@riddledc/riddle-proof-packs 0.4.0 → 0.4.2

package/dist/index.cjs

@@ -751,11 +751,14 @@ var explore_songs_and_mixes_default = {
         label: "install-neon-exploration-proof",
         timeout_ms: 1e4,
         store_return_to: "__neonProof.install",
-        script: "window.__neonProofApi=()=>window.__NEON_MIX_PROOF__||window.__RIDDLE_SEQUENCER_PROOF__; window.__neonProofWindows=async()=>{const api=window.__neonProofApi(); if(api?.renderProofWindows) return await api.renderProofWindows({}); if(api?.renderOfflineMetrics) return {ok:true,proofKind:'offline-audio-window-set',windows:[await api.renderOfflineMetrics({bars:1,seed:'neon-explore-fallback',monitorProfile:'smallSpeaker'})]}; return {ok:false,error:'missing renderProofWindows'};}; const api=window.__neonProofApi(); const state=api?.getState?.()||api?.getSummary?.()||{}; return {available:Boolean(api),selectedSong:state.selectedSong||null};",
+        script: "window.__neonProofApi=()=>window.__NEON_MIX_PROOF__||window.__RIDDLE_SEQUENCER_PROOF__; const api=window.__neonProofApi(); const state=api?.getState?.()||api?.getSummary?.()||{}; return {available:Boolean(api),hasExplorationSweep:typeof api?.runExplorationSweep==='function',selectedSong:state.selectedSong||null};",
         return_summary_fields: [
           {
             path: "available"
           },
+          {
+            path: "hasExplorationSweep"
+          },
           {
             path: "selectedSong"
           }
@@ -769,8 +772,17 @@ var explore_songs_and_mixes_default = {
       },
       {
         type: "window_call",
-        label: "render-proof-windows",
-        path: "__neonProofWindows",
+        label: "run-exploration-sweep",
+        path: "__NEON_MIX_PROOF__.runExplorationSweep",
+        args: [
+          {
+            maxSongs: 4,
+            maxPartsPerSong: 2,
+            maxWindowsPerPart: 2,
+            monitorProfile: "smallSpeaker",
+            loadAllSources: true
+          }
+        ],
         store_return_to: "__neonProof.exploration",
         capture_return: true,
         timeout_ms: 18e4,
@@ -778,14 +790,26 @@ var explore_songs_and_mixes_default = {
           {
             path: "ok"
           },
+          {
+            path: "status"
+          },
           {
             path: "proofKind"
           },
           {
-            path: "windows.length"
+            path: "entryCount"
+          },
+          {
+            path: "findingCount"
           }
         ]
       },
+      {
+        type: "assert_window_value",
+        path: "__neonProof.exploration.proofKind",
+        expected_value: "neon-exploration-sweep",
+        timeout_ms: 1e4
+      },
       {
         type: "assert_window_value",
         path: "__neonProof.exploration.ok",
@@ -829,10 +853,12 @@ var explore_songs_and_mixes_default = {
     pack_id: "neon_step_sequencer",
     pack_public_name: "Neon Step Sequencer Pack",
     evidence_role_pattern: "current_target",
-    purpose: "Exploration sweep over the app-provided proof windows for prioritizing the next weakest song/mix combination.",
+    purpose: "Bounded exploration sweep over Neon songs, parts, mix profiles, and proof windows for prioritizing the next weakest review target.",
     required_receipts: [
-      "proof-window set returns ok",
-      "window receipts are captured",
+      "exploration sweep contract is available",
+      "song/part entries are captured",
+      "all sampled entries pass objective guardrails",
+      "prioritized findings are captured",
       "browser health is clean",
       "screenshot is captured"
     ],
@@ -1665,7 +1691,7 @@ var ratchet_loop_mix_level_search_default = {
       },
       {
         type: "window_call",
-        label: "run-ratchet-loop-mix-level-search",
+        label: "test-claim-candidates-mix-level-search",
         path: "__NEON_MIX_PROOF__.runRatchetLoop",
         args: [
           {
@@ -1679,8 +1705,7 @@ var ratchet_loop_mix_level_search_default = {
             maxIterations: 6,
             monitorProfile: "smallSpeaker",
             restore: true,
-            applyBest: false,
-            minImprovement: 0.1
+            applyBest: false
           }
         ],
         store_return_to: "__neonMixProof.ratchetLoop",
@@ -1697,10 +1722,13 @@ var ratchet_loop_mix_level_search_default = {
             path: "strategy"
           },
           {
-            path: "improvement"
+            path: "best.claimVerdict.status"
+          },
+          {
+            path: "best.claimCandidate.action.track"
           },
           {
-            path: "best.candidate.track"
+            path: "supportedClaimCandidateCount"
           }
         ]
       },
@@ -1750,16 +1778,18 @@ var ratchet_loop_mix_level_search_default = {
     pack_id: "neon_step_sequencer",
     pack_public_name: "Neon Step Sequencer Pack",
     evidence_role_pattern: "interaction_snapshots",
-    purpose: "Run a bounded automated ratchet loop using the Neon mix-level-search strategy.",
+    purpose: "Run a bounded loop that tests Neon mix-level change-claim candidates using the mix-level-search strategy.",
     required_receipts: [
       "ratchet loop returns ok",
       "baseline proof-window summary is captured",
-      "candidate attempts are captured",
-      "best candidate or human-review status is captured"
+      "claim candidates are captured",
+      "each candidate records claim receipts",
+      "best supported claim candidate or human-review status is captured",
+      "state restoration receipt is captured"
     ],
     does_not_prove: [
       "subjective mix quality",
-      "that the best objective candidate should be kept",
+      "that a supported claim candidate should be kept",
       "all possible mix edits"
     ]
   }

