\/p>\n\n\n\n

generation was observed and did not change*. Nobody made that claim.<\/p>\n\n\n\n

The fix is a single snapshot from a single directory listing:<\/p>\n\n\n\n

“`python<\/p>\n\n\n\n

LineageSnapshot(presence, digest)<\/p>\n\n\n\n

#   NO_LINEAGE_NAMESPACE | LINEAGE_HOLDS_NO_GENERATION | LINEAGE_PRESENT<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

Two listings can describe two filesystem instants, so presence answered by one<\/p>\n\n\n\n

and digests taken from the other is a record about no single moment. And a<\/p>\n\n\n\n

listing that fails for any reason other than a missing namespace raises<\/p>\n\n\n\n

`LINEAGE_INSPECTION_REFUSED` and publishes nothing \u2014 *we could not look*<\/em> is not<\/p>\n\n\n\n

an answer to *what was there*<\/em>.<\/p>\n\n\n\n

An earlier draft of that record also carried a `lineage_present` boolean beside<\/p>\n\n\n\n

the enum. It was removed. A serialized copy is a second answer that can<\/p>\n\n\n\n

disagree with the one next to it, and a reader has no way to tell which one the<\/p>\n\n\n\n

writer meant. The convenience survives as a derived property; the evidence does<\/p>\n\n\n\n

not store it twice.<\/p>\n\n\n\n

## Names Are Checked By Machine, And Rejected Rather Than Judged<\/strong><\/p>\n\n\n\n

SCYTHE keeps two disjoint verdict vocabularies: **merit**<\/strong> codes say something<\/p>\n\n\n\n

about the subject, **executability**<\/strong> codes say whether a verdict could be<\/p>\n\n\n\n

reached at all. They are named disjointly and counted separately, and a name<\/p>\n\n\n\n

that reads like a member of the other set is a bug even when the code is<\/p>\n\n\n\n

correct.<\/p>\n\n\n\n

A test enforces this mechanically. It parses every non-test module in the tree<\/p>\n\n\n\n

via AST \u2014 never raw text \u2014 and discovers the universe rather than trusting a<\/p>\n\n\n\n

hand-maintained list:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

discovered tokens          395<\/p>\n\n\n\n

declared merit codes        25<\/p>\n\n\n\n

declared executability      50<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The check is component-level containment in both directions, plus negation-pair<\/p>\n\n\n\n

detection for `UN`\/`NON`\/`NOT` prefixes. Every hit is either a real collision or<\/p>\n\n\n\n

a recorded judgement with a reason. There is no third state.<\/p>\n\n\n\n

The interesting part is the names it has killed. In the last three slices alone:<\/p>\n\n\n\n

| candidate | collided with | outcome |<\/p>\n\n\n\n

| — | — | — |<\/p>\n\n\n\n

| `DECLARED_NOT_EXERCISED` | `LEDGER_GENERATION_UNDECLARED` | renamed `CONFIGURED_NOT_EXERCISED` |<\/p>\n\n\n\n

| `ARTEFACT_DECLARED` | `LEDGER_GENERATION_UNDECLARED` | renamed `CARRIED_FROM_ARTEFACT` |<\/p>\n\n\n\n

| `ARTEFACT_ATTESTED` | `LOCK_EXCLUSION_UNATTESTED` | rejected |<\/p>\n\n\n\n

| `LINEAGE_ABSENT` | `ABSENT` (a coordinate kind) | renamed `NO_LINEAGE_NAMESPACE` |<\/p>\n\n\n\n

| `LINEAGE_INSPECTION_FAILED` | `FAILED` | renamed `LINEAGE_INSPECTION_REFUSED` |<\/p>\n\n\n\n

Each rejection is itself a test. `DECLARED_NOT_EXERCISED` would have<\/em> collided,<\/p>\n\n\n\n

and a test asserts it \u2014 so the reason survives after the prose explaining it has<\/p>\n\n\n\n

been forgotten.<\/p>\n\n\n\n

## Every Property Has A Mutation That Breaks It<\/strong><\/p>\n\n\n\n

A passing test suite proves that code runs. It does not prove the tests are<\/p>\n\n\n\n

watching anything. SCYTHE’s answer is a negative control per property: a<\/p>\n\n\n\n

deliberate mutation that must make exactly<\/strong> that property’s tests fail and no<\/p>\n\n\n\n

others.<\/p>\n\n\n\n

The last four slices added 53, on a set that numbers 132 at the time of<\/p>\n\n\n\n

writing. A representative sample:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

status inferred from populated sample rate and gain    -> 4\/4 broken<\/p>\n\n\n\n

status inferred from device identity                   -> 4\/4 broken<\/p>\n\n\n\n

the observer writes the status into the record itself  -> 1\/1 broken<\/p>\n\n\n\n

a measurement value spliced into the observer’s source -> 1\/1 broken<\/p>\n\n\n\n

the artefact opened twice, declaration and verdicts apart -> 1\/1 broken<\/p>\n\n\n\n

presence inferred from an empty digest map             -> 2\/2 broken<\/p>\n\n\n\n

an unreadable namespace reported as an empty lineage   -> 2\/2 broken<\/p>\n\n\n\n

a second answer serialized beside the enum             -> 2\/2 broken<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

