RED-first fixes for the three bugs the source-blind field test caught. Two
are the same failure the whole cycle keeps circling: prose that tells a
caller something the code does not do.
- The chart not-found refusal prescribed `aura exec --tap <NODE.FIELD>=<FOLD>`.
The selector takes a declared tap NAME, so following the refusal verbatim
earns a second refusal ("the tap plan names 'sub.value', but the blueprint
declares no such tap"), and the flag is not needed at all — a plain run
records every declared tap. This sentence predates the cycle and was carried
over verbatim while the rest of the refusal was corrected; the field test
ran it as written, which is exactly why it ran it.
- `aura measure ic <family-handle>` answered "no recorded run '…' under
runs/traces" for a handle that IS recorded — the value the campaign run
printed, that `chart` renders, whose directory exists. A family holds
members rather than one series, so IC is measured over a member. The
refusal now says that and lists the member keys, which also closes the
friction that nothing on any text surface prints them: reaching a member
previously meant the very `ls runs/traces/` this cycle set out to retire.
- C28's #297 note still gave the entry points' signature as the retired
`(_, Vec<String>)` pair in the present tense, contradicting C27 one contract
over. The close commit claimed lockstep; this makes it true.
Both code fixes have tests that fail without them. The field test also found
four friction items and two spec gaps, filed rather than fixed here.
refs #309
Architect review over 9636b00..9221bcd. What holds, confirmed against the
diff rather than the commit body: no stored record shape moved (neither
`aura-engine/src/report.rs` nor the registry compat mirror appears in the
change at all, and the byte pins pass unedited); the handle travels the
beside-the-report route C27/#297 established for the unbound-tap names, with
the library still printing nothing; trace enumeration landed inside
`aura-registry` where C22 puts trace file I/O; no C14 exit class moved, both
refusal arms falling through to the same exit 1.
Two high drift items, both fixed here:
- The corrected refusal claimed a campaign run prints its handle as
`trace_name` on stdout. It prints `campaign_run.trace_name` — the campaign
leg wraps its record, as the glossary states outright — so the chain the
refusal advertised returns null there. This is the third instance of one
error class in this cycle: a false claim inside the very prose meant to
stop misdirecting callers. Grounding caught the first two, review the
third.
- The measurement leg shipped without the handle pin the spec asked for. All
three new pins drove the strategy path, so the measurement leg emitted a
handle nothing asserted. It has its own test now, on the shared bind pair
the two legs cannot drift across.
Ledger and vocabulary brought into lockstep: C27's current state records the
`RunOutcome` carrier that replaced the pair; the glossary's tap entry names
the single-run handle and warns that a family id is not one; the authoring
guide's own single-run tap example now says the name is printed rather than
guessed. The suggestion line gained the `aura:` prefix every other stderr
line carries.
Bench: all five fingerprints OK — the decisive signal, since they hang on the
record bytes this cycle was required not to move. Timing deltas are load
noise (loadavg 8.7 under concurrent agents), not regressions.
refs #309
A single run persisted its taps and then said nothing about where: the only
way to learn the directory name was `ls runs/traces/`, and the chart intake's
not-found refusal pointed at "the handle a sweep/walk-forward/campaign run
printed" — two verbs retired with #319, and a single run printed no handle at
all. A caller holding a family id from a families listing was stuck: that id
is not a trace handle, and nothing said so.
The handle now rides BESIDE the report, not inside it. Both declared-tap entry
points return `RunOutcome { report, skipped, trace_name }` in place of the
`(report, skipped)` pair, and the CLI composes report + handle into one stdout
object through a `serde(flatten)` wrapper. `RunReport`, `MeasurementReport`,
`RunManifest` and the registry's compat mirror are untouched, so no stored
record shape moves and a tap-free run's line stays byte-identical.
Placement is the load-bearing decision, and it follows a precedent rather than
inventing one: the report is the durable C18 run record — its manifest states
what the run *was* — while a trace directory is where its output went. The
project settled this exact question one cycle earlier for the sibling value,
where the unbound-tap names ride beside the report so the CLI, not the
library, prints the note (C27/#297). An embedding host gets the handle as a
value, never as text to parse back.
The chart refusal gains a second arm. A family id
(`{campaign8}-{strategy_ordinal}-{instrument}-w..-r..-s..-{run}`) is
recognised syntactically — the trailing segment shape plus an 8-hex head, the
form `derive_trace_name` mints — and answered with the campaign's REAL
recorded handles, filtered to those `chart` would accept, suggesting one only
when exactly one matches. It never derives a handle by truncating the id: the
id's second segment counts strategies while the handle's counts runs, and one
campaign run mints family ids at several strategy ordinals whose traces all
live under that single run's handle. `TraceStore::names()` supplies the
enumeration, keeping trace file I/O in the store where C22 puts it.
Verification caught three errors worth recording. The first spec draft claimed
the handle was the id's leading pair — refuted by a green test where a run-0
campaign mints ordinal-1 ids. The second draft's replacement refusal said a
campaign run "prints one handle per family"; it prints one per run. Review
then found the `NotFound` filter untested — deleting it left the suite green —
so the case now has a test that fails without it.
closes#309
RED-first fixes for the two code bugs the safe-to-embed milestone
fieldtest caught: (1) run_signal_r panicked (exit 101) on a blueprint
exposing neither bias nor a declared tap — the CLI pre-validated but the
library seam, the very surface the milestone promises is kill-free, did
not; the guard now refuses class 2 with the CLI's prose family before
wrap_r (run_measurement has no wrap seam and no hole). (2) The override
unknown-param refusal exited 1 through the one inline site the #297
adjudication missed; it and its wrapped-retired sibling are class 2 now,
three existing pins re-pinned with the adjudication comment.
Doc truth from the same fieldtest: the guide's campaign no-data class
corrected (contained cell faults exit 3, not 1), TapPlanError's rustdoc
no longer advertises the retired exit-1 register, and C30 records that
rev + the embedding's committed lockfile are the reproducible-build
contract (branch-referenced engine deps resolve at lock time).
refs #296
reproduce_family returns Result<ReproduceReport, RunnerError> — divergence
is a result the embedding caller reads (all_identical()), never an error;
the CLI renders the per-member lines and summary byte-identically and owns
the exit (1 on divergence, no stderr line). The library's print block and
the empty-message refusal are gone; exit_on_runner_error always prints
(no empty producer remains).
reproduce_family_in gains the identity/pip guard, firing before any store
fetch: an instrument mismatch (incl. synthetic-for-real) and a broker/pip
mismatch — forward-built label compare against the stored broker string
(RunManifest carries no pip field; the mint's label constructors became
shared pub(crate) helpers, so guard and stamp cannot drift). The class is
context-borne: the explicit-source seam refuses 2 (caller-named source),
the derived simple path refuses 1 (data drift), same prose. A window
guard is deliberately absent — no reproduce path takes a caller window.