package/dist/index.js

@@ -716,11 +716,14 @@ var explore_songs_and_mixes_default = {
         label: "install-neon-exploration-proof",
         timeout_ms: 1e4,
         store_return_to: "__neonProof.install",
-        script: "window.__neonProofApi=()=>window.__NEON_MIX_PROOF__||window.__RIDDLE_SEQUENCER_PROOF__; window.__neonProofWindows=async()=>{const api=window.__neonProofApi(); if(api?.renderProofWindows) return await api.renderProofWindows({}); if(api?.renderOfflineMetrics) return {ok:true,proofKind:'offline-audio-window-set',windows:[await api.renderOfflineMetrics({bars:1,seed:'neon-explore-fallback',monitorProfile:'smallSpeaker'})]}; return {ok:false,error:'missing renderProofWindows'};}; const api=window.__neonProofApi(); const state=api?.getState?.()||api?.getSummary?.()||{}; return {available:Boolean(api),selectedSong:state.selectedSong||null};",
+        script: "window.__neonProofApi=()=>window.__NEON_MIX_PROOF__||window.__RIDDLE_SEQUENCER_PROOF__; const api=window.__neonProofApi(); const state=api?.getState?.()||api?.getSummary?.()||{}; return {available:Boolean(api),hasExplorationSweep:typeof api?.runExplorationSweep==='function',selectedSong:state.selectedSong||null};",
         return_summary_fields: [
           {
             path: "available"
           },
+          {
+            path: "hasExplorationSweep"
+          },
           {
             path: "selectedSong"
           }
@@ -734,8 +737,17 @@ var explore_songs_and_mixes_default = {
       },
       {
         type: "window_call",
-        label: "render-proof-windows",
-        path: "__neonProofWindows",
+        label: "run-exploration-sweep",
+        path: "__NEON_MIX_PROOF__.runExplorationSweep",
+        args: [
+          {
+            maxSongs: 4,
+            maxPartsPerSong: 2,
+            maxWindowsPerPart: 2,
+            monitorProfile: "smallSpeaker",
+            loadAllSources: true
+          }
+        ],
         store_return_to: "__neonProof.exploration",
         capture_return: true,
         timeout_ms: 18e4,
@@ -743,14 +755,26 @@ var explore_songs_and_mixes_default = {
           {
             path: "ok"
           },
+          {
+            path: "status"
+          },
           {
             path: "proofKind"
           },
           {
-            path: "windows.length"
+            path: "entryCount"
+          },
+          {
+            path: "findingCount"
           }
         ]
       },
+      {
+        type: "assert_window_value",
+        path: "__neonProof.exploration.proofKind",
+        expected_value: "neon-exploration-sweep",
+        timeout_ms: 1e4
+      },
       {
         type: "assert_window_value",
         path: "__neonProof.exploration.ok",
@@ -794,10 +818,12 @@ var explore_songs_and_mixes_default = {
     pack_id: "neon_step_sequencer",
     pack_public_name: "Neon Step Sequencer Pack",
     evidence_role_pattern: "current_target",
-    purpose: "Exploration sweep over the app-provided proof windows for prioritizing the next weakest song/mix combination.",
+    purpose: "Bounded exploration sweep over Neon songs, parts, mix profiles, and proof windows for prioritizing the next weakest review target.",
