Incorporate the dimwit market cell with measured chart vision, and stop the RustCrypto dependency split

.github/dependabot.yml

@@ -35,3 +35,16 @@ updates:
       time: "05:45"
       timezone: America/Chicago
     open-pull-requests-limit: 3
+    groups:
+      # The RustCrypto crates share a `digest` major version. Bumped one at a time
+      # they cannot compile: sha2 0.11 moves to digest 0.11 while hmac 0.12 still
+      # expects digest 0.10, so Sha256 stops satisfying hmac's CoreProxy bound and
+      # new_from_slice/verify_slice fall off CoreWrapper<HmacCore<Sha256>>. Keeping
+      # them in one pull request makes the major bump atomic and compilable.
+      rustcrypto:
+        patterns:
+          - "sha2"
+          - "sha1"
+          - "hmac"
+          - "digest"
+          - "hkdf"

DUMBMONEY_WORKSPACE.json

@@ -95,9 +95,9 @@
       "snapshot_executed": false,
       "runtime_binding": "dumbmoney.runtime-bindings.v2:dimwit",
       "source_repository": "https://github.com/ObtuseAI/dimwit",
-      "public_base_commit": "1a6dbf6ae043bacd203f2ff17fe03a3e4a07bb43",
-      "integration_commit": "1a6dbf6ae043bacd203f2ff17fe03a3e4a07bb43",
-      "source_tree": "77eddc59e2a70ce18647f8555b4dcb3f06722200"
+      "public_base_commit": "276527a38c30df1427d4cf8effee899a55198d24",
+      "integration_commit": "276527a38c30df1427d4cf8effee899a55198d24",
+      "source_tree": "d5bbc2fc2feff1183d4a16cd726c3b454acedebf"
     },
     {
       "name": "nimrod",

contracts/runtime-bindings.v2.json

@@ -154,7 +154,7 @@
       "cell": "dimwit",
       "upstream_snapshot": {
         "path": "projects/dimwit",
-        "source_tree": "77eddc59e2a70ce18647f8555b4dcb3f06722200",
+        "source_tree": "d5bbc2fc2feff1183d4a16cd726c3b454acedebf",
         "product_role": "proof_bearing_game_production_studio",
         "executed_by_runtime": false
       },
@@ -168,7 +168,8 @@
         "implementation_status": "PARTIAL",
         "limitations": [
           "deterministic_ohlcv_indicators_plus_held_out_ta_evidence",
-          "chart_pixels_and_visual_perception_not_implemented",
+          "chart_pixel_recovery_is_verified_only_for_charts_this_runtime_rendered",
+          "foreign_chart_screenshots_are_not_interpreted",
           "frozen_dimwit_is_a_game_production_studio_and_is_not_executed"
         ],
         "executes_upstream_snapshot": false

projects/dimwit/.github/workflows/tests.yml

@@ -52,8 +52,12 @@ jobs:
         #     test_bot_balance_telemetry, test_elite_toolchains,
         #     test_metahuman_output_attempt, test_packaged_build_validation,
         #     test_performance_baseline.
+        # The 10 dimwit/tests/test_market_*.py modules are the market evidence
+        # lane (dimwit/market/): pure Python plus numpy/pillow, no Unreal, no
+        # Blender, no Windows path assumptions and no network, so they belong in
+        # the clean-clone set rather than the local-only one.
         # This selection is the honest intersection a clean Linux runner can
-        # actually satisfy: 549 tests across 58 modules, including source-controlled
+        # actually satisfy: 788 tests across 68 modules, including source-controlled
         # engine fixtures, improvement outcomes, and cross-engine proof receipts.
         run: |
           python -m pytest -q -p no:cacheprovider --no-header \
@@ -109,6 +113,16 @@ jobs:
             dimwit/tests/test_ui_settings_persistence.py \
             dimwit/tests/test_universal_mobile_factory.py \
             dimwit/tests/test_unreal_game_builder_engine.py \
+            dimwit/tests/test_market_bars.py \
+            dimwit/tests/test_market_cell_contract.py \
+            dimwit/tests/test_market_chart.py \
+            dimwit/tests/test_market_chart_vision.py \
+            dimwit/tests/test_market_evidence.py \
+            dimwit/tests/test_market_indicators.py \
+            dimwit/tests/test_market_knowledge.py \
+            dimwit/tests/test_market_patterns.py \
+            dimwit/tests/test_market_scan.py \
+            dimwit/tests/test_market_sports.py \
             tests/test_animation_roster.py \
             tests/test_mode_contract.py \
             tests/test_progression.py \

projects/dimwit/AGENTS.md

@@ -25,10 +25,25 @@ Autonomy **STOPS** at the review ceiling. The operator (a human) owns everything
 6. **Validate before trust.** After ANY change you make, run `python scripts/pipeline/run_validation.py` (exit 0 only on suite
    PASS). Your change is a **CANDIDATE** until it passes. Stop at the review ceiling.
 