In-process tests pin the promise: divergence returns Ok, drift refuses
class 1, the host survives. Fork minutes: issues/299#issuecomment-4875,
-4877, -4879.
closes#299
All 14 process::exit sites in aura-runner (member x8, measure x5,
translate x1) convert to RunnerError { exit_code, message } propagation;
run_signal_r / run_measurement return Result tuples carrying the
skipped-tap names beside the C18 reports (record shapes untouched), and
the CLI owns printing via the shared exit_on_runner_error arm (the
dispatch_reproduce pattern — reproduce now uses the same helper).
cost_knob becomes fallible: the sweep worker maps it into the cell's
MemberFault (closing the one hole in 'a worker never process-exits'),
the persist path keeps its string channel. The skipped-tap note is
CLI-printed from returned data, byte-identical.
The one deliberate behaviour change is the minuted C14 adjudication:
refusals in the content of what argv named (binding, synthetic mismatch,
compile, tap-bind content, exposes-neither) move exit 1 -> 2; the
boundary is FORM vs VALUE — an out-of-domain override value keeps the
runtime class via the panic containment, matching the campaign leg.
Environment/data/IO refusals stay 1 (TapPlanError::exit_class is the
single source). Named re-pins in exec.rs, run_refuses_unrunnable_
blueprint.rs, tap_recording.rs, graph_construct.rs; new library-level
test pins the milestone promise (a refusal returns in-process, the
embedding host survives). Prose is byte-identical everywhere.
Docs in lockstep: C28 deferred block closed (14 sites, hole closed),
C14 partition + form-vs-value boundary, C27 note emission, guide exit
classes. Fork minutes: issues/297#issuecomment-4868 and -4873.
closes#297
- parse_override_tokens distinguishes the grammar fault (no '=' / no
dotted path — prose unchanged) from a shape-valid token whose VALUE
fails the scalar-literal parse: the refusal now names the value and
the literal requirement (styled after the kind-mismatch sibling)
instead of falsely claiming a NODE.PARAM=VALUE violation. RED-first;
exit stays 2 on both branches.
- the authoring guide's multi-regime family ids corrected to the
executor's real shape (infix -r{k} after the window segment, -r0
explicit on the default regime), and §1's name-op paragraph no
longer claims naming dissolves same-name trace overwriting (that
trace-dir question is tracked on #311, not resolved here).
Architect drift review (scope 77ad046..HEAD): What holds — the #343
reference-semantics hash is code-backed on both output surfaces
(record line + trace index), the RiskRegime::Fixed stamp/re-derive
lockstep is complete both ways, and both new guide examples build
against the binary. Drift resolved in this commit: C24's
root-name-gated intake enumeration gains the introspect --taps FILE
branch plus its missing per-site pin (byte-identical refusal to the
register sibling — the #331 class-enumeration lesson applied); C27
gains the #337 realization note (positive discovery view closes the
recovery-only gap); C12's inline-hash sentence updated to the topo
parameter; five stale dispatch_run comments reworded to name
exec_blueprint_leg as the surviving owner.
Remaining drift routed: the unpinned guide op-scripts are a durable
debt class, filed as #344 (idea) rather than swept here.
Regression gate: aura-bench all five fingerprints OK (report-only
deltas within noise on a loaded box; no baseline move, nothing to
ratify).
refs #337
refs #343
Harvest sweep, review re-check fix.
The prior fix corrected only the stdout record line: run_signal_r
persists the manifest to runs/traces/<name>/index.json (bound.finish)
BEFORE the post-hoc overwrite ran, so an overridden run's trace store
still carried the reopened hash — diverging from stdout in exactly the
case the revised C24/C18 clause describes. run_signal_r now takes
topo: Option<&str> (None = inline computation as before; Some = the
caller's reference-semantics hash, the run_blueprint_member precedent)
so record line and trace index read the one hash built before any
persistence; exec's override branch passes the base document's id and
the post-run mutation is gone. RED-first: the new trace-store pin
failed against the divergent state.
refs #343
Harvest sweep, whole-diff review fix block.
The review refuted the #343 ratification's factual premise: the
campaign leg does NOT mint a reopened member's hash — runner.rs passes
the stored strategy_id as topo even for reopened members (reference
semantics). Decision revised on the issue (fix, not ratify): exec's
override run now stamps the LOADED base document's content id
(computed before reopen_all; RED-first — the inverted pin failed on
the old reopened-hash behaviour), keeping topology_hash
store-resolvable and identical in meaning across both legs; the
variation lives in manifest.params. C24/C12/C18 rewritten to the
corrected reading.
Also per review: C24's retired build-free-introspection sentence
updated (old wording archived in the history sidecar, #339 item 4);
C14's partition key precisely scoped to the routing seam, naming the
legs' differing content-validation exits honestly; measure.rs's false
unreachability comment corrected; --taps added to the guide's
introspect cheat-sheet and pinned on the content-id path; the
zero-trade-cell note routed through the diag macro; doc-comment,
issue-ref, and gate-name corrections; the Fixed-regime NaN branch
pinned.
refs #339
refs #342
refs #343
Harvest sweep, batch 6 of 6.
- #343 ratified (per the decision minuted on the issue): an override IS
a topology-parameter variation — the executed run's topology_hash is
the reopened topology's own, deliberately not required to resolve in
the store; the manifest's params map plus the base document
reconstruct the variant. C24 carries the full clause, C12/C18
cross-reference it; pinned by a bare-vs-noop-override hash test
(distinct, and stable across runs).
- #342 item 5: C14 pins exec's unresolved-target exit partition on the
existing usage/runtime line — file-content faults (not-JSON,
op-script, wrong-kind) exit 2, missing-state faults (unreadable
non-id target, unregistered content id) exit 1. One sanctioned
realignment: the wrong-kind refusal moves 1 → 2 to join its
file-content siblings (RED-first on the updated pin); the
unregistered-id row gains its missing sibling pin.
- #341 item 4: the authoring guide gains a worked cost-wiring op-script
(ConstantCost → CostSum via PositionManagement geometry, built and
verified), honest about being a wiring reference — production cost
models still come from a campaign document's cost block (C10).
closes#341closes#342closes#343
Harvest sweep, batch 5 of 6.
- introspect --unwired resolves use refs through the registry store:
the parse+resolve phase is extracted from graph build's path into the
shared parse_and_resolve_ops, so a use-bearing document introspects
identically to how it builds (fetch, C29 gate, label resolution) —
retiring both the hard miss and the 'content id' mislabeling of
by-name refs. The engine's subgraph closure contract is unchanged
(store-freedom binds the engine, not the CLI caller).
- ganging a spliced instance's member path refuses with the rule
(OpError::GangOfSplicedInstance) instead of the typo-shaped
no-such-param prose; the leaf-typo case keeps its UnknownParam shape
under a new sibling pin.
- the --node pending note names the follow-up moves on the current
surface (build, then introspect --params / --unwired), and the
authoring guide gains a worked LinComb op-script example
(args-then-bind order, term[i]/weights[i], verified by building it).
refs #339
refs #341
Harvest sweep, batch 4 of 6.