     required_receipts: [
-      "proof-window set returns ok",
-      "window receipts are captured",
+      "exploration sweep contract is available",
+      "song/part entries are captured",
+      "all sampled entries pass objective guardrails",
+      "prioritized findings are captured",
       "browser health is clean",
       "screenshot is captured"
     ],
@@ -1630,7 +1656,7 @@ var ratchet_loop_mix_level_search_default = {
       },
       {
         type: "window_call",
-        label: "run-ratchet-loop-mix-level-search",
+        label: "test-claim-candidates-mix-level-search",
         path: "__NEON_MIX_PROOF__.runRatchetLoop",
         args: [
           {
@@ -1644,8 +1670,7 @@ var ratchet_loop_mix_level_search_default = {
             maxIterations: 6,
             monitorProfile: "smallSpeaker",
             restore: true,
-            applyBest: false,
-            minImprovement: 0.1
+            applyBest: false
           }
         ],
         store_return_to: "__neonMixProof.ratchetLoop",
@@ -1662,10 +1687,13 @@ var ratchet_loop_mix_level_search_default = {
             path: "strategy"
           },
           {
-            path: "improvement"
+            path: "best.claimVerdict.status"
+          },
+          {
+            path: "best.claimCandidate.action.track"
           },
           {
-            path: "best.candidate.track"
+            path: "supportedClaimCandidateCount"
           }
         ]
       },
@@ -1715,16 +1743,18 @@ var ratchet_loop_mix_level_search_default = {
     pack_id: "neon_step_sequencer",
     pack_public_name: "Neon Step Sequencer Pack",
     evidence_role_pattern: "interaction_snapshots",
-    purpose: "Run a bounded automated ratchet loop using the Neon mix-level-search strategy.",
+    purpose: "Run a bounded loop that tests Neon mix-level change-claim candidates using the mix-level-search strategy.",
     required_receipts: [
       "ratchet loop returns ok",
       "baseline proof-window summary is captured",
-      "candidate attempts are captured",
-      "best candidate or human-review status is captured"
+      "claim candidates are captured",
+      "each candidate records claim receipts",
+      "best supported claim candidate or human-review status is captured",
+      "state restoration receipt is captured"
     ],
     does_not_prove: [
       "subjective mix quality",
-      "that the best objective candidate should be kept",
+      "that a supported claim candidate should be kept",
       "all possible mix edits"
     ]
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@riddledc/riddle-proof-packs",
-  "version": "0.4.0",
+  "version": "0.4.2",
   "description": "Reusable proof pack profiles and metadata helpers for the Riddle proof framework.",
   "license": "MIT",
   "author": "RiddleDC",

package/packs/neon-step-sequencer/README.md

@@ -33,16 +33,17 @@ This pack is the first app-specific lab for the open Riddle Proof architecture.
 
 The loop is not mix-specific as a proof concept. The proof concept is a bounded ratchet loop: propose a candidate, apply it, collect evidence, classify the result, restore or keep state, and repeat until the budget is exhausted.
 
