# Mad Moon: verified capabilities and presentation claims

**Evidence pass: September 11, 2026.** This report covers the 27 Mad Moon features and eight inherited Codex capabilities in the frozen catalogue. It catalogues what can be presented, the evidence supporting it and the limits that belong beside it. It does not choose stories or design the website.

Mad Moon has concrete engineering worth presenting now: a defined development method, durable project ideas, native memory and code-navigation integrations, exact workspace identities, guarded local operations, delivery checks and supported activity capture. Its strongest authorship claim is the environment and integration boundaries Mylo engineered around those capabilities. **This pass does not establish that the combined system makes development faster, cheaper, more accurate or more reliable overall.**

Six focused batches of existing tests produced **210 passes and one explicit native-check skip**. Current Serena and CGC reads returned a source-matching symbol, and Serena returned a confirmed reference. Hindsight returned project knowledge. Retained native records supply additional navigation, normal-restart and desktop-capture evidence, with their original identities and limitations. No product code, new tests, evaluator, dataset or permanent verification infrastructure was added.

## What to lead with

| Capability | Defensible presentation wording | Evidence limit |
| --- | --- | --- |
| Intent, review and human decisions | “I built a working method that keeps scope, plans, decisions and proof attached to inspectable artifacts.” | The method is guidance with contract checks. Host permissions and command guards have separate enforcement. Review quality has not been measured. |
| Durable context | “Project ideas have stable records and revisions, and the agent can retrieve native project memory.” | Parking Lot durability has focused fixture proof. Hindsight retrieval was observed, but this pass did not stage a new capture-to-fresh-recall cycle or measure relevance. |
| Understanding code | “I integrated semantic symbol/reference tools and a shared structural graph, then checked their returned code against the actual checkout.” | Current reads and dated native cases support specific results. They do not establish universal query/edit correctness or an improvement in task speed. |
| Project identity and preserved work | “Workspace commands bind work to explicit repositories and Git identities, and report work that survives a later failure.” | Real disposable Git/filesystem tests support selected behavior. They do not make every multi-step operation atomic. |
| Controlled changes and delivery | “I engineered preimage checks, recovery records and delivery checks over actual Git history.” | The controls cover owned operations and known prohibited material. Cleanup, migration and retirement have distinct scope. |
| Inspectable activity | “Supported desktop activity can be exported through a bounded launcher and inspected in Langfuse.” | Dated records document export/readback and known omissions. Original transcript paths in two receipts are now absent; current cloud retention and complete replay were not established. |

These statements explain capabilities and engineering choices. Benefit language such as less rework, better recall, fewer mistakes or lower cost remains an intended outcome until measured.

## Material findings and wording changes

**The catalogue's PR status is stale.** [PR 31](https://github.com/mylo-james/mad-moon/pull/31) is now merged at `9a4c0be56da1b236f659768393afeb56f64af0e1`. Its candidate head is still `19bb879f783049e83647eef860475152d42e5b3f`, with base `6187e99fbbf2b7832497040e331b92ff84e0f435`. The selected local development source and runtime receipt still point to that earlier base. Present F27 as merged with retained candidate proof, while stating that this selected local source has not incorporated it. This task did not update or restart the service.

**Serena works after selecting the existing project.** The first native configuration read found no active project. After the owner asked about its use, the existing development project was selected. Serena reported Python LSP ready and returned `canonical_digest` with the same body as current source and CGC, plus a confirmed cleanup reference. Both the initial result and successful follow-up remain in the record. No reinstall, onboarding or semantic edit was needed.

**“Memory works” needs a narrower meaning.** Hindsight configuration identity, loopback health, retrieval, capture, fresh-session recall and useful context are different assertions. This pass observed project knowledge retrieval. The sandboxed diagnostic could not establish loopback health, and no new capture/retrieval cycle or relevance study was performed.