- present_campaign's cell loop gains the non-walk-forward sibling of
the #313 note: a cell whose families are all non-wf and whose every
report traded zero times gets 'aura: note: ... metrics are vacuous,
not a break-even result' on stderr (emit-independent; wf cells keep
their window note, gate-truncated cells stay silent).
- exec's single-run bias leg (always RunData::Synthetic — a direct
blueprint exec never binds real data) notes on a zero-trade run that
the built-in 18-cycle stream is a smoke fixture and points at a
campaign document's data section for a meaningful run.
- blueprint_serde refuses to serialize a still-pending arg-bearing
builder (SerializeError::PendingBuilder naming node type + missing
args + the try_args remedy) instead of emitting an args-free v1
document whose fault only surfaces at load as MissingArg — the
write-side twin of the load-path fence; only the Rust GraphBuilder
path can reach this state.
closes#324
refs #341
Harvest sweep, batch 3 of 6 (second half).
The third externally-tagged variant ('{"fixed":{"distance":..}}',
additive — stored Vol/VolTf documents keep their bytes): doc-tier
validation mirrors FixedStop::new's distance>0 assert (NaN refused)
so the assert is unreachable from a document; stop_rule_for_regime
binds the shipped FixedStop composite; run_blueprint_member stamps
stop_distance beside the vol knobs' precedent and
stop_rule_from_params re-derives Fixed from it (stamps are mutually
exclusive; the pre-#233 no-knob fallback is unchanged). BadRegime
prose widened to name the distance rule. C10's risk-axis paragraph
updates from 'additive when needed' to shipped; glossary gains the
fixed variant.
closes#338
Harvest sweep, batch 3 of 6 (first half).
Composite::declared_taps() walks the blueprint depth-first collecting
(tap name, source wire, column kind), names bare at every depth —
mirroring inline_composite's hoist, where a tap's name IS the flat
graph's load-bearing identifier — and bounds-total over invalid wires
(compile's validate_wiring stays the real gate). The CLI view loads
like --params (FILE gates the authored root name, a content id does
not), prints one 'name node.field Kind' row per tap, and renders the
tap-less case as a stderr note with empty stdout, exit 0.
Closes the recovery-only discovery loop: since #333 the only way to
learn a blueprint's declared tap names was provoking the undeclared-tap
refusal roster.
closes#337
Harvest sweep, batch 2 of 6.
- exec's file-target classification is now a four-way shape peek
(campaign / blueprint / op-script / wrong-kind): a valid-JSON op-script
array refuses at the routing seam with a build-first pointer ('aura
graph build') instead of the blueprint leg's false 'not valid JSON'
claim, and a kind-bearing non-campaign document names the found kind
plus exec's two executable classes. Exit-code classes unchanged
(op-script usage-class 2, wrong-kind runtime-class 1).
- the bare-tap measurement leg renders CompileError::UnboundRootRole by
the signal's own root-role NAME (captured pre-consumption), retiring
the raw Debug-form leak; every other variant keeps the total fallback.
refs #342
refs #339
Harvest sweep, batch 1 of 6.
- graph build's unknown-key refusal now parses the op-list element-by-
element and attributes any per-element parse fault as 'op N (kind):',
immune to serde's deny-unknown-fields token position landing on the
following element (RED-first: the misattribution was pinned failing
before the fix).
- the UnsupportedVersion refusal names the full accepted range
('versions 1..=2'), not just the ceiling.
- a bare unresolved type id inside a data-only project now carries the
same Rust-escalation pointer as the namespaced form ('aura nodes new');
outside any project a bare id stays hint-free (a std-name typo is the
likelier cause there).
closes#336
refs #341
RED-first fixes for the four M42 A/B findings. campaign introspect
--unwired now drills into a present-but-empty presentation section
exactly as it drills into data, listing persist_taps/emit as open slots
— the miss caused three of the cold start's five dead ends (the process
sibling has no analogous nested-required shape, verified). The override
panic catch wraps known non-diagnostic payloads (capacity overflow) in
domain prose while passing real constructor messages through
byte-identical; a kind-mismatched override renders as prose naming
path, expected and got kinds with a write-this example instead of the
ParamKindMismatch debug struct; chart's unknown-handle refusal names
the current trace surface (exec --tap / presentation.persist_taps)
instead of the retired --trace flag.
closes#319
RED-first fixes for the two field-test bugs on the new flag's blueprint
leg, both formerly exit-101 panics. An out-of-domain override value now
renders the node constructor's own message as a prose refusal, exit 1 —
the single-run bootstrap rides the same catch mechanism the campaign
path's #272 per-cell containment uses (a campaign-leg pin confirms the
identical value was already contained there, exit 3, unchanged). The
retired wrapped axis form refuses before the params zip with a
did-you-mean naming the raw node.param namespace and the discovery verb,
mirroring the campaign leg's existing prose family.
refs #319
Architect review over 9cfe296..dc5f174 (drift_found -> all five items
fixed in this close): C24's family-machinery sentence re-amended to
current truth (synthetic walks are reproduce-only inputs since the
builder retirement; superseded text to the sidecar); the ported
generalize test's doc no longer promises the retired CrossInstrument
family handle and its stale section header is gone;
FamilyKind::CrossInstrument carries its C18 dead-machinery/read-back
note in code; live comments in campaign_run, bench surfaces, and
graph_construct name exec instead of the retired verbs; the five
inlined topology_hash copies in the bin test module consolidated into
one test helper. What holds, per the architect: C24<->code gated-intake
lockstep exact, C12's single-hash claim true in code, exec dispatch
matches C14's amended text.
Bench (regression gate) exit 0, all five fingerprints OK:
engine_throughput bars_per_s 14014049 (+0.1%), ingest_throughput
12935441 (-0.5%), campaign_sweep wall 1.407s (+0.9%), campaign_heavy
wall 5.688s (+2.1%), cli_fixed_cost run_ms 3.6 (+6.6%, spawn-floor
jitter under parallel-agent load, report-only). No baseline moved.