+## The market cell (`dimwit/market/`)
+Dimwit's market evidence lane applies the studio's law to prices and game state. Same doctrine, three extra
+non-negotiables:
+1. **Observations only.** Never emit `forecast_probability` or `expected_return_bps` as anything but `None`.
+   Probability and edge claims belong to a downstream evidence court, after held-out evidence.
+2. **No lookahead, structurally.** Indicators must be prefix-stable (`indicator(bars)[i]` equals the value on
+   `bars[:i+1]`); patterns must expose `detected_at_index`, and rules may read only that, never `index`.
+3. **Disclose the search and the baseline.** Any new rule raises `family_size`; any claim is scored on excess
+   over the unconditional same-side baseline and must beat its own placebo. Never add a rule without re-running
+   `python -m dimwit market audit --deep`.
+
+No network, no credentials, no broker calls in that package. A new module there must be added to
+`evidence.ATTESTED_MODULES` or the contract test fails.
+
 ## How to drive
 | Purpose | Command |
 |---|---|
 | The gate (full fail-closed suite; exit 0 = PASS) | `python scripts/pipeline/run_validation.py` |
+| Market cell anti-costume gate (exit 0 = clean) | `python -m dimwit market audit --deep` |
 | Fast gate (UE/eyes validators → BLOCKED, cannot PASS — honest) | `python scripts/pipeline/run_validation.py --no-ue` |
 | List every validator | `python scripts/pipeline/run_validation.py --list` |
 | The loop (sweep to PROMOTED_TO_REVIEW) | `python scripts/pipeline/run_director.py` |

projects/dimwit/CHANGELOG.md

@@ -7,6 +7,28 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- `dimwit.market`: the market evidence lane — the studio's evidence law applied to prices and game state
+  instead of pixels and packages, with the same fail-closed discipline.
+  - 46 prefix-stable indicators and 24 pattern detectors, each carrying its confirmation lag.
+  - Deterministic candlestick renderer (PNG + SVG, three themes) with invertible pixel geometry, plus chart
+    vision that reads the prices back out and reports recovery error in pixels.
+  - Walk-forward rule scanner over 35 rules with baseline-excess scoring, overlap deflation, Bonferroni and
+    Benjamini–Hochberg search disclosure, and a placebo control.
+  - Sports game-state analysis, margin/win-probability charting, and a cross-game rule scanner whose
+    observations are genuinely independent.
+  - A 111-term citable knowledge pack merged from the code registries plus 41 methodology, microstructure and
+    assurance concepts.
+  - Downstream-compatible observation export, an implementation attestation over the market modules' own bytes,
+    and a hash-chained evidence ledger with a length anchor so tail truncation is detectable.
+  - `python -m dimwit market <cmd>` operator surface; each command exits non-zero on a BLOCKED verdict.
+  - `dimwit market audit` — the anti-costume gate: runs the cell and reports what actually executed.
+- New `MARKET` capability domain (24 read-only capabilities). The two writing verbs, `EXECUTE/chart.export` and
+  `EXECUTE/evidence.record`, stay behind the existing MCP mutation gate.
+- MCP path confinement for the chart-vision verbs that accept a local image path.
+- 239 tests in `dimwit/tests/test_market_*.py`.
+
 ## [0.1.1] - 2026-07-26
 
 ### Fixed

projects/dimwit/README.md

@@ -9,8 +9,9 @@ Every executable lane returns receipts. Every promotion stops for human review.
 
 [![Tests](https://github.com/ObtuseAI/dimwit/actions/workflows/tests.yml/badge.svg)](https://github.com/ObtuseAI/dimwit/actions/workflows/tests.yml)
 [![Python 3.12](https://img.shields.io/badge/Python-3.12-3776AB?logo=python&logoColor=white)](https://www.python.org/)
-[![CI proof](https://img.shields.io/badge/clean_clone-549_tests-8B5CF6)](.github/workflows/tests.yml)
+[![CI proof](https://img.shields.io/badge/clean_clone-788_tests-8B5CF6)](.github/workflows/tests.yml)
 [![Human gate](https://img.shields.io/badge/ceiling-PROMOTED__TO__REVIEW-22D3EE)](AGENTS.md)
+[![Market lane](https://img.shields.io/badge/market_lane-239_tests-F59E0B)](docs/DIMWIT_MARKET_LANE.md)
 [![Source available](https://img.shields.io/badge/license-source--available-22C55E)](LICENSE)
 
 </div>
@@ -23,6 +24,8 @@ It turns a brief into a typed, dependency-aware production graph. Each runnable
 
 Dimwit can autonomously prepare a candidate up to `PROMOTED_TO_REVIEW`. Human acceptance, source integration, signing, publishing, payment, account access, store submission, and active-slice promotion remain operator-owned.
 