**Capture evidence is useful but bounded.** Retained ordinary owner-started desktop records include automatic export, prompt/tool/response readback and trace reopening. A separate app-created record omits its initiating prompt. Those events used an older dirty source state, and their referenced original transcript files are now absent. The records can be described with those qualifications; they do not prove exact-current-source capture or a lossless replay system.

**Normal restart proof is not crash safety.** F27's retained small and representative native records match the exact candidate gateway hash and document selected-watch restoration, missed offline changes, normal shutdown and database reopening. They do not establish forced-termination or power-loss recovery. The recorded representative startup took 231.6 seconds; that observation belongs to its fixture and is not a general performance result.

**The current full CI gate uses macOS.** The workflow's Linux preference is guidance, not an implemented Linux-first matrix. The PR metadata records a successful applicable macOS job and GitGuardian check, alongside a separate earlier skipped run. No CI-cost reduction was measured or new hosted run launched.

## Identity, evidence and scope

The inspected development source is `6187e99fbbf2b7832497040e331b92ff84e0f435`, package 0.0.15, using the existing Python 3.14.7 / pytest 9.1.1 environment. The current checkout is dirty because of the local planning/progress packet; no tracked product source changed during this pass. All five input-report hashes matched the approved baseline.

M means a source mechanism, method or tested fixture behavior; C means a selected configuration; O means a bounded native event; B means a measured benefit. A pass in one class does not establish another. “Supported with narrower wording” means the adjacent scope and limitation are part of the claim. Deferred claims are accounted-for gaps, not successful checks. The contradiction recorded for F27 is the catalogue's historical open-PR status, not a failing feature.

The evidence index records the exact commands, testcase results, native reads, original receipt hashes, failures/limits and provenance. The detailed assertion record retains owners, Mad Moon's contribution, expected observations, falsifiers, proposed wording and refresh conditions. Every one of the 293 feature placement passages is preserved and linked to its assertion family; their engineering purposes remain proposed outcomes rather than separate proven events.

## Benefits and research

The local evidence supports what the environment does and selected observations of it doing those things. External research can explain why those choices are worth examining. It cannot supply a missing Mad Moon result.

- LongMemEval separates different long-term conversational-memory abilities. Its abstract supports treating memory as several evidence questions; it does not test this repository or Hindsight installation. [S07, original paper](https://arxiv.org/abs/2410.10813v2).
- METR's 2025 randomized study found slower completion with the tested AI tools for experienced maintainers in its particular setting. This cautions against assuming a universal gain; it is not an estimate of Mad Moon or current tools. The official design/limitations summary was reviewed, not a new local experiment. [S09, original study explanation](https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/).
- SWE-agent studies interfaces for repository navigation, editing and execution. Its abstract supports treating interface choices as an empirical question; its tested system and benchmarks differ from this environment. [S11, original paper](https://arxiv.org/abs/2405.15793v3).

## What remains deferred

No substantial verification code was needed. The remaining gaps are new Hindsight capture/fresh recall and relevance; universal host/reviewer isolation and continuity; Serena edit correctness beyond the observed reads; enabled native YouTube intake; fresh exact-source/cloud capture; forced-crash graph recovery; and comparative outcomes. Existing source or disposable fixtures cover selected local-operation behavior without requiring live cleanup, migration, stable installation or consumer delivery here.

The next useful evidence should usually come from an already authorized real operation or a maintained native check. When that evidence is not readily available, keep the limitation visible. No additional implementation project is needed to complete this report.