Also rides: the reviewer re-check's two comment nits (classify-once
wording, SIGPIPE example verb).
refs #319
In-cycle fixes for the independent review of the #319 package, RED-first
for both Importants. A repeated --override path now refuses at the shared
lexer (exit 2, path named) instead of panicking exit 101 through
reopen_all — which also un-misattributes the campaign leg's collision
prose for duplicates. override_paths' refusal points at aura graph
introspect --params (the retired --list-axes reference swept from every
live comment, one explicitly-historical mention remains). exec classifies
file targets through a Result-bearing peek: unparseable JSON refuses
neutrally before any leg is chosen. RunPresentation (single-variant)
removed; run_campaign_returning's contract note names its one deliberate
usage-class exit; C14's routing sentence and C20's family-builder
paragraph amended to current truth (superseded text to the history
sidecars); bench error strings and deletion blank-line runs tidied.
refs #319
Slice 8, closing the #319 retirement's prose and bench debt. Bench argv
rides exec (fixed_cost single-run probe, campaign surfaces, and the seed
helper that still called the retired sweep at runtime — caught by the
bench run itself); all five fingerprints unchanged against the committed
baselines, confirming record-line parity through the retirement. Live
docs (README, authoring guide, project layout, glossary) describe only
exec + campaign documents + graph introspect --params; three glossary
entries explicitly mark their verb retired. Ledger amendments per the C29
discipline — C25's executor-verb-set re-settled (exec + the --override
residue), C14's dual grammar reduced to the one dispatch, C24's
document-built runs/families + gated-intake route list (lockstep with the
code comment), C12's override clause and identity anchor repointed to the
runner-layer hash computation, C18/C22/C27/C01 mention rewrites —
superseded sentences moved verbatim to the history sidecars (c25's
created). Stale dual-grammar doc comment on is_blueprint_file rewritten.
refs #319
Slice 7 of the #319 retirement. aura-runner's blueprint_sweep_family /
blueprint_walkforward_family / blueprint_mc_family lost their only
production callers with the quintet and are gone (family.rs 889 -> 199
lines) together with their zero-caller helpers; reproduce's live inputs
(DataSource, showcase/walkforward sources) stay. Coverage was ported
first: member-computation identity + family-shared topology_hash now pin
through a campaign-document twin in tests/exec.rs, reproduce's multi-member
and MC-seed-label branches re-mint their fixtures directly. The CLI-side
topology_hash helper goes with them — its keep-rationale was disproven
(the record line's hash is computed in the runner's member machinery;
minuted on the issue) — and c28_layering's file list drops the stale
verb_sugar entry. The synthetic per-seed MC family remains an intentional
retirement casualty, recorded in the test dispositions.
refs #319
Slice 6 of the #319 retirement, the destructive half. The five flag verbs,
their five Args structs and dispatch arms, the whole argv->document
translator (verb_sugar.rs, 1257 lines) and 33 quintet-only helpers are
gone; retired verbs now refuse at the clap layer as unknown commands.
main.rs shrinks 4445 -> 2764 lines. The gated-intake route list narrows to
the canonical layer + exec's blueprint leg; exec's own refusal prose no
longer names dead verbs; test seeding recipes ride graph register.
cli_run.rs walked the full disposition table: 94 tests retired with their
dead flag surface (each with a named green twin where the property
survives), 75 ported — exec ports keep assertions byte-identical, campaign
ports inherit their golden grades unchanged (walkforward's 90/30/30 real
roller and the archive window clipping the sugar applied silently are now
explicit document fields — the equivalence held exactly, one reconstructed
walkforward pooled expectancy_r within 1e-12, cause documented inline).
Surface deltas recorded by the ports: per-cell refusals are #272 contained
faults (exit 3 + warning) rather than hard exit 1; the synthetic per-seed
mc family and generalize's merged cross-instrument family record were
sugar-only constructs and retire with it (the campaign record's
generalizations[] carries the data).
refs #319
Slice 5 of the #319 retirement. The scaffolder's project CLAUDE.md and the
CLI concepts help now teach only the surviving surface (exec over both
document classes, --override, graph introspect --params for discovery;
traces via exec --tap / presentation.persist_taps). Sweep vehicles moved
onto campaign documents driven by exec (project_new's starter quickstarts,
cli_broken_pipe — whose bare-sweep EPIPE probe was latently vacuous and now
streams a real 4-member campaign, project_sweep_campaign's real-data leg,
graph_construct's gang-axis parity pair as paired one-cell campaigns);
pure discovery sites became graph introspect --params, and the line-identity
test kept its literal --params expectations minus the sweep half. Help pins
rewritten; the scaffold template's in-src pin follows the new bytes.
refs #319
Slice 4 of the #319 retirement: 34 mechanical argv swaps across six test
files (tap_recording 12, project_load 13, run_measurement 3, measure_ic 1,
run_refuses_unrunnable_blueprint 1, graph_construct 4 run-halves) — every
site drove run as a synthetic vehicle with at most --tap, so no one-cell
campaign conversions were needed; assertions untouched. graph_construct's
sweep sites and the remaining vehicle files stay for the next slice.
refs #319
Slice 3 of the #319 retirement. The campaign leg's --override injects a
single-value axis over the named bound param into every strategy entry —
after intrinsic validation, before the referential gate, so an unknown
path speaks validate_campaign_refs' existing did-you-mean prose and a
kind mismatch its existing fault; collision with a document-declared axis
refuses (an override overrides a bound value, never an axis, exit 2).
The stored bytes and the recorded campaign_id stay those of the original
document; the audit record of an override is the raw member manifests.
The walk-forward zero-trade aura: note: diagnostic now has a campaign-
path producer call (present_campaign per-cell), so it survives the
quintet's removal at today's strength.
refs #319
Slice 2 of the #319 retirement. The blueprint leg now carries both run
shapes: a bias signal via run_signal_r, a no-bias measurement blueprint
(>=1 declared tap) via run_measurement — record lines byte-identical to
run's. --override NODE.PARAM=VALUE (repeatable) is the one deliberate
sugar residue: validated through the shared override_paths prose, the
closed-guard runs on the reopened probe (space == override set), values
bind in reopened param_space order so the manifest records them raw under
params ("what varied"). A measurement blueprint refuses --override
explicitly (its space is not wrap_r-wrapped) rather than dropping it
silently.
refs #319
The #319 retirement's constructive half, slice 1: a new top-level
`exec <target>` executes a campaign document (file, register-then-run, or
64-hex content id — byte-behaviour of the retired `campaign run`, which no
longer parses) or a fully-bound signal blueprint (synthetic single run,
record line byte-identical to `aura run`'s, root-name gate migrated).
Flag discipline is per-leg: --parallel-instruments on campaign targets,
--tap on blueprint targets (refused on campaign targets with prose).
File routing peeks the document's top-level "kind" key: is_blueprint_file
alone cannot tell two .json classes apart.
--override (the #246 residue) and the measurement leg follow in later
slices; run/sweep/walkforward/mc/generalize stay alive until the removal
slice.
refs #319
Reshape the aura data namespace around one principle: every verb emits
JSON, no format flag (the headless-agent posture #264 named).
- aura data info <symbol> (new): one flat JSON object — symbol plus the
six neutral geometry fields (digits, pipSize, tickSize, lotSize,
baseAsset, quoteAsset) from DataServer::symbol_meta when the
<SYMBOL>.meta.json sidecar yields geometry; a symbol is known via bar
files OR sidecar (a sidecar-only symbol still reports its geometry);
an unknown symbol refuses with prose, exit 1. description stays out
until the neutral reader exposes it (data-server#4).
- aura data list: NDJSON — one JSON string per line; an empty archive
emits zero stdout lines, with the human affordance moved to a stderr
note in the #278 class vocabulary.