+The evidence machinery is not specific to art. Dimwit applies the same law to numbers in its [market evidence lane](#12-market-evidence-lane): 46 indicators proven prefix-stable by recomputation, chart recovery error reported in pixels, and rule survival reported only after a disclosed search space and a placebo run. It emits observations and never a forecast.
+
 ## Product truth
 
 | Contract | Current public state |
@@ -32,6 +35,7 @@ Dimwit can autonomously prepare a candidate up to `PROMOTED_TO_REVIEW`. Human ac
 | **Engineering evidence baseline** | [`1019e4df`](https://github.com/ObtuseAI/dimwit/commit/1019e4dfb6191a2e751829253fde7359c5bb2d67) · configured [test workflow](https://github.com/ObtuseAI/dimwit/actions/workflows/tests.yml) |
 | **Proved now** | The public source implements typed production graphs, capability detection, bounded command plans, receipts, artifact hashing, fail-closed validators, recovery candidates, and review promotion. |
 | **Authority ceiling** | Automation stops at `PROMOTED_TO_REVIEW`. Human operators own acceptance, integration, publishing, accounts, payment, signing, and active-slice promotion. |
+| **Second production lane** | **Market evidence lane** — [`dimwit/market/`](dimwit/market): deterministic technical analysis, chart rendering, chart *vision*, walk-forward rule evidence, and sports game-state analysis. Observations only; no network, no credentials, no broker. |
 | **Clean demonstration** | Use the [quick start](#quick-start) to inspect the graph, run provider-free checks, and exercise bounded studio surfaces. |
 | **Known limit** | DCC execution, live captures, visual judgment, mobile SDKs, and packaged-build proof depend on the actual local toolchains and current project evidence; missing evidence remains blocked. |
 
@@ -51,6 +55,7 @@ Dimwit can autonomously prepare a candidate up to `PROMOTED_TO_REVIEW`. Human ac
 | **Does every engine execute everywhere?** | No. Adapters report detected, ready, plan-only, blocked, or reference-only based on actual project and toolchain evidence. |
 | **What is the core output?** | A review candidate with build receipts, logs, artifact hashes, validator results, provenance, rollback notes, and unresolved blockers. |
 | **What makes it different?** | It treats creative-production claims as evidence contracts rather than accepting successful process exit, generated files, or AI confidence as proof. |
+| **Does it only do art?** | No. The [market evidence lane](#12-market-evidence-lane) applies the same law to prices and game state instead of pixels and packages: technical analysis, chart vision, and walk-forward rule evidence. `python -m dimwit market audit --deep` runs it and reports what actually executed. |
 
 ## The production problem
 
@@ -373,6 +378,41 @@ The local Studio IDE brings production state into one work surface:
 
 The server binds to `127.0.0.1` and prints a tokenized URL. The token protects source views and operator actions. The interface does not expose an arbitrary shell or generic filesystem mutation endpoint.
 