## Complete parent catalogue

| ID and capability | What this pass supports and limits | Detailed rows |
| --- | --- | --- |
| F01. Intent specification and planned execution | Defines procedure, exact artifacts and evidence expectations. 1 supported; 1 deferred. Benefit remains intended. | F01 (private evidence record) |
| F02. Scoped human approvals | Binds written human decisions to named artifacts and effects. 1 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F02 (private evidence record) |
| F03. Independent review with declared limits | Defines reviewer role and honest fallback conditions. 2 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F03 (private evidence record) |
| F04. Bounded delegation and model routing | Defines bounded assignments and retained primary responsibility. 1 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F04 (private evidence record) |
| F05. Visible progress and continuity for long work | Requires goals, visible phases, truthful completion and preserved context. 1 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F05 (private evidence record) |
| F06. A durable Parking Lot for future ideas | Implements per-project capture, revisions and durable publication. 4 supported; 1 deferred. Benefit remains intended. | F06 (private evidence record) |
| F07. Native project memory through Hindsight | Prepares and validates selected native mappings without replacing memory algorithms. 1 supported; 2 supported with narrower wording; 1 unverified; 2 deferred. Benefit remains intended. | F07 (private evidence record) |
| F08. Semantic code navigation and edits through Serena | Prepares pinned stdio integration and external provider state. 1 supported; 2 supported with narrower wording; 2 deferred. Benefit remains intended. | F08 (private evidence record) |
| F09. Structural repository navigation through CGC | Provides shared transport, selected roots and guarded configuration. 5 supported with narrower wording; 1 supported; 1 deferred. Benefit remains intended. | F09 (private evidence record) |
| F10. Explicit optional tool selection | Preserves selection and separates preparation from activation/readiness. 2 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F10 (private evidence record) |
| F11. Source-aware research intake | Pins/configures the provider and requires attributable bounded intake. 2 supported; 2 deferred. Benefit remains intended. | F11 (private evidence record) |
| F12. Native integration and portability boundaries | Validates explicit commands, environments and selected artifact identities. 2 supported; 2 deferred. Benefit remains intended. | F12 (private evidence record) |
| F13. Independent projects in one workspace | Validates independent project and worktree identities. 2 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F13 (private evidence record) |
| F14. Persistent worktree roles and partial outcomes | Creates/reuses exact roles and reports durable partial effects. 1 supported with narrower wording; 1 supported; 1 deferred. Benefit remains intended. | F14 (private evidence record) |
| F15. External operational records for consumer applications | Keeps consumer operational records external and names manual handoffs. 1 supported; 1 deferred. Benefit remains intended. | F15 (private evidence record) |
| F16. Checks across outgoing consumer Git history | Checks actual index/outgoing commits and preserves prior hook routes. 2 supported; 1 deferred. Benefit remains intended. | F16 (private evidence record) |
| F17. Previewed changes with ownership and recovery | Discloses effects and validates preimages with recovery records. 2 supported; 1 deferred. Benefit remains intended. | F17 (private evidence record) |
| F18. Distinct development and stable runtime lanes | Separates source development from verified stable release activation. 2 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F18 (private evidence record) |