- aura data coverage: retired — #272's per-cell fault isolation subsumed
the prophylactic pre-run check; aura-runner's data_coverage_report and
its presentation helpers go with it, while interior_gap_months stays
as the campaign path's single gap walk (#295).
The three derived design decisions (description omitted, broader
refusal prose, empty-archive stderr notice) are minuted on the issue.
closes#273
Fieldtest finding B1: the one remaining authored-envelope FILE route,
sweep --list-axes, skipped the root-name gate — a hand-edited
"name":"../escape" listed mangled, non-bindable wrapped axis names
at exit 0 while every sibling intake refused. The discovery surface now
gates before any axis line prints (exit 1, shared prose), restoring
both the #331 every-authored-intake claim and #328's discovery-surface
identity. RED-first: the e2e refusal test was written and observed
failing (exit 0, axis lines printed) before the fix.
Re-enumeration of every read_to_string-fed envelope load confirms the
class is now closed; C24 additionally ratifies the fieldtest's SG1 —
the refusal core sentence is byte-uniform across sites while the
leading context prefix varies by site convention.
refs #331, #328
Architect drift review (drift_found) resolved fix-path on all three
items. [high] The sweep/walkforward/mc family builders and
validate_and_register_axes read authored envelope files and
put_blueprint'd them ungated — falsifying the C29 exemption's premise
that the store is populated through gated intakes; all four sites now
gate the root name before the family/axis build (placing the gate
after the build produced a misleading unknown-axis error instead of
the gate refusal — RED-verified), with an e2e pin on the synthetic
sweep route. The decision was gate-the-routes, not weaken-the-prose:
the exemption's premise is the invariant, not a wording choice.
[medium] C24 now states the gate rule as the class it is (the op
intake plus every CLI intake reading an authored envelope from a
file; store read-back exempt, C29). [low] C23 reconciles
names-as-debug-symbols with the render name's one operational role
(trace-directory key — never compilation, identity, or execution
semantics).
Ratify: aura-bench cli_fixed_cost baseline re-pinned
(run_line_fnv 6bb0d796f760d140 -> 9bdbc3acf7b2926a). The moved metric
was caused by the #328 tidy (4474814), which switched single-run
manifest defaults from wrapped to raw axis names after that cycle's
bench rerun had already passed; bbac29d and this cycle's HEAD produce
fingerprint-identical record lines (verified with both binaries on
one scratch project), so this cycle only inherits the stale pin. All
five surfaces fingerprint OK after the re-pin.
refs #331, #328
An eleventh op-script op {"op":"name","name":"<n>"} sets the composite
render name (engine Op::Name + CLI OpDoc mirror); omitting it keeps the
default "graph" byte-identically. The default leaked everywhere one
research project has more than one strategy: every store entry named
graph, every default tap recording sharing traces/graph/, and two
use-splices of unnamed blueprints colliding on the default instance
identifier — the authored name dissolves all three through existing
mechanics (no downstream site edited).
Mechanics: at-most-once per script (doc-op precedent), position-free
(read only at finish). A shared deterministic name_gate (aura-engine:
non-empty, no path separators, not . or ..) guards every seam where a
name is born from authored data, because the name keys a trace
directory unsanitized: the op intake (GraphSession::set_name) and all
four CLI fresh-file envelope intakes (register, introspect
--content-id FILE, the bare graph FILE viewer, aura run FILE — the run
route reached begin_run(signal.name()) ungated, and introspect
--params FILE rode the same parse). Store read-back (reproduce, use
resolution, introspect/params by content id) stays deliberately
ungated — C29: registered artifacts are never retroactively
invalidated — pinned by a test that plants a bad-root-name blueprint
via the registry API and asserts introspect --params still answers.
Refusal prose is single-sourced (name_gate_fault_prose) so every seam
reads identically, op-indexed on the op route.
Identity semantics: the authored name hashes into the content id (a
named document is a different document) and never into the identity id
(names are stripped as debug symbols, C23) — pinned by a twin test.
The previously untested use-splice instance-name default
(construction.rs) got its ratifying pin before the collision claim
leans on it. The registry label (register --name) stays orthogonal.
Review rounds caught and fixed: the envelope gate initially fired on
store read-back (C29 violation at the introspect surface), and the
run/params fresh-file routes were unenumerated intake seams — closed
with the call-site classification now recorded in the wrapper docs.
Verification: cargo build/test/clippy -D warnings all green (99 test
targets, 0 failures, independently re-run); the new run-route test was
RED-verified by hand-removing the gate.
closes#331
refs #328, #311
Spec: blueprint-name-op (fork minutes on #331)
Architect drift review (cycle a3785a6..b3b7115) found the cycle's raw
switch incomplete on two high items; both fixed forward here rather
than minuted as residue, since #319 builds on this namespace:
- The real/campaign executor still minted WRAPPED manifest.params and
manifest.defaults (AC2 overclaim in the feat commit body). The mint
seams now reshape to raw on every route (member.rs raw_bound_defaults,
runner.rs manifest_space mirror of the family.rs pattern); the fix
surfaced a latent regression in persist_campaign_traces, which keyed
point_from_params by exact wrapped name and is now translation-aware
(same recipe as reproduce_family_in). Old on-disk families stay
replayable (C29; reproduce reads both shapes).
- The walkforward --axis intake silently accepted wrapped names — the
silent alias fork 2 rejected. It now runs the same
refuse_wrapped_synthetic_axes preflight as the plain synthetic sweep
(new e2e: aura_walkforward_synthetic_blueprint_refuses_a_wrapped_form
_axis_name); family builders translate raw names onto wrapped
SweepBinder slots on all three entry points. C24's "retired from the
surface" claim is thereby made true instead of softened.
- Low items: stale graph_construct docstring updated; new
graph_params_and_sweep_list_axes_are_line_identical pins the two
discovery surfaces against format drift.
- Bench infra: aura-bench's campaign_sweep surface seeded wrapped axis
literals and was correctly refused by the new intake — seeds converted
to raw; full bench rerun: all 5 surfaces fingerprint OK, deltas within
load noise (report-only, no baseline update needed).
Library-unit tests calling family builders directly keep wrapped input
(valid: translation is idempotent on exact wrapped names; intake-level
refusal is the user contract). Suite + clippy verified green post-fix.
refs #328
Reconciles the sweep-axis namespace (closes#328): the raw form
<node>.<param> (splice paths keep their interior path) is now the only
user-facing axis name — printed by --list-axes, accepted by --axis on
BOTH sweep routes, recorded in the synthetic run manifest, discovered
by graph introspect --params (bound params included, default= lexicon),
and validated by campaign documents. The wrapped <blueprint>.<...> form
is retired with translation refusals on both seams.
Mechanics:
- classify_axis_intake (aura-runner axes.rs): explicit raw-first
acceptance predicate shared by both --axis intake routes. Deliberately
not built on raw_matches_wrapped, whose equality branch would have
accepted the retired form silently (skeptic finding, minuted on #328).