+### 12. Market evidence lane
+
+The same evidence law, applied to numbers instead of pixels. [`dimwit/market/`](dimwit/market) is eleven modules
+of deterministic market analysis: technical indicators, chart rendering, chart *vision*, walk-forward rule
+evidence, and sports game-state analysis.
+
+| Question | Answer |
+| --- | --- |
+| **What does it do?** | Reads OHLCV bars and game timelines, renders charts, reads charts back, and produces walk-forward evidence about rules over them. |
+| **Why is it here?** | Because "the process exited zero" proves nothing about a number either. A backtest is the purest form of a claim that looks like proof and is not. |
+| **What is measured, not asserted?** | 46 indicators proven prefix-stable by recomputation on truncated prefixes; chart recovery error reported in **pixels** (≤ 0.5 px); rule survival reported after Bonferroni and Benjamini–Hochberg on a disclosed search space. |
+| **Does it forecast?** | No. Every result carries `forecast_probability: None`, and a test asserts it across all fifteen public result types. |
+| **Does it touch money?** | No. No network, no credentials, no broker, no orders. Bars arrive from the caller with a declared classification. |
+| **Authority ceiling** | Unchanged. The lane emits observations; nothing it produces promotes, sizes, or executes anything. |
+
+Three properties do the work:
+
+- **Prefix stability.** `indicator(bars)[i] == indicator(bars[:i+1])[-1]`, tested per indicator. That converts
+  "no lookahead" from an assurance into a property a test can fail. Patterns carry both `index` (where the shape
+  is) and `detected_at_index` (when it was knowable), because a fractal pivot is not visible until its
+  confirmation bars exist.
+- **Falsifiable chart vision.** The renderer returns its exact pixel geometry; the reader inverts it; the
+  round-trip is scored against known values. "Dimwit can see a chart" becomes a number a test can fail, not a
+  claim. Foreign screenshots are read for shape only, with the price scale pinned to `UNKNOWN`.
+- **Honest observation accounting.** Settled independent observations are the scarce resource, so rules are
+  scored on **excess over the unconditional same-side baseline**, t-statistics are deflated for window overlap,
+  and a placebo run with displaced entries must not match the real result. On a random walk the scanner finds
+  nothing; on an injected conditional edge it recovers the responsible rule.
+
+Provenance is recomputable rather than declared: every export embeds a digest of the actual bytes of the eleven
+attested market modules, and `python -m dimwit market audit` runs the lane end to end and reports what really
+executed — including what is deliberately absent.
+
+Full detail: [market evidence lane](docs/DIMWIT_MARKET_LANE.md).
+
 ## What proof looks like
 
 | Production claim | Required evidence |
@@ -412,6 +452,19 @@ python dimwit.py studio --status
 python scripts\pipeline\run_validation.py --list
 ```
 
+Exercise the market evidence lane against its deterministic synthetic fixture (no network, no data files):
+
+```powershell
+python -m dimwit market audit --deep
+python -m dimwit market indicators --self
+python -m dimwit market vision --self --bars 120
+python -m dimwit market scan --self --null
+python -m dimwit market chart --self --bars 140 --theme tote --out today
+python -m dimwit market know adverse_selection
+```
+
+Each command prints one JSON document and exits non-zero on a `BLOCKED` verdict, so any of them works as a shell gate.
+
 Launch the local Studio IDE:
 
 ```powershell
@@ -481,6 +534,7 @@ dimwit/
 |-- adapters/              shared engine and production contracts
 |-- capabilities/          typed registry and read/mutate boundaries
 |-- ledger/                evidence, outcomes, integrity, and repair
+|-- market/                market evidence lane: TA, chart render and vision, scans, sports
 |-- pipelines/             production, validation, runtime, and review flows
 |-- toolchains/            Blender, Unreal, engine, and mobile execution plans
 |-- studio/                DAG scheduling, receipts, and IDE data surfaces
@@ -503,6 +557,7 @@ Recommended reading:
 2. [Elite Studio toolchains](docs/DIMWIT_ELITE_STUDIO_TOOLCHAINS_20260711.md) — Blender, Unreal, and the studio DAG.
 3. [Universal game and mobile factory](docs/DIMWIT_UNIVERSAL_GAME_FACTORY_20260711.md) — engine adapters and mobile evidence.
 4. [Open-source adaptation audit](docs/DIMWIT_BLUNDER_OPEN_SOURCE_AUDIT_20260711.md) — reused ideas, upstream policy, and boundaries.
+4b. [Market evidence lane](docs/DIMWIT_MARKET_LANE.md) — technical analysis, chart vision, and observation accounting.
 5. [Autonomy capability matrix](docs/superpowers/specs/2026-06-28-autonomy-capability-matrix-design.md) — capability states and validation integration.
 6. [Cross-pipeline contract auditor](docs/superpowers/specs/2026-06-28-cross-pipeline-contract-auditor-design.md) — drift detection between production lanes.
 7. [Real-game validation loop](docs/superpowers/specs/2026-06-28-real-game-validation-loop-design.md) — runtime and capture proof.
@@ -534,7 +589,8 @@ Dimwit does not:
 - fabricate an iOS archive on Windows or a DCC result when the DCC is absent;
 - delegate signing credentials, purchases, account access, publishing, or payment;
 - promote its own work beyond `PROMOTED_TO_REVIEW`;
-- convert a successful command, generated file, or AI opinion into complete production proof.
+- convert a successful command, generated file, or AI opinion into complete production proof;
+- forecast prices or outcomes, hold broker credentials, place orders, or claim an edge — the market lane emits observations, and the bars it reads arrive from the caller with a declared classification.
 
 The intended outcome is not unattended publication. It is a better-prepared, better-instrumented, and more honestly reviewed candidate.