Each control also runs every other<\/em> test in its slice, to catch a mutation<\/p>\n\n\n\n

that breaks half the suite and therefore proves nothing about which property<\/p>\n\n\n\n

the tests are watching.<\/p>\n\n\n\n

Controls earn their keep by finding real weaknesses, and three did:<\/p>\n\n\n\n

– the closed-set test read its constants at import and never saw the sets open;<\/p>\n\n\n\n

– the AST test compared strings by equality, so a token spliced into a longer<\/p>\n\n\n\n

  literal passed it;<\/p>\n\n\n\n

– an undeclared-name control reached only the reader’s copy of a set the<\/p>\n\n\n\n

  producer binds separately.<\/p>\n\n\n\n

All three were test bugs, found by the mutation rather than by review.<\/p>\n\n\n\n

## Merge Is Not Acceptance<\/strong><\/p>\n\n\n\n

One process lesson is worth recording because it cost three repairs.<\/p>\n\n\n\n

The governing discipline here is proposal \u2192 acceptance \u2192 implementation<\/strong>, each<\/p>\n\n\n\n

a separate commit, merged before any code depends on it. Three times, an<\/p>\n\n\n\n

amendment was merged while the document still described itself as `PROPOSED`.<\/p>\n\n\n\n

Each was repaired by a forward acceptance commit rather than a rewritten merge:<\/p>\n\n\n\n

treating the conversation as acceptance while the document said otherwise would<\/p>\n\n\n\n

leave a contract contradicting itself, which is worse than a visible two-step.<\/p>\n\n\n\n

The tempting generalization \u2014 a merge instruction implies acceptance<\/em> \u2014 was<\/p>\n\n\n\n

drafted, reviewed, and thrown out. It deletes a decision the contract exists to<\/p>\n\n\n\n

require. A proposal may be merged precisely to preserve it as a proposal<\/p>\n\n\n\n

without<\/em> authorizing what it describes. What landed instead is a gate:<\/p>\n\n\n\n

> A proposed amendment requires an explicit acceptance decision and an<\/p>\n\n\n\n

> acceptance commit before implementation or execution. Merge approval alone<\/p>\n\n\n\n

> does not supply acceptance unless it expressly says that the amendment’s<\/p>\n\n\n\n

> substance is accepted.<\/p>\n\n\n\n

## Where It Stands<\/strong><\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

Promotion Execution Contract    3,596 lines, amendments A-M<\/p>\n\n\n\n

Test suite                      1,493 tests, OK (skipped=1)<\/p>\n\n\n\n

Negative controls               132, all discriminating<\/p>\n\n\n\n

Slices landed                   3, 4, 6, 6b, 7, 8, 9, 10a-10g<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

Counts as of merge `f6bdf69`. They move every slice; the contract line count<\/p>\n\n\n\n

and the vocabulary figures move more slowly.<\/p>\n\n\n\n

The directories for the bounded run are pinned, resolved against the host, and<\/p>\n\n\n\n

created at mode `0700`:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

lineage root   \/home\/spectrcyde\/scythe-ledger\/promotion             absent<\/p>\n\n\n\n

derived        \/home\/spectrcyde\/scythe-live-observation\/derived     empty<\/p>\n\n\n\n

records        \/home\/spectrcyde\/scythe-live-observation\/records     empty<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The namespace-refusal check was run against the real function, not by eye \u2014<\/p>\n\n\n\n

a detail worth stating, because the rule is wider than it looks. The lineage<\/p>\n\n\n\n

root is a *filename prefix*<\/em>, not a directory, so the forbidden namespace is its<\/p>\n\n\n\n

parent. Placing the lineage root directly inside the observation tree refuses<\/p>\n\n\n\n

both<\/strong> output directories, and that is the arrangement anyone tidying paths<\/p>\n\n\n\n

would reach for first.<\/p>\n\n\n\n

## The Last Blocker Is A Person<\/strong><\/p>\n\n\n\n

Everything mechanical is done. What remains is not mechanical.<\/p>\n\n\n\n

The run needs twelve operator-declared<\/strong> fixes: real positions, observed<\/p>\n\n\n\n