-This pack's first concrete strategy is `mix-level-search`, which tries small level edits against Neon proof windows and returns objective evidence for the best candidate. It still does not decide subjective mix taste; the output is a listening and review packet.
+This pack's first concrete strategy is `mix-level-search`, which turns small level edits into change-claim candidates. Each candidate says what action will be attempted, what receipts must support the claim, and what evidence should be reviewed afterward. The loop may include a ranking metric to order review, but the verdict comes from receipts and invariants, not from a universal mix-quality number. It still does not decide subjective mix taste; the output is a listening and review packet.
 
 ## Example evidence
 
-The `examples/` directory contains four local Playwright proof results captured against LilArcade Neon Step Sequencer on May 24, 2026:
+The `examples/` directory contains local Playwright proof results captured against LilArcade Neon Step Sequencer on May 24, 2026:
 
 - `run-001-fast-mix-health`: passing `current_target` audit with proof contract, source readiness, mix RMS `0.1234`, peak `0.8321`, and no clipping.
 - `run-002-mix-change`: passing `interaction_snapshots` proof where a bass-level edit moved bass RMS from `0.0507` to `0.1071` and mix RMS from `0.073` to `0.1264` without clipping.
 - `run-003-full-matrix`: passing `current_target` viewport matrix across desktop, phone, iPad Mini, and iPad with `0 px` horizontal overflow.
-- `run-004-ratchet-loop-mix-level-search`: passing `interaction_snapshots` proof where a bounded loop tested six mix-level candidates and returned `chord -0.10` as the best objective candidate without keeping the edit.
+- `run-004-ratchet-loop-mix-level-search`: passing `interaction_snapshots` proof where a bounded loop tested six mix-level change-claim candidates, found a supported `chord -0.10` candidate, recorded receipt-level verdicts, and restored app state without keeping the edit.
+- `run-005-explore-songs-and-mixes-final`: passing `current_target` exploration sweep across four songs and eight song/part entries, with `8` passing entries, `0` prioritized findings, and no clipping after the local app-contract and mix-headroom ratchet.
 
 ## Naming note
 

package/packs/neon-step-sequencer/case-study/findings.md

@@ -41,7 +41,7 @@ Profile render window, song fixture, or authoring guide.
 
 ### Subjective boundary
 
-Metrics pass, but deciding whether the mix is better requires listening.
+Receipts can support that a requested mix-level change happened and stayed inside guardrails, but deciding whether the mix is musically better requires listening.
 
 Likely change:
 
@@ -93,14 +93,25 @@ Pack summary guidance first; Riddle Proof core only if a general display primiti
 ### Run 004 proved the bounded loop shape without making it mix-specific
 
 - run: `run-004-ratchet-loop-mix-level-search`
-- claim: a bounded ratchet loop can try candidate edits, collect proof-window evidence, rank candidates, and restore state.
-- observed evidence: `mix-level-search` tested `6` candidates; baseline score was `28.83345`; best score was `27.0708`; best candidate was `chord -0.10` to level `0.28`; loop status was `candidate_found`; console fatal count was `0`.
-- classification: none; passing `interaction_snapshots` proof with subjective listening caveat.
+- claim: a bounded ratchet loop can try change-claim candidates, collect proof-window evidence, record receipt verdicts, select a supported candidate for review, and restore state.
+- observed evidence: `mix-level-search` tested `6` candidates; baseline candidate-ranking metric was `28.8336`; best candidate-ranking metric was `27.07095`; best supported claim candidate was `chord -0.10` to level `0.28`; loop status was `claim_candidate_supported`; receipt checks covered edit acceptance, contract level agreement, rendered target metric movement, required instrument activity, no clipping, and no low-level proof window; console fatal count was `0`.
+- classification: none; passing `interaction_snapshots` proof with subjective listening caveat. The ranking metric is a review-order hint, not a proof verdict.
 - smallest layer changed: generic app proof-contract loop plus one Neon strategy.
 - change made: added `runRatchetLoop` with a `mix-level-search` strategy and a proof-pack profile that calls it.
 - rerun: passed on May 24, 2026.
 - next sharper question: can the loop run multiple strategy classes without changing the proof primitive?
 