| F19. Local feature integration with retained provenance | Archives feature ancestry and integrates eligible product content. 2 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F19 (private evidence record) |
| F20. Evidence-bound feature retirement | Retires only a verified feature worktree while retaining branch/archive. 2 supported; 1 deferred. Benefit remains intended. | F20 (private evidence record) |
| F21. Narrow cleanup and explicit migration | Separately implements receipt-bound cleanup and explicit mapped relocation. 2 supported; 2 supported with narrower wording; 2 deferred. Benefit remains intended. | F21 (private evidence record) |
| F22. Diagnostics that distinguish kinds of readiness | Reports identity/configuration/health limitations without cold repair. 1 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F22 (private evidence record) |
| F23. Requirement to evidence traceability | Requires credible requirement-to-proof links at stable boundaries. 1 supported; 1 deferred. Benefit remains intended. | F23 (private evidence record) |
| F24. Candidate validation and CI cost discipline | Classifies diffs and applies candidate checks/concurrency rules. 1 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F24 (private evidence record) |
| F25. Recorded agent activity through Langfuse | Connects eligible task events to maintained exporter. 2 supported; 2 supported with narrower wording; 1 deferred. Benefit remains intended. | F25 (private evidence record) |
| F26. Scoped trace capture and credential handling | Validates capture eligibility and bounds child execution/status records. 3 supported; 1 supported with narrower wording; 1 deferred. Benefit remains intended. | F26 (private evidence record) |
| F27. Selected graph watches restored after restart | Restores explicitly selected watches and prepares native database shutdown. 3 supported with narrower wording; 1 contradicted; 1 supported; 2 deferred. Benefit remains intended. | F27 (private evidence record) |
| C01. Layered instructions and on-demand skills | Supplies instruction layers and discoverable reusable skills. 1 supported with narrower wording; 1 deferred. Benefit remains intended. | C01 (private evidence record) |
| C02. User steering, questions and decisions | Connects user corrections to current scope and decisions. 1 supported with narrower wording; 1 deferred. Benefit remains intended. | C02 (private evidence record) |
| C03. Native tool execution and MCP access | Integrates available tool interfaces under declared ownership. 1 supported with narrower wording; 1 deferred. Benefit remains intended. | C03 (private evidence record) |
| C04. Native worker lifecycle and routing | Uses bounded workers with primary integration responsibility. 1 supported with narrower wording; 1 deferred. Benefit remains intended. | C04 (private evidence record) |
| C05. Working context, Git context and inspection surfaces | Names exact project/checkout and exposes inspectable artifacts. 1 supported; 1 deferred. Benefit remains intended. | C05 (private evidence record) |
| C06. Host permissions and execution boundaries | Combines declared human authority with operation-specific checks. 1 supported with narrower wording; 1 deferred. Benefit remains intended. | C06 (private evidence record) |
| C07. Native goals, task naming and available progress controls | Uses goals, naming and waits without inventing missing controls. 1 supported with narrower wording; 1 deferred. Benefit remains intended. | C07 (private evidence record) |
| C08. Task history, context continuity and native hook events | Uses durable artifacts and distinct native retention paths. 1 supported with narrower wording; 1 supported; 1 deferred. Benefit remains intended. | C08 (private evidence record) |
## Saved evidence