Wrapped hit -> translation refusal naming the raw candidate (shared
prose builder wrapped_axis_refusal, one literal for both routes).
- The synthetic route resolution moves to the raw frame the real route
(bind_axes) already uses: override_paths/wrapped_bound_overrides_of
match via raw_matches_wrapped (also removing a latent slicing panic),
blueprint_sweep_family translates raw names onto wrapped SweepBinder
slots, manifest.params records raw keys (name-only reshaping, zip is
positional). The two sweep routes now share one convention.
- reproduce translates recorded manifest names raw-or-wrapped onto the
wrapped space: old registered artifacts stay replayable (C29), new
raw manifests round-trip.
- introspect --params gains the bound default= lines via the same
render_value the --list-axes bound pass uses - the two discovery
surfaces are byte-identical by construction.
- AxisNotInParamSpace carries raw_candidate; ref_fault_prose renders
the did-you-mean iff present. Sweep-terminal bind errors
(MissingKnob/KindMismatch) render the raw knob name (render seam
only; acceptance criterion: no user-facing surface prints a wrapped
axis name).
- Downstream consequence fixes: scaffold quickstart example spoke
wrapped; ~90 wrapped test literals converted to raw across six test
files (deliberately-bogus refusal names and the untouched internal
walkforward wrapped route kept, with comments).
Fork decisions and the skeptic correction are minuted on #328
(comments 2026-07-24/25). Grounding-check PASS first attempt (12
assumptions ratified). Review (opus): translation ambiguity cleared
(uniform single prefix per blueprint), reproduce both-shapes traced;
repair pass added the synthetic-route refusal e2e + the shared prose
builder. Verified: cargo test --workspace green (99 targets, 0
failures), clippy -D warnings clean.
refs #246, refs #319, refs #331
The data plane reaches non-scalar structural config (closes#271, the
final Data-authorability-boundary item). Op::Add gains args
(string pairs, applied via try_args after resolve and before the bind
loop — no pending builder ever enters a graph; an arg-bearing type
without args refuses as MissingArg). OpError::BadArg carries the
precise ArgOpError; the CLI renders it as prose naming the arg, its
kind, and the kind's closed hint (exit 1, #175 content-fault class).
Serde: PrimitiveData.args (skip-if-empty — args-free documents stay
byte-identical, golden-pinned, so every existing content id is stable
per C18) with a data-driven format version: the writer emits 1 for
args-free documents and 2 when any primitive recursively carries args;
the loader accepts 1..=2. The old v2-refusal pin flips deliberately to
v3 (named contract change). Identity JSON keeps args — they are
structural, and the Rust configured() path and the args op-script
bridge to the same identity id (pinned).
Roster: Session / LinComb / CostSum enter (33 -> 36, both count pins;
the LinComb-rejection assertion flips to the resolved side). introspect
--node shows arg rows (name, kind, hint) plus the pending note;
OP_REFERENCE's add row teaches the args form. Registry comment
de-staled (LinComb is now roster-reachable).
Sequencing note: this commit lands op seam and roster together so no
intermediate tree state exposes a pending builder.
The op-script gains its tenth verb end-to-end: OpDoc::Use carries a tagged ref
({"content_id": id-or-unique-prefix} | {"name": label}); DTO conversion resolves
it (label sidecar / #302 prefix semantics / verbatim id), fetches, deserializes
eagerly (a vocab/parse failure fails fast naming the id prefix — review
finding, no swallowed .ok()), re-walks the C29 doc gate over the fetched
composite (a doc-less store entry cannot enter a NEW composition; existing
reads stay valid — no retroactive invalidation), echoes every resolution as
the existing benign marker (aura: note: use "<instance>": <ref> -> <id>,
stderr; stdout stays clean payload), and hands the session a pure cached
lookup closure.
graph register <file> --name <label> records the label and echoes repoints
(label "x" -> <new> (was <old>)); graph introspect --registered lists
label / 12-char id prefix / root doc line — the discovery surface for what
use can reference. Refusals are by-identifier, name their rule, and enumerate
the label set (empty store reads as prose); all op-list content faults are
runtime exit 1 per the pinned #175 attribution, argv misuse stays exit 2.
Ten new/extended e2e tests cover label round-trip + repoint, echo discipline,
unknown-label enumeration, the C29 refusal shape, end-to-end sweep axes
through a spliced instance (graph.<instance>.<node>.<param>), the open-pattern
build flow, and the standalone-run bootstrap refusal (asserts the existing
Debug-form fault; an index->name rendering at the run boundary is recorded
residue, not silently claimed). OP_REFERENCE and its help pin move 9 -> 10.
Empirical note for the record: aura run's bias arm re-roots the loaded signal
via wrap_r, so the standalone-run refusal for an open pattern is exercised on
the bare-tap measurement path; the bias path surfaces openness as a role-
binding refusal instead (pre-existing shape, untouched).
closes#317
Two fold-selector fieldtest findings on the --tap seam, RED-first.
Skipped-tap note (#334): an explicit --tap plan replaces the record-all
default entirely (C27: unbound is inert) but gave no runtime signal that
unlisted declared taps went unbound — a forgotten tap surfaced only as a
missing trace file. bind_tap_plan's unbound arm now emits the C14 benign
note (aura: note: declared tap "<name>" unbound this run) per skipped
tap, exit unaffected. No new carrier: the arm is only reachable when
plan.default_named is None (an explicit plan) — under record-all it is
structurally dead, so the default emits nothing. Emitted runner-side
beside the existing eprintln registers; the runner->CLI print migration
is #297.
Undeclared-tap refusal (#333): --tap on a tap the blueprint does not
declare was refused without naming the valid taps, forcing the author
back into the blueprint JSON. TapPlanError::UnknownTap now carries the
declared-tap roster and Display appends it, mirroring UnknownLabel's
fold-roster idiom (C29: a refusal names the closed set it checks
against). Decision minuted on #333 (refusal roster over a new introspect
view).
Rider (#335): the fold roster doc lines now state the summary-row
timestamp — the accumulator folds (count/sum/mean/min/max) land one row
at the last warm (contributing) value's ts, and last gains the same
at-its-own-timestamp wording first already carried. The ratification
record lands in the sibling ledger/glossary commit.
closes#334closes#333
Two fold-selector/op-script surface-honesty bugs from the 2026-07-24
fieldtests, both RED-first.
--folds (#332): graph introspect --folds rendered the aura-std FoldKind
table — capitalized ids (Mean) and no record entry — while aura run
--tap resolves labels against the fold-registry roster (lowercase,
record included). The discovery surface the --tap help points at thus
misdescribed the accepted input twice. It now renders
FoldRegistry::core().roster() (label + doc per entry); the pinned
FoldKind-rendering test was rewritten to pin the roster form (8 rows,
no capitalized id leaks, record present).