+### Run 005 restored the exploration flywheel
+
+- run: `run-005-explore-songs-and-mixes-final`
+- claim: a bounded current-target sweep can explore multiple Neon songs/parts, classify findings, and end with a useful confidence map.
+- observed evidence: final run sampled `4` songs and `8` song/part entries; all `8` entries passed; prioritized findings dropped to `0`; source preparation loaded drums `samples`, bass/chord/guitar `hybrid`, and vocal `voice_oohs`; final sampled peaks stayed below the clipping threshold; console fatal count was `0`.
+- classification: resolved chain of `proof_insufficient`, `app_contract_gap`, and `product_regression` findings during the local ratchet.
+- smallest layer changed: app proof contract, app snapshot normalization, app fixture/mix data, and profile JSON.
+- change made: made the exploration profile call `runExplorationSweep`; added the app-contract sweep method; normalized arbitrary song/part tempo and bar count; preserved bass/chord/guitar lane enable flags in song snapshots; lowered only hot built-in Yakety and Monkberry Sheet mix data enough to clear clipping receipts.
+- rerun: passed on May 24, 2026 with `8` entries, `8` passed, and `0` findings.
+- next sharper question: can this exploration workflow become the normal local pack/profile loop before any changeset or npm release?
+
 ### Local runner shutdown needs a small ergonomics follow-up
 
 - run: `run-002-mix-change`, `run-003-full-matrix`, `run-004-ratchet-loop-mix-level-search`
@@ -108,6 +119,6 @@ Pack summary guidance first; Riddle Proof core only if a general display primiti
 - observed evidence: complete passing artifacts were written, but the wrapper process lingered after artifact write and had to be stopped.
 - classification: `proof_insufficient` for operator ergonomics, not a Neon product regression.
 - smallest layer changed: none in this pack.
-- change made: documented the issue and used an outer timeout for the matrix run.
-- rerun: not yet rerun after a runner fix.
+- change made: fixed the local Playwright runner timeout cleanup in `@riddledc/riddle-proof-runner-playwright`.
+- rerun: later runner smoke tests and the published `0.4.3` package exited cleanly.
 - next sharper question: should the local runner force-close browser handles or expose a clearer artifact-written exit phase?

package/packs/neon-step-sequencer/case-study/ratchet-card.md

@@ -33,8 +33,9 @@ The project shows that a complex audio app can improve proof confidence mostly b
 - Run 001: a `current_target` audit connected the Neon route, proof contract, source readiness, and offline mix-health metrics.
 - Run 002: an `interaction_snapshots` proof showed a bass-level edit moving bass RMS from `0.0507` to `0.1071` and mix RMS from `0.073` to `0.1264` without clipping.
 - Run 003: a `current_target` matrix passed across desktop, phone, iPad Mini, and iPad with `0 px` horizontal overflow.
-- Run 004: a bounded loop tested six `mix-level-search` candidates, returned `chord -0.10` as the best objective candidate, and restored app state.
+- Run 004: a bounded loop tested six `mix-level-search` change-claim candidates, returned `chord -0.10` as a supported review candidate with receipt-level evidence, and restored app state.
+- Run 005: a `current_target` exploration sweep sampled four songs and eight song/part entries. The first real sweep exposed app-contract normalization gaps and hot built-in song presets; after the local ratchet fixes it passed with `8` entries, `0` findings, and no clipping.
 