- Detailed assertions and placement crosswalk (private evidence record)
- Evidence index and test/native records (private evidence record)
- Baseline and exact identities (private evidence record)
- Native route assessment (private evidence record)
- Research and justified deferrals (private evidence record)
- Independent report review (private evidence record)
- Final identities and handoff (private evidence record)

## Assertion matrix

The table below includes the decomposed assertions, their dispositions and evidence IDs. Full fields and preserved stage purposes are in the linked detailed record. These are evidence statuses, not feature grades.

| Assertion | Statement | Disposition | Evidence |
| --- | --- | --- | --- |
| F01-M01 | The method orders specification, plan/tasks, implementation, review and candidate approval. | supported | B01 |
| F01-B01 | Measured local improvement: Clearer scope and less rework. | deferred | R01 |
| F02-M01 | Written approvals identify exact artifacts and named effects. | supported | B01 |
| F02-O01 | This packet retains the user's specification and combined-plan decisions against exact hashes. | supported with narrower wording | E004 E010 |
| F02-B01 | Measured local improvement: Fewer mistaken authority assumptions. | deferred | R01 |
| F03-C01 | The reviewer profile requests its model, effort and read-only settings. | supported | B01 E001 |
| F03-M01 | The method requires generic labeling when effective reviewer isolation is unproven. | supported | B01 |
| F03-O01 | Prior package review returned independent findings and a delta re-review. | supported with narrower wording | E010 |
| F03-B01 | Measured local improvement: More defects found before acceptance. | deferred | R01 |
| F04-M01 | Worker briefs must declare bounded scope, ownership, checks and stop conditions. | supported | B01 E001 |
| F04-O01 | This task dispatched two bounded read workers and integrated their returned artifacts. | supported with narrower wording | E010 |
| F04-B01 | Measured local improvement: Better allocation of effort, cost and elapsed time. | deferred | R01 |
| F05-M01 | Guidance requires available native progress controls and truthful completion state. | supported | E001 |
| F05-O01 | A native goal and task title were created for this verification phase. | supported with narrower wording | E010 |
| F05-B01 | Measured local improvement: Better continuity and visibility during long work. | deferred | R01 |
| F06-M01 | Capture keys make repeated idea capture idempotent. | supported | B02 |
| F06-M02 | Revision checks reject stale competing edits without overwriting the winner. | supported | B02 |
| F06-M03 | Native locks and atomic publication preserve a valid store after interruption. | supported | B02 |
| F06-M04 | Stable idea IDs, explicit states and read operations preserve retrievable records. | supported | B02 |
| F06-B01 | Measured local improvement: Less lost context and distraction from future ideas. | deferred | R01 |
| F07-M01 | Mapping updates preserve unrelated native settings and refuse stale/conflicting mappings. | supported | B03 |
| F07-C01 | Selected Hindsight artifact and configuration identities were reported valid. | supported with narrower wording | E004 |
| F07-O01 | The Hindsight knowledge interface returned project-specific pages in this task. | supported with narrower wording | E005 |
| F07-O02 | The selected loopback daemon is healthy for this diagnostic caller. | unverified | E004 |
| F07-O03 | A new development decision is captured and later retrieved in a fresh session. | deferred | E001 E005 |
| F07-B01 | Measured local improvement: Less repeated explanation and more useful project recall. | deferred | R01 |
| F08-M01 | Serena preparation renders selected stdio settings and external provider-state paths. | supported | B03 |
| F08-O01 | Retained native Serena activation, symbol and reference results match the recorded source. | supported with narrower wording | E006 |
| F08-O02 | After selecting existing development, Serena reports a ready LSP and returns the current canonical_digest body and references. | supported with narrower wording | E005 |
| F08-O03 | Native semantic edits preserve intended behavior in this selected session. | deferred | E001 E006 |
| F08-B01 | Measured local improvement: Faster and more accurate symbol-oriented work. | deferred | R01 |
| F09-M01 | The shared gateway serializes requests through a lifespan-bound native client. | supported with narrower wording | E001 E006 |
| F09-M02 | Loopback and Host/Origin checks reject the inspected hostile endpoint requests. | supported with narrower wording | E006 E008 |
| F09-M03 | Allowed roots constrain filesystem scope without automatically indexing every allowed root. | supported | B03 E006 |
| F09-O01 | CGC currently lists the development checkout as indexed and watched. | supported with narrower wording | E005 |
| F09-O02 | A current CGC lookup returned canonical_digest with a matching source body. | supported with narrower wording | E005 |
| F09-O03 | Retained CGC native results contain the expected caller-to-callee relationship. | supported with narrower wording | E006 |