Unknown op-list keys (#326): a typo'd key in an op-list element
("params" for "bind") was silently dropped and the wrong graph built
with zero signal, contradicting the closed-vocabulary posture (C25).
OpDoc now carries serde deny_unknown_fields; the refusal names the
offending key. Exit class is 1, not the issue's suggested 2: op-lists
arrive on stdin, and the pinned #175 attribution principle classes
stdin-content faults as runtime — the new refusal fires from that same
deserialize. Blast radius verified: all 21 committed *.ops.json under
fieldtests/ still build cleanly.
closes#332closes#326
campaign introspect (--block std::process_ref and --unwired) described
the slot as "content id of a process document" — a plausible bare-string
reading the validator rejects; the required {"content_id":"..."} tagging
appeared in no introspection output (binary-only M1 fieldtest finding).
The SlotKind::ContentRef label and the open-slot hint now lead with the
tagged { content_id } shape. Deliberately narrower than the issue's
suggestion of mirroring std::strategy's { content_id } | { identity_id }
form: validate_campaign refuses an identity_id process ref
(DocFault::ProcessRefMustBeContentId) and a pinned test forbids
advertising it, so the hint states the one accepted key only (C29
honesty; correction minuted on the issue). RED-first: the extended
process_ref_describe_advertises_content_id_only assertion failed against
the old label before the fix.
closes#329
Architect (cycle 9124275..1baee77) holds: the C27 boundary is preserved
and ledger-anchored (selection = run-mode authority, fold growth stays
Rust/role 2; validation solely in the shared bind_tap_plan seam before
store I/O; unlisted taps inert); C1 byte-identity pinned for the
explicit all-record plan vs the record-all default; ledger discipline
kept (superseded C27 sentence verbatim in the new history sidecar); the
C25 projection tension was explicitly adjudicated in C27's Current
state, deferral #312/#327 minuted.
Drift resolution:
- [medium] C14 exit-class: --tap-content refusals (unknown label /
undeclared tap) exit 1 through the assembly crate's runner-side
registers while C14 assigns argv/blueprint-content faults to class 2.
Pre-existing register family (#283 duplicate-tap, #286 exposes-
neither). RESOLVED AS TRACKED-FORWARD to #297 (RunnerError
propagation is where the CLI regains class authority; evidence
comment with concrete sites posted there). A CLI-side shadow
pre-validation was rejected: it duplicates the bind_tap_plan seam
and can drift.
- [low] FIXED here: the three hand-rolled `aura run` usage strings and
the loaded-blueprint-grammar comment now name --tap.
- [low] ADJUDICATED equivalent-as-delivered: the fold-mean expectation
is derived from the pinned R_SMA_PRICES const rather than a second
record run — an independent derivation of the same series; the
spec's intent (fold row equals the series mean) is pinned.
Regression: aura-bench report-only, exit 0, all five surfaces within
tolerance, fingerprints OK (engine 13.78M bars/s Δ-1.6%, ingest 13.13M
Δ+1.0%, campaign_sweep rss 93.9MB Δ+1.9%, campaign_heavy rss 106.9MB
Δ+5.4%, cli run 3.6ms Δ-0.3%; loadavg-5.3 warning noted, report-only).
No baseline update.
Cycle spec + plan + the spent m1 fieldtest spec removed from the
working tree (git-ignored files, shell rm per lifecycle).
refs #310, refs #297
Decide #310's data-authorability fork as (b), split by run-mode
authority: selecting a Named fold subscription is data-reachable on the
one-shot run path; adding a fold stays a Rust entry (role 2, C25).
`aura run` gains a repeatable --tap TAP=FOLD selector feeding the
existing TapPlan seam: no flag keeps the record-all default; any flag
replaces the plan entirely (unlisted taps stay unbound — C27 inert,
non-error). Validation stays in bind_tap_plan (roster-enumerating label
refusal, typed UndeclaredTap, both before store I/O); parse-level
refusals (malformed pair, duplicate tap) exit 2 via the verb's usage
convention.
Rejected alternatives (minuted on #310 with the skeptic-pass
rationale): an op-script/blueprint carrier (C27 forbids recording
policy in the serialized fragment); ratifying Rust-only (softens a
designed-in promise with effort as the only rationale). The
campaign/document carrier and the persist_taps/declared-tap namespace
reconciliation are deferred to the Measurement-reachable milestone
(#312/#327, minuted on #312).
Ledger: C27 Current state records the boundary; the superseded sentence
moved verbatim to the new c27-declared-taps.history.md sidecar; the
glossary tap-plan paragraph updated, fixing the pre-existing inaccuracy
that named `aura measure` a record-all-passing verb (it is the post-hoc
IC analysis and constructs no tap plan).
Verification: 5 new unit tests (parse branches), 5 new binary-level
selector tests incl. a byte-identity pin of an explicit all-record plan
vs the record-all default (C1), 1 measurement-arm test; full workspace
suite green; clippy -D warnings clean. Independent opus diff review:
approved, no Important/Minor findings (nits: inert-arm generic refusal,
a=b=c falling to the roster refusal — accepted as-is).
closes#310
Independent post-cycle review (opus, fresh context) on the branch diff
returned one Minor and one Nit; both adjudicated against the source and
fixed:
- Minor: the concepts paragraph and the sweep/walkforward/mc long helps
claimed the document desugaring unconditionally, but only the --real
branches route through verb_sugar — the synthetic branches execute
in-process and register no documents. The claims now carry the --real
qualifier (and the paragraph names the in-process synthetic path).
generalize keeps the unconditional wording: its only mode is --real
(usage requires it), so the claim is true as stated.
- Nit: the vocabulary listing's `None` fallback would have silently printed
a bare, C29-incomplete row if a rostered type id ever failed to resolve.
Replaced with a loud expect — an invariant breach should not degrade into
exactly the surface this cycle exists to remove.
refs #315
The shipped wording lifted #319's loose "quintet is pure translation"
phrasing into "the research verbs (run, sweep, ...) are sugar — each writes
registered process/campaign documents". Source check (dispatch_run): only
the four verb_sugar bridges (sweep, walkforward, mc, generalize) generate
and register documents; `run` is a direct single backtest, exactly why its
own long help already avoided the desugaring claim. The paragraph now names
the four document-bridged verbs and places run as their single-backtest
sibling — consistent with `run --help`.
Found by re-auditing the shipped prose claims against the source after the
owner asked how I know no error was made (skills-less-cycle experiment,
refs #315): prose claims in help text are exactly the class the green test
suite cannot semantically verify.
The binary is the only always-present teacher (C29); the 2026-07-22 external
field test showed the help teaching nothing but the verb inventory. This
cycle makes every closed vocabulary the binary ships listable with a one-line
meaning, and opens the help with the concepts the field agent had to discover
forensically.
#315 — help + introspection:
- `aura --help` opens with the two-layer concepts paragraph (research verbs
as sugar over registered process/campaign documents; the directly
authorable data plane; bias-as-target-position + protective-stop execution
model in R; how traces come to exist).