during a walk, read off a screen and typed in by a person, roughly fifteen<\/p>\n\n\n\n

seconds apart, each stamped with the host’s `monotonic_ns` at the instant the<\/p>\n\n\n\n

runner accepts it. That is the host’s *ingestion*<\/em> time, not the receiver’s fix<\/p>\n\n\n\n

time, and the artefact says so \u2014 the two differ by however long the<\/p>\n\n\n\n

transcription took, and no elapsed time may be computed across a phone’s wall<\/p>\n\n\n\n

clock and a host monotonic value.<\/p>\n\n\n\n

The authority chain is recorded in three fields rather than one, because<\/p>\n\n\n\n

collapsing them would lose the step where the authority actually degrades:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

position source       PHONE_DISPLAYED_FIX<\/p>\n\n\n\n

transfer              HAND_TRANSCRIPTION<\/p>\n\n\n\n

accepted authority    OPERATOR_DECLARED<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The phone may hold a genuine GNSS fix. What reaches the artefact is what a<\/p>\n\n\n\n

person read and retyped. **A fix read off a screen and typed in is a typed<\/p>\n\n\n\n

fix**, and it is not elevated to device-attested GNSS merely because a phone<\/p>\n\n\n\n

displayed it.<\/p>\n\n\n\n

So the run waits. Not on a missing feature, and not on a device \u2014 on somebody<\/p>\n\n\n\n

walking.<\/p>\n\n\n\n

It is worth being exact about what the code enforces here, because the<\/p>\n\n\n\n

temptation is to describe the intent and let a reader hear a guarantee.<\/p>\n\n\n\n

**The cadence is attested, not enforced.**<\/strong> The runner requires an interactive<\/p>\n\n\n\n

terminal, which rejects a pipe or a redirected file. It cannot prove a person<\/p>\n\n\n\n

was not pasting a buffer they prepared earlier: a paste into a live terminal<\/p>\n\n\n\n

arrives through the same descriptor a typed line does. So the fifteen-second<\/p>\n\n\n\n

spacing is an operator’s attestation and a target, never a refusal \u2014 M.5<\/p>\n\n\n\n

declares only the count and the duration as active bounds. What the artefact<\/p>\n\n\n\n

carries is the *actual*<\/em> monotonic interval between accepted fixes, which means a<\/p>\n\n\n\n

reviewer can see a cadence that does not look like walking even though nothing<\/p>\n\n\n\n

refused it at the time.<\/p>\n\n\n\n

**A short run publishes nothing.**<\/strong> The runner will not interpolate a missing<\/p>\n\n\n\n

fix, will not repeat one to make the count, and will not substitute constructed<\/p>\n\n\n\n

evidence. What it does with an incomplete set is refuse: on the 240-second<\/p>\n\n\n\n

deadline it raises `ENTRY_DEADLINE_REACHED` and no artefact is written at all.<\/p>\n\n\n\n

An earlier draft of this post said such a run “ends on its bounds with what it<\/p>\n\n\n\n

has” \u2014 that describes the amendment’s intent for a run and not the code’s<\/p>\n\n\n\n

behaviour, which has no notion of a partial artefact. Eleven fixes produce<\/p>\n\n\n\n

nothing.<\/p>\n\n\n\n

The no-fallback prohibition is the one that would be tempting at the moment it<\/p>\n\n\n\n

mattered, which is exactly why it is written down.<\/p>\n\n\n\n

—<\/strong><\/p>\n\n\n\n

*The Promotion Execution Contract lives at `docs\/PROMOTION_EXECUTION_CONTRACT.md`.<\/p>\n\n\n\n

The vocabulary rules it conforms to are in `SCYTHE_VERDICT_VOCABULARIES.md`.<\/p>\n\n\n\n

Nothing described here has promoted anything to GraphOps: promotion authority<\/p>\n\n\n\n

remains a separate explicit act, and the ARMED path stays blocked until a<\/p>\n\n\n\n

completed observation record is reviewed.*<\/p>\n","protected":false},"excerpt":{"rendered":"

Date: September 13, 2026 Author: SCYTHE Team Category: Evidence-Centered Computing, RF, GraphOps, Promotion Discipline — An SDR is plugged into this workstation. A `Realtek RTL2838` sits on `Bus 001 Device 026`, tuned to nothing, doing nothing. Over the last several weeks SCYTHE built the entire path that would let a live walk produce evidence good […]<\/p>\n","protected":false},"author":1,"featured_media":68,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-16","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-scythe"],"_links":{"self":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts\/16","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=16"}],"version-history":[{"count":1,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts\/16\/revisions"}],"predecessor-version":[{"id":17,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts\/16\/revisions\/17"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/media\/68"}],"wp:attachment":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=16"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=16"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=16"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}