| F09-B01 | Measured local improvement: Better discovery of code relationships. | deferred | R01 |
| F10-M01 | Optional-registration updates preserve activation choices and unrelated configuration. | supported | B02 B03 |
| F10-M02 | Absent optional registrations are classified without writing them. | supported | B02 B05 |
| F10-C01 | Saved global Serena/CGC registrations are enabled and YouTube is disabled. | supported with narrower wording | E004 |
| F10-B01 | Measured local improvement: Less irrelevant tool exposure and better tool selection. | deferred | R01 |
| F11-M01 | The pinned transcript integration validates its selected configuration and command boundary. | supported | B03 |
| F11-M02 | The intake method requires pagination, attribution and separation of captions from visual proof. | supported | E001 |
| F11-O01 | This session can perform verified native transcript discovery and complete caption intake. | deferred | B03 E004 |
| F11-B01 | Measured local improvement: More useful attributable research. | deferred | R01 |
| F12-M01 | Provider calls validate explicit executables, supported literal argv and selected environment keys. | supported | B03 |
| F12-M02 | Selected artifact/payload identity checks reject the inspected tampering cases. | supported | B03 |
| F12-O01 | The same integrated behavior works across all compatible agent hosts. | deferred | E001 E006 |
| F12-B01 | Measured local improvement: Reusable integrations with less custom maintenance. | deferred | R01 |
| F13-M01 | Registration validates independent project locators and strict manifest identities. | supported | B04 |
| F13-M02 | Worktree roles are checked against the expected Git common directory and branch identity. | supported with narrower wording | B04 |
| F13-O01 | This task's source and runtime receipt select the declared development checkout and base commit. | supported | E004 |
| F13-B01 | Measured local improvement: Fewer wrong-project operations. | deferred | R01 |
| F14-M01 | Existing commands create or reuse explicitly named worktree roles. | supported with narrower wording | B04 |
| F14-M02 | A later registration failure preserves completed Git work and reports partial effects. | supported | B04 |
| F14-B01 | Measured local improvement: Less lost work and clearer recovery after partial failure. | deferred | R01 |
| F15-M01 | Consumer CURRENT records are manually read external handoffs, not automatic native resolution. | supported | E002 |
| F15-B01 | Measured local improvement: Less mixing of private development records with product code. | deferred | R01 |
| F16-M01 | Consumer checks inspect the actual index and outgoing commits, including content later removed. | supported | B02 |
| F16-M02 | Hook wrappers preserve prior hooks and select the invoking linked worktree. | supported | B02 |
| F16-B01 | Measured local improvement: Fewer known operational-record leaks in delivery. | deferred | R01 |
| F17-M01 | Owned writes refuse stale preimages, unsafe identities and symlink conflicts. | supported | B04 |
| F17-M02 | Recovery records retain preimages and classify bounded partial/durability failures. | supported | B04 |
| F17-B01 | Measured local improvement: More inspectable changes and less accidental overwrite. | deferred | R01 |
| F18-M01 | Development activation rejects inspected branch/checkout drift while preserving the prior pointer. | supported | B05 |
| F18-M02 | Stable release transport/promotion rejects inspected invalid release/tag identities. | supported with narrower wording | B05 |
| F18-C01 | The current source lane selects the base commit; stable has no verified active lane. | supported | E004 |
| F18-B01 | Measured local improvement: Less confusion about which code is running. | deferred | R01 |
| F19-M01 | Local integration archives feature ancestry while delivering eligible product content. | supported | B05 |
| F19-M02 | Post-CAS checkout failure retains effects and reports manual recovery. | supported | B05 |
| F19-M03 | Provider handoff readiness is distinct from completed Git integration. | supported with narrower wording | E002 |
| F19-B01 | Measured local improvement: Cleaner delivery with recoverable provenance. | deferred | R01 |
| F20-M01 | Retirement requires bound closeout/integration evidence and preserves branch/archive after worktree removal. | supported | B05 |
| F20-M02 | A manifest conflict after native removal reports manual recovery. | supported | B05 |
| F20-B01 | Measured local improvement: Safer removal of completed worktrees. | deferred | R01 |
| F21-M01 | Cleanup preview is itemized, digest/receipt bound and writes nothing. | supported | B05 |
| F21-M02 | Cleanup apply requires the exact current approved eligible digest. | supported with narrower wording | E002 |
| F21-O01 | Destructive cleanup successfully removes eligible files in a newly verified fixture. | deferred | E002 |
| F21-M03 | Migration moves explicitly selected same-filesystem worktrees while retaining excluded work. | supported | B05 |
| F21-M04 | Migration recovery preserves/restores declared Git state after inspected late failures. | supported with narrower wording | B05 |
| F21-B01 | Measured local improvement: More predictable maintenance and relocation. | deferred | R01 |
