sourcecode 5.9.3__tar.gz → 5.9.6__tar.gz

{sourcecode-5.9.3 → sourcecode-5.9.6}/CHANGELOG.md

@@ -1,5 +1,44 @@
 # Changelog
 
+## [5.9.6] — 2026-08-27
+
+*Roadmap correction and release-alignment patch release.*
+
+**Fixed**
+
+- Makes explicit cache warming durable across processes.
+- Reconciles Spring presence detection for foreign member annotations.
+- Removes split Markdown badge fragments from README-derived summaries.
+- Makes empty baseline history an explicit `no_history` result.
+- Explains the intentional CLI/MCP population difference in `self-audit`.
+
+**Validation**
+
+- Added dedicated regression coverage using `/Users/user/Documents/workspace/testing/teavm`.
+- Updated README, English and Spanish guides, roadmap, ledger and CLI version output.
+
+## [5.9.4] — 2026-08-27
+
+*External-audit reliability and contract-alignment release.*
+
+**Fixed**
+
+- Rejects missing consumer repositories, malformed CVE feeds and unresolved
+  `pack gate --since` refs as structured input errors.
+- Rejects unknown profiles in `posture --diff` when the repository declares a
+  profile population, while preserving annotation-only profile fixtures.
+- Prevents architectural baseline captures from overwriting the snapshot for
+  the same commit.
+- Makes `endpoints --compact` disclose inline that its empty row list is an
+  intentional omission, and makes unresolved `impact`/`impact-chain` runs exit
+  non-zero even when no Java files are present.
+
+**Documentation and validation**
+
+- Updated README, English and Spanish user guides, roadmap status and CLI help.
+- Dedicated regression tests pass against repositories under
+  `Documents/workspace/testing`.
+
 ## [5.9.3] — 2026-08-27
 
 *Workspace entitlement correctness patch release.*

{sourcecode-5.9.3 → sourcecode-5.9.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sourcecode
-Version: 5.9.3
+Version: 5.9.6
 Summary: Persistent structural context and ultra-fast repeated analysis for AI coding agents
 Project-URL: Homepage, https://github.com/HarounDominique/sourcecode
 Project-URL: Repository, https://github.com/HarounDominique/sourcecode
@@ -59,7 +59,7 @@ Description-Content-Type: text/markdown
 > question. All analysis is local and deterministic; see [Privacy](#configuration--privacy)
 > before inspecting a repository whose telemetry policy you need to control.
 
-![Version](https://img.shields.io/badge/version-5.9.3-blue)
+![Version](https://img.shields.io/badge/version-5.9.6-blue)
 ![Python](https://img.shields.io/badge/python-3.9%2B-green)
 
 Copyright (c) 2026 Dominique Haroun. All rights reserved. ASK Engine 5.7.0 and
@@ -196,7 +196,7 @@ set `ASK_MAX_RSS_MB=<megabytes>`; if the process exceeds it, ASK refuses the
 answer with `MEMORY_TOO_LARGE`, including the observed peak and limit. Leave it
 unset, or set it to `0`, to disable the ceiling explicitly.
 
-**Current performance status.** Release 5.9.3 retains the shared Canonical IR
+**Current performance status.** Release 5.9.6 retains the shared Canonical IR
 path used by `validation` and now passes that same CIR into labelled `data-exposure`
 runs; the no-label path remains parse-free. These are implementation properties,
 not duration promises. A controlled external warm-path audit has opened performance
@@ -218,6 +218,14 @@ in `overlapping_paths`; a missing or non-directory path is rejected as structure
 invalid input rather than treated as an empty repository. This prevents silent
 entitlement errors in CI and does not change per-repository gates or billing.
 
+Release 5.9.6 also hardens the boundary contracts: missing consumer repositories,
+malformed CVE feeds and unresolved `pack gate --since` refs fail as structured
+input errors; duplicate same-commit baselines are never overwritten. `posture
+--diff` validates unknown profiles when the repository declares profiles. In
+`endpoints --compact`, an empty `endpoints` list is explicitly marked as an
+intentional omission, not an empty API; unresolved `impact` and `impact-chain`
+queries exit non-zero, including on repositories without Java sources.
+
 **What a warm actually covers.** `ask cache warm` rebuilds the declared root views and their
 shared structural layers (L1/L2 + the Repository Intelligence Snapshot + the shared Canonical
 IR). Pass `--views compact` or `--views compact,agent` to warm only selected views; without
@@ -243,7 +251,7 @@ brew tap haroundominique/sourcecode && brew install sourcecode
 # pip / pipx
 pipx install sourcecode        # or: pip install sourcecode
 