 ## Honest boundary
 
-These runs prove objective claims about a running app target. They do not prove that the mix is tasteful, that every song section is healthy, or that a release candidate is better than production. The ratchet loop is a generic proof-loop shape; `mix-level-search` is only the first Neon strategy plugged into it.
+These runs prove objective claims about a running app target. They do not prove that the mix is tasteful, that every song section is healthy, or that a release candidate is better than production. Run 005 is still bounded to configured song/part limits. The ratchet loop is a generic proof-loop shape; `mix-level-search` is only the first Neon strategy plugged into it.

package/packs/neon-step-sequencer/case-study/ratchet-log.md

@@ -177,7 +177,7 @@ Can the pack explore song/mix combinations and produce a prioritized confidence
 
 Claim:
 
-Neon can run a bounded ratchet loop that proposes mix-level candidates, applies each candidate, captures proof-window evidence, ranks candidates by objective metrics, and restores app state.
+Neon can run a bounded ratchet loop that proposes mix-level change-claim candidates, applies each candidate action, captures proof-window evidence, records receipt-level verdicts, selects a supported candidate for review, and restores app state.
 
 Profile:
 
@@ -185,16 +185,16 @@ Profile:
 
 Evidence to capture:
 
-- baseline proof-window score
-- per-candidate proof-window score
-- best candidate and objective improvement
+- baseline candidate-ranking metric
+- per-candidate receipt verdicts
+- supported claim candidate and ranking metric delta
 - state restoration receipt
 - compact caveats for human handoff
 
 Possible outcomes:
 
-- `candidate_found`: at least one candidate improves objective metrics.
-- `needs_human_review`: evidence is valid but no objective candidate clears the threshold.
+- `claim_candidate_supported`: at least one candidate has the receipts needed to support its change claim.
+- `needs_human_review`: evidence is valid but no candidate has enough receipts to support the proposed claim automatically.
 - `proof_insufficient`: the app contract or proof window does not provide enough evidence.
 - `profile_calibration`: the chosen tracks, windows, or thresholds do not fit the target.
 
@@ -206,17 +206,18 @@ Observed evidence:
 
 - strategy `mix-level-search`
 - tested `6` candidates across `bass`, `chord`, `guitar`, and `rhythmSynth`
-- baseline score `28.83345`
-- best score `27.0708`
-- objective improvement `1.7627`
-- best candidate `chord -0.10` to level `0.28`
-- loop status `candidate_found`
+- baseline candidate-ranking metric `28.8336`
+- best candidate-ranking metric `27.07095`
+- ranking metric delta `1.7627`
+- best supported claim candidate `chord -0.10` to level `0.28`
+- claim receipts recorded for edit acceptance, contract level agreement, rendered target metric movement, required instrument activity, no clipping, and no low-level proof window
+- loop status `claim_candidate_supported`
 - app state restored after the run
 - console fatal count `0`
 
 Failure classification:
 
-None. This was a passing `interaction_snapshots` loop proof, with an explicit listening-review caveat.
+None. This was a passing `interaction_snapshots` loop proof, with an explicit listening-review caveat and a ranking metric that is only a review-order hint.
 
 Smallest layer changed:
 
@@ -226,6 +227,61 @@ Next sharper question:
 
 Can the pack explore song/mix combinations and produce a prioritized confidence map?
 