- Each document-bridged sugar verb's long help names the process shape it
desugars to (sweep -> std::sweep; walkforward -> [std::grid,
std::walk_forward]; mc -> + std::monte_carlo; generalize -> [std::sweep
(selection), std::generalize]). `run` is not verb_sugar-bridged, so its
help points at the canonical document-first form instead of claiming a
desugaring it does not perform.
- `graph introspect --vocabulary` lists each node type with its schema doc
(bare names taught nothing); `--node <T>`'s head line carries the meaning.
- `graph introspect --folds` (new): the tap-fold vocabulary with bind rule,
output kind, and meaning — it existed only as source comments. The rows
live in aura-std (`fold_vocabulary`), unit-pinned against
`fold_binds_at`/`fold_output_kind` so the prose cannot drift from the
executable rules, and doc_gate-checked (C29 entry seam). The discovery
surface is graph introspect because folds bind at graph-declared taps
(C27); the document layer still refuses a folds slot pending #310.
- `process introspect --metrics` lines carry each metric's meaning (the doc
column existed since #316 but was never printed).
- `campaign introspect --block std::presentation` lists the four persisted
taps' meanings under the persist_taps slot.
- The scaffolded project CLAUDE.md teaches the execution semantics (bias as
held target, protective stop, R) and the document data plane.
#323 — C29 authoring surfaces:
- `introspect --unwired` enumerates the optional `description` envelope slot
for both document kinds (absent-only, like risk/cost); the campaign lib
fixture gains a description, and the risk/cost-bound e2e binds description
too so "no open slots" keeps meaning complete.
- `graph build --help` carries the op-list reference: all nine op kinds with
fields and a worked element each (OP_REFERENCE lives beside OpDoc so a new
variant is one screen from its help line).
- The authoring guide's S0 worked literal is byte-pinned to the scaffold's
LIB_RS template (rendered with the guide's my_lab namespace); the scaffold
e2es compile that template for real, so the pin is a transitive compile
guard. The template gains the guide's C29 comment, the guide gains the
template's rename note — one worked example, two surfaces, zero drift.
Deliberately out of scope (ratified #319, 2026-07-24): no doc-strings for
the quintet's 48 flag fields — the sugar surface retires; only the pointer
lines above bridge toward the document layer.
Verification: full workspace suite green, clippy -D warnings clean,
aura-bench exit 0 with all fingerprints OK (help_ms +1.6% — the longer help
text, in tolerance), cargo doc warning count unchanged (7 pre-existing).
9 new e2e pins in tests/help_self_description.rs; guide + CLAUDE.md pins in
scaffold unit tests; fold-vocabulary rules + doc_gate unit-pinned in
aura-std.
closes#315closes#323
Milestone-fieldtest friction: a single-OOS-window walk-forward printed
"all 1 walk-forward windows recorded zero trades" (captured RED
evidence: fieldtests/milestone-stderr-honesty/captured/
example2_singular.err). The message text moves into a pure
zero-trade_note_text builder — singular "the single walk-forward
window recorded zero trades", plural unchanged so the existing e2e
phrase pins hold — with a unit pin on both forms. refs #278
Orchestrator probe after 51096a3: blueprint_mc_family is the one
remaining family builder running members bare — a closed poison
blueprint (SMA length bound 0) through aura mc --seeds 2 exits 101
with a raw panic. RED e2e pins the shared contract: no 101, an
aura: warning: fault line, no panic text or crates/ path, exit 3.
Single-seed escape confirms the bare seam (not the parallelism) is
the cause. refs #278
Milestone-fieldtest bug (source-blind run, 2026-07-24): a member fault
on the synthetic walkforward/sweep path escapes as an uncaught panic
(exit 101, no class marker, internal source path leaked) while the
real path contains the identical fault via the #272 boundary as
aura: warning: + exit 3. Cause: the family builders in
aura-runner/src/family.rs run members bare, without the catch_unwind
containment that lives only in aura-campaign/src/exec.rs. Two RED
e2es pin the derived contract (decision log on #278): no exit 101,
a warning-marked fault line, no panic text or crates/ path on stderr,
exit 3 per the C14 partition. refs #278
Iteration stderr-markers-2, tasks 3-4 of the stderr-class-markers plan.
aura-runner's four continuing-run diagnostics join the class grammar as
hand-written literals (the #295 boundary keeps the macro owner in the
presentation crate): the stale-dylib warning gains the `aura: ` namespace,
and the three tap/no-nominee notices become `aura: note: ` — the latter
two extracted into pure note-builder fns with unit pins. The two e2e
pins on the cost-model tap string moved with the retag
(research_docs.rs). A new project_nodes e2e exercises the real stale-
dylib load path: fresh build emits nothing, a source newer than the
dylib emits exactly one class-marked warning and still succeeds.
The #313 notice ships on both walk-forward paths: `aura: note: all N
walk-forward windows recorded zero trades` before the summary JSON,
exit 0 untouched (a null result is a valid research result, #198).
Wording and condition live in one shared diag helper taking the
per-window trade counts (orchestrator tidy of the reviewer's
duplicated-predicate residue); the call sites keep only the
path-specific field extraction. Three e2es pin it: the equal-length
degenerate on the synthetic path (exactly one note, no warnings, exit
0, summary unchanged), a traded negative twin, and the real/sugar-path
degenerate gated on the GER40 archive.
refs #278, refs #313
Iteration stderr-markers-1, tasks 1-2 of the stderr-class-markers plan
(spec signed via grounding-check PASS, decisions logged on #278).
The two stderr diagnostic classes become machine-separable (refs #278):
a new crate-internal diag module owns the grammar as note!/warning!
macros (`aura: note: <text>` benign continuing-run diagnostic,
`aura: warning: <text>` recorded fault the run survives); the four
undifferentiated sites are retagged — gate-emptied cell -> note,
failed cell / walkforward cell / mc cell -> warning — and the four
scaffold literals migrate onto the shared macro (output bytes
unchanged). The always-on record summary line gains the bare `aura: `
namespace it was missing; bare-prefix stays the vocabulary for error
lines accompanying a non-zero exit (C14 partition) and plain info
lines. Exit codes unchanged throughout.
Verification: RED->GREEN warning pin on the per-cell-fault e2e; new
gate-emptied e2e (impossible n_trades threshold) pins the note marker,
exit 0, and the summary-line prefix; scaffold e2es and the full
workspace suite green; clippy -D warnings clean on aura-cli. The C28
shell-module roster test gained the diag.rs entry.
Known residue for the next slices: aura-runner literals (stale-dylib
warning, three tap/no-nominee notes) and the #313 zero-trade notice
follow in tasks 3-5. The verb_sugar warning call sites remain e2e-
uncovered (a hostless walkforward fault fixture is structurally
unreachable — single-instrument validation refuses before the cell
loop); their grammar is pinned via the shared macro and the campaign
warning e2e.
refs #278
claude