| F22-M01 | Diagnostics distinguish unavailable, inactive, conflict and unverified states without cold repair. | supported | B05 |
| F22-O01 | Current status distinguishes valid identities from unproven session, protocol and freshness state. | supported with narrower wording | E004 |
| F22-B01 | Measured local improvement: Faster and more accurate diagnosis. | deferred | R01 |
| F23-M01 | The method requires normative requirements to map to credible stable-boundary evidence. | supported | B01 E001 |
| F23-B01 | Measured local improvement: More defensible completion decisions. | deferred | R01 |
| F24-M01 | CI source classifies complete diffs and uses native concurrency with a macOS full gate. | supported | B06 E002 |
| F24-O01 | PR31 records a successful applicable macOS gate and GitGuardian check for its candidate. | supported with narrower wording | E009 |
| F24-B01 | Measured local improvement: Less wasted CI while preserving useful validation. | deferred | R01 |
| F25-M01 | The native Stop launcher connects eligible tasks to the pinned maintained exporter. | supported | B06 E003 |
| F25-O01 | Retained owner-started desktop verification records report a prompt/tool/response trace and cloud readback. | supported with narrower wording | E007 |
| F25-O02 | Retained app-created task readback omits the initiating delegated prompt while retaining response/tool data. | supported with narrower wording | E007 |
| F25-M02 | The capture integration covers supported exporter activity while native provider histories remain separate. | supported | E003 E007 |
| F25-B01 | Measured local improvement: Better inspection of agent activity. | deferred | R01 |
| F26-M01 | Capture validation excludes old, foreign and ineligible history by task/checkout/cutoff identity. | supported | B06 |
| F26-M02 | A bounded selected child environment passes fixture credentials only to the exporter route. | supported | B06 |
| F26-M03 | Timeout handling stops owned descendants and status receipts preserve foreign file boundaries. | supported | B06 |
| F26-C01 | Current capture metadata selects the development checkout and matching pinned exporter payload. | supported with narrower wording | E003 |
| F26-B01 | Measured local improvement: More controlled capture and fewer unrelated exports. | deferred | R01 |
| F27-M01 | Candidate watch_roots restoration is distinct from allowed roots and waits for native reconciliation. | supported with narrower wording | E008 |
| F27-O01 | Retained candidate fixtures restored selected watches and offline changes across normal restart. | supported with narrower wording | E008 |
| F27-O02 | The representative candidate database reopened in a distinct native process after shutdown. | supported with narrower wording | E008 |
| F27-O03 | PR31 is still an open pending candidate. | contradicted | E009 |
| F27-C01 | This selected development source still uses the earlier base gateway after the remote merge. | supported | E004 E008 E009 |
| F27-O04 | Forced termination or power loss preserves complete graph recovery. | deferred | E008 |
| F27-B01 | Measured local improvement: Less missing structural context after normal restart. | deferred | R01 |
| C01-O01 | This task received repository instructions and explicitly read the relevant skills. | supported with narrower wording | E010 |
| C01-B01 | Measured local improvement: More consistent task behavior. | deferred | R01 |
| C02-O01 | The user's effort correction and exact approvals are reflected in the saved packet. | supported with narrower wording | E010 |
| C02-B01 | Measured local improvement: More useful human control during development. | deferred | R01 |
| C03-O01 | Native shell, patch and selected MCP calls returned inspectable results in this task. | supported with narrower wording | E005 E010 |
| C03-B01 | Measured local improvement: More effective source and tool use. | deferred | R01 |
| C04-O01 | Native workers returned bounded findings for primary integration. | supported with narrower wording | E010 |
| C04-B01 | Measured local improvement: Useful parallelism and independent assessment. | deferred | R01 |
| C05-O01 | Cwd, Git and import observations identify the source checkout used by this task. | supported | E004 |
| C05-B01 | Measured local improvement: Less context confusion between projects. | deferred | R01 |
| C06-C01 | Host tool contracts declare restricted filesystem/network behavior and scoped escalation. | supported with narrower wording | E010 |
| C06-B01 | Measured local improvement: Fewer unintended effects. | deferred | R01 |
| C07-O01 | Native goal/title controls were used; update_plan was unavailable and disclosed. | supported with narrower wording | E010 |
| C07-B01 | Measured local improvement: Better visibility of long-running work. | deferred | R01 |
| C08-O01 | Retained history/hook records and durable artifacts support distinct continuity paths. | supported with narrower wording | E006 E007 E010 |
| C08-M01 | The method keeps context, Hindsight retention and Langfuse capture as separate evidence paths. | supported | E001 E003 E007 |
| C08-B01 | Measured local improvement: Better continuity and inspectability across sessions. | deferred | R01 |