+## Run 005 - Bounded song/mix exploration sweep
+
+Claim:
+
+Neon can run a bounded current-target exploration sweep across multiple songs and parts, produce a confidence map, classify weak layers, and close the loop after small user-controlled changes.
+
+Profile:
+
+`profiles/explore-songs-and-mixes.json`
+
+Evidence to capture:
+
+- route and proof-contract availability
+- sample/source preparation receipt
+- bounded song/part entries
+- per-window active-instrument, peak, headroom, and clipping receipts
+- prioritized findings
+- screenshot, console health, and layout health
+
+Possible outcomes:
+
+- `app_contract_gap`: the app cannot render arbitrary song/part proof states yet.
+- `proof_insufficient`: the exploration profile runs but cannot support the claim.
+- `profile_calibration`: proof windows or required-active receipts target the wrong musical window.
+- `product_regression`: a sampled song/part clips or violates objective guardrails.
+
+Observed status:
+
+Passed on May 24, 2026 with `local-playwright` after three local ratchet iterations.
+
+Observed evidence:
+
+- final run sampled `4` songs and `8` song/part entries
+- final run passed `8` entries with `0` prioritized findings
+- source preparation loaded drums `samples`, bass/chord/guitar `hybrid`, and vocal `voice_oohs`
+- final sampled peaks stayed below the clipping threshold: Yakety `0.9589` and `0.9756`, Monkberry Sheet `0.9734` and `0.9550`, Monkberry Full OMR `0.8345` and `0.8327`, Monkberry Tab proof windows `0.8328` to `0.8423`
+- console fatal count `0`
+- horizontal overflow `0 px`
+
+Failure classification:
+
+Resolved during the ratchet:
+
+- `proof_insufficient`: the first corrected sweep hit an `OfflineAudioContext` zero-frame error because historical song/part proof states did not normalize tempo/bar count.
+- `app_contract_gap`: the next sweep showed saved/song snapshots preserved `rhythmSynthEnabled` but not bass/chord/guitar enable flags.
+- `product_regression`: after lane flags were fixed, the sweep found clipping in Yakety Dark and Monkberry Sheet presets.
+
+Smallest layer changed:
+
+App proof contract, app snapshot normalization, app fixture/mix data, and the local exploration profile. Riddle Proof core did not need a change.
+
+Next sharper question:
+
+Can the same exploration shape become a reusable pack/profile workflow where a user can choose a bounded target set, run locally during iteration, and publish only after the evidence is worth sharing?
+
 ## Project note
 
 The ratchet is not a pass. The ratchet is the next sharper question.

package/packs/neon-step-sequencer/case-study/reusable-lessons.md

@@ -28,4 +28,4 @@ Most ratchet steps should change profile JSON, pack docs, app proof contracts, o
 
 ## Loops are generic
 
-A ratchet loop should be domain-neutral: propose a candidate, apply it, collect evidence, classify the result, restore or keep state, and repeat within a budget. Neon `mix-level-search` is a strategy plugged into that loop, not the loop's identity.
+A ratchet loop should be domain-neutral: propose a claim candidate, apply its action, collect evidence, classify receipt-level support, restore or keep state, and repeat within a budget. Neon `mix-level-search` is a strategy plugged into that loop, not the loop's identity.

package/packs/neon-step-sequencer/examples/README.md

@@ -11,12 +11,14 @@ The raw `profile-result.json` files are real runner outputs. They intentionally
 | `run-001-fast-mix-health` | `current_target` | passed | The current Neon target exposes a proof contract, source receipts, and a bounded offline mix-health render. |
 | `run-002-mix-change` | `interaction_snapshots` | passed | A bass-level edit changes rendered bass and mix metrics without clipping. |
 | `run-003-full-matrix` | `current_target` | passed | The mix-health proof holds across desktop, phone, iPad Mini, and iPad viewports. |
-| `run-004-ratchet-loop-mix-level-search` | `interaction_snapshots` | passed | A bounded ratchet loop tests mix-level candidates, returns the best objective candidate, and restores app state. |
+| `run-004-ratchet-loop-mix-level-search` | `interaction_snapshots` | passed | A bounded ratchet loop tests mix-level change-claim candidates, records receipt verdicts, chooses a supported candidate for review, and restores app state. |
+| `run-005-explore-songs-and-mixes-final` | `current_target` | passed | A bounded exploration sweep samples four songs and eight song/part entries, producing a zero-finding confidence map after app-contract and mix-headroom fixes. |
 
 ## What these examples do not prove
 
 - They do not prove subjective mix taste.
-- They do not prove every song, section, or mix preset.
+- They do not prove every song, section, or mix preset. Run 005 is bounded to the configured song, part, and proof-window limits.
 - They do not prove production CDN asset availability; these were local dev-server runs.
 - They do not prove a reference/candidate release delta. The mix-change run uses pre-action/post-action snapshots inside one proof run, not a separate baseline deployment.
 - The ratchet-loop run does not prove that the loop primitive is mix-specific; `mix-level-search` is only this pack's first concrete strategy.
+- The ratchet-loop run does not prove that the supported candidate should be kept. Its ranking metric is a review-order hint, not a taste verdict.