1 reporting.selector_screen
reporting.selector_screen
Frozen selector-screen reporting over canonical plan and store admission.
Every intended attempt remains visible. Distribution shadows are conditional on factual states of their originating arm, never pooled across producing policies. Scorer-derived interior prefixes and sealed independent terminal audits retain separate meanings. This adapter makes no corpus admission or promotion decision.
1.1 Classes
| Name | Description |
|---|---|
| ArmIndexSummary | Equal-scene canonical reduction within one factual originating arm. |
| AttemptPrefix | One intended unit and horizon, including absent or nonfinite endpoints. |
| OriginRow | One immutable inspection observation with its producing arm and unit. |
| PrefixSupport | Full intended support denominator for an arm and acquisition prefix. |
| RootPairing | Observed root-shell equality and complete eight-arm support per block. |
| SelectorAttempt | One immutable planned attempt; source binding is strict JSON bytes. |
| SelectorReadout | Captured source-bound screen; all nested evidence is immutable. |
1.1.1 ArmIndexSummary
reporting.selector_screen.ArmIndexSummary(arm, intended_attempts, summary)Equal-scene canonical reduction within one factual originating arm.
1.1.1.1 Attributes
| Name | Description |
|---|---|
| arm | Originating acquisition arm; shadows never cross this boundary. |
| intended_attempts | Full planned arm denominator, eight for this frozen screen. |
| summary | Conditional index mean with retained at-risk and finite support counts. |
1.1.2 AttemptPrefix
reporting.selector_screen.AttemptPrefix(horizon, reached, gain)One intended unit and horizon, including absent or nonfinite endpoints.
1.1.2.1 Attributes
| Name | Description |
|---|---|
| horizon | Acquisition prefix length, 8/12/16. |
| reached | Whether a validated factual chain reached this prefix. |
| gain | Finite scorer-derived cumulative raw gain, otherwise unavailable. |
1.1.3 OriginRow
reporting.selector_screen.OriginRow(arm, unit_id, row)One immutable inspection observation with its producing arm and unit.
1.1.3.1 Attributes
| Name | Description |
|---|---|
| arm | Factual originating arm, distinct from the shadow temperature/weight. |
| unit_id | Planned unit identity; local store row IDs never become global IDs. |
| row | Validated canonical observation, including its original local identities. |
1.1.4 PrefixSupport
reporting.selector_screen.PrefixSupport(
arm,
horizon,
intended_attempts,
reached_attempts,
defined_attempts,
scene_equal_gain,
defined_scenes,
)Full intended support denominator for an arm and acquisition prefix.
1.1.4.1 Attributes
| Name | Description |
|---|---|
| arm | Originating acquisition arm. |
| horizon | Requested acquisition prefix. |
| intended_attempts | All planned units in this arm, including failures and unreached. |
| reached_attempts | Validated chains physically reaching this prefix. |
| defined_attempts | Reached prefixes with finite scorer-derived gain. |
| scene_equal_gain | Equal-scene mean conditional on finite prefix support. |
| defined_scenes | Independent physical scenes supporting the conditional mean. |
1.1.5 RootPairing
reporting.selector_screen.RootPairing(
block,
sample_key,
observed_arms,
status,
fatal_scientific_pairing_failure,
)Observed root-shell equality and complete eight-arm support per block.
1.1.5.1 Attributes
| Name | Description |
|---|---|
| block | Planned source-root block. |
| sample_key | Exact matched source sample. |
| observed_arms | Arms with a selected root shell available for comparison. |
| status | matched_complete, matched_partial, unavailable, or mismatch. |
| fatal_scientific_pairing_failure | True on observed shell or independent prepared-root identity disagreement. |
1.1.6 SelectorAttempt
reporting.selector_screen.SelectorAttempt(
unit_id,
block,
arm,
sample_key,
scene,
status,
reason,
elapsed_seconds,
prefixes,
root_shell_sha256,
terminal_audit_status,
binding_json,
terminal_audits_json,
geometry_json=b'{}',
)One immutable planned attempt; source binding is strict JSON bytes.
1.1.6.1 Attributes
| Name | Description |
|---|---|
| unit_id | Unique planned attempt identity. |
| block | Matched source-root block, independent of acquisition arm. |
| arm | Originating acquisition arm, preserved for shadows as well as facts. |
| sample_key | Exact source sample from the frozen shard entry. |
| scene | Physical scene copied from that source row. |
| status | Canonical receipt/lifecycle disposition, including unreached. |
| reason | Recorded failure or lifecycle explanation. |
| elapsed_seconds | Observed outer receipt duration, never inferred for missing attempts. |
| prefixes | Every requested prefix; no missing unit is imputed as zero gain. |
| root_shell_sha256 | Exact observed root shell digest, absent without a selected root state. |
| terminal_audit_status | Canonical deep-shard endpoint seal status, independent of prefix support. |
| binding_json | Immutable source, writer, producer, store and target-pool identities. |
| terminal_audits_json | Sealed terminal-only receipt projections; absent audits serialize as []. |
| geometry_json | Canonical trajectory/BEV projections as captured strict JSON bytes. |
1.1.6.2 Methods
| Name | Description |
|---|---|
| to_jsonable | Return a fresh mapping without exposing captured mutable containers. |
1.1.6.2.1 to_jsonable
reporting.selector_screen.SelectorAttempt.to_jsonable()Return a fresh mapping without exposing captured mutable containers.
1.1.7 SelectorReadout
reporting.selector_screen.SelectorReadout(
source_json,
attempts,
rows,
summaries,
prefix_support,
pairing,
)Captured source-bound screen; all nested evidence is immutable.
1.1.7.1 Attributes
| Name | Description |
|---|---|
| source_json | Captured plan, canonical report and reducer identities as strict JSON. |
| attempts | Exactly all 64 planned units, in frozen order. |
| rows | Factual and immediate shadow observations with explicit origin. |
| summaries | Per-origin equal-scene reductions with metric availability counts. |
| prefix_support | H8/H12/H16 support for every arm, including wholly unreached arms. |
| pairing | Root-shell parity for every intended source block. |
1.1.7.2 Methods
| Name | Description |
|---|---|
| to_jsonable | Return a fresh strict JSON projection of captured evidence. |
1.1.7.2.1 to_jsonable
reporting.selector_screen.SelectorReadout.to_jsonable()Return a fresh strict JSON projection of captured evidence.
1.2 Functions
| Name | Description |
|---|---|
| assemble_selector_readout | Reduce captured observations within origin arms, preserving all-attempt support. |
| main | Export strict JSON and optional interactive HTML from one captured readout. |
| read_published_selector_readout | Verify sealed published evidence and regenerate its existing reductions. |
| read_selector_bundle | Cold-read one externally anchored E28 publication without original stores. |
| read_selector_experiment | Capture a consistent screen through canonical receipt and deep-shard gates. |
| selector_figures | Plot factual equal-scene index means with defined/at-risk support in hover. |
| selector_html | Render deterministic offline HTML from captured evidence only. |
| selector_trajectory_figures | Overlay canonical calibrated-camera BEV chains per matched source block. |
1.2.1 assemble_selector_readout
reporting.selector_screen.assemble_selector_readout(
source,
attempts,
observations,
)Reduce captured observations within origin arms, preserving all-attempt support.
Store-local row IDs are remapped only in the temporary reduction projection; the returned observations preserve their original identities. This lets the canonical inspection reducer reject duplicates without confusing separate stores in the same physical scene. Scene means always precede across-scene means. Missing attempted prefixes contribute support counts, never rewards.
1.2.2 main
reporting.selector_screen.main()Export strict JSON and optional interactive HTML from one captured readout.
1.2.3 read_published_selector_readout
reporting.selector_screen.read_published_selector_readout(
published,
*,
frozen_plan,
expected_sha256,
)Verify sealed published evidence and regenerate its existing reductions.
The externally supplied publication SHA is the trust anchor. The captured plan bytes, producer, measurement and every planned source/unit are checked without opening historical absolute paths. Archived configs/<basename> writer bytes are required and hash-verified for every arm; missing controls fail admission, including arms without observations. Display coordinates must match the canonical archived policy and every factual row. All blocks require matched single source rows; complete parsed writer controls require H16, explicit seeds and equality outside the two treatments. No kernel probabilities, raw labels, deep store seals or acquisition evidence are refreshed; those claims remain captured historical evidence. Canonical-report contrasts, lifecycle records, elapsed/resource observations, terminal audit values and geometry are hash-protected captured fields, not regenerated. The returned provenance enumerates that boundary separately from regenerated index, prefix, support and pairing reductions. Paired contrast values are copied; only the pairing gate is reapplied. This does not establish training reproduction or promotion.
1.2.4 read_selector_bundle
reporting.selector_screen.read_selector_bundle(
bundle,
*,
expected_manifest_sha256,
)Cold-read one externally anchored E28 publication without original stores.
Every manifest declaration must have a unique safe relative path, lowercase SHA256 and nonboolean nonnegative byte size. Symlink components are rejected lexically. Consumed artifacts must be regular files before opening. Publication, frozen plan and declared archived writer configs are byte-verified; ancillary manifest products are not refreshed by this reader. Historical captured identities stay unchanged, while the returned source separately identifies current readback code. This verifies captured bytes and their reductions, not current producer kernels, raw-store seals or independent measurements. Original absolute paths and Git are not opened.
1.2.5 read_selector_experiment
reporting.selector_screen.read_selector_experiment(plan_path)Capture a consistent screen through canonical receipt and deep-shard gates.
The producer revision comes from the frozen plan, independent of this reporting checkout. Exactly resolved frozen writer TOML is compared with the persisted staging configuration produced by the shard writer. Canonical receipts bind source rows and target-pool sampling; every completed store is deep-admitted again after selector and terminal projection to detect drift.
1.2.6 selector_figures
reporting.selector_screen.selector_figures(readout)Plot factual equal-scene index means with defined/at-risk support in hover.
1.2.7 selector_html
reporting.selector_screen.selector_html(readout)Render deterministic offline HTML from captured evidence only.
Semantic figure IDs fix Plotly’s otherwise random div UUID. Trace order is arm-sorted and index/block-sorted; inline JavaScript is included once. The same captured readout and Plotly version produce identical HTML bytes.
1.2.8 selector_trajectory_figures
reporting.selector_screen.selector_trajectory_figures(readout)Overlay canonical calibrated-camera BEV chains per matched source block.
The canonical projection supplies metric coordinates with a fixed root camera origin and root-to-target world-Z-up orientation. Shape alone makes no statement about mesh clearance or physical human dynamics.