-ask version                    # ask 5.9.3 — and, on a build that has aged,
+ask version                    # ask 5.9.6 — and, on a build that has aged,
                                # how many releases have probably shipped since
 ```
 

{sourcecode-5.9.3 → sourcecode-5.9.6}/README.md

@@ -10,7 +10,7 @@
 > question. All analysis is local and deterministic; see [Privacy](#configuration--privacy)
 > before inspecting a repository whose telemetry policy you need to control.
 
-![Version](https://img.shields.io/badge/version-5.9.3-blue)
+![Version](https://img.shields.io/badge/version-5.9.6-blue)
 ![Python](https://img.shields.io/badge/python-3.9%2B-green)
 
 Copyright (c) 2026 Dominique Haroun. All rights reserved. ASK Engine 5.7.0 and
@@ -147,7 +147,7 @@ set `ASK_MAX_RSS_MB=<megabytes>`; if the process exceeds it, ASK refuses the
 answer with `MEMORY_TOO_LARGE`, including the observed peak and limit. Leave it
 unset, or set it to `0`, to disable the ceiling explicitly.
 
-**Current performance status.** Release 5.9.3 retains the shared Canonical IR
+**Current performance status.** Release 5.9.6 retains the shared Canonical IR
 path used by `validation` and now passes that same CIR into labelled `data-exposure`
 runs; the no-label path remains parse-free. These are implementation properties,
 not duration promises. A controlled external warm-path audit has opened performance
@@ -169,6 +169,14 @@ in `overlapping_paths`; a missing or non-directory path is rejected as structure
 invalid input rather than treated as an empty repository. This prevents silent
 entitlement errors in CI and does not change per-repository gates or billing.
 
+Release 5.9.6 also hardens the boundary contracts: missing consumer repositories,
+malformed CVE feeds and unresolved `pack gate --since` refs fail as structured
+input errors; duplicate same-commit baselines are never overwritten. `posture
+--diff` validates unknown profiles when the repository declares profiles. In
+`endpoints --compact`, an empty `endpoints` list is explicitly marked as an
+intentional omission, not an empty API; unresolved `impact` and `impact-chain`
+queries exit non-zero, including on repositories without Java sources.
+
 **What a warm actually covers.** `ask cache warm` rebuilds the declared root views and their
 shared structural layers (L1/L2 + the Repository Intelligence Snapshot + the shared Canonical
 IR). Pass `--views compact` or `--views compact,agent` to warm only selected views; without
@@ -194,7 +202,7 @@ brew tap haroundominique/sourcecode && brew install sourcecode
 # pip / pipx
 pipx install sourcecode        # or: pip install sourcecode
 
-ask version                    # ask 5.9.3 — and, on a build that has aged,
+ask version                    # ask 5.9.6 — and, on a build that has aged,
                                # how many releases have probably shipped since
 ```
 

sourcecode-5.9.6/docs/AUDIT-2026-08-27-EXTERNAL-5.9.3.md

@@ -0,0 +1,96 @@
+# External Audit Intake — ASK 5.9.3
+
+**Received:** 2026-08-27  
+**Scope:** reliability, semantic consistency, agent ergonomics and commercial fit  
+**Evidence base:** two external reports covering five parallel batches, eight-plus
+real Java repositories and approximately 180 CLI invocations
+
+## Executive disposition
+
+The reports use different rubrics and must not be collapsed into one KPI:
+
+- **7.2/10** for the first reliability and commercial review. It reports five
+  silent failures, one factual help error and strong agent-oriented bounding.
+- **6.5/10** for the broader semantic and cross-repository review. It reports a
+  critical nondeterminism in `selftest`, a high-confidence Spring detection false
+  positive and several lower-severity contract inconsistencies.
+
+The audit confirms that no analysed repository was modified: five batches remained
+clean, only ASK cache artefacts were created, and no Java files or commits were
+written. It also confirms strong structured errors when an error path is reached,
+default non-destructiveness, deterministic bounding in most commands and accurate
+Spring analysis across Boot 4, Boot 2 and classic Spring repositories.
+
+## Accepted defects and regressions
+
+The following are accepted into the engineering queue because the reports provide
+reproduction details or a directly observable contract violation:
+
+| ID | Severity | Finding | Triage |
+|---|---:|---|---|
+| `AUD-610-B01` | P0 | `ask selftest .` is nondeterministic: intermittent empty exit-1, hang or success across four repositories; individual rows are stable. | Open. Treat as an aggregation/concurrency or uncaught-worker failure until a controlled trace proves otherwise. A silent failure in the product's own reliability command blocks strong reliability claims. |
+| `AUD-610-B02` | P1 | Root `ask . --compact/-agent -f yaml` can exit 1 with both streams empty. | Open. Reproduce on a clean cache and capture the failing subprocess boundary before choosing a fix. |
+| `AUD-610-B03` | P1 | `ask audit-report .` fails silently in markdown, JSON and YAML, with and without output file. | Open. The documented flagship evidence path must return a payload or structured diagnostic, never zero bytes. |
+| `AUD-610-B04` | P1 | `ask risk . --limit 0` crashes despite `0` being a legal documented value. | Open. Add the boundary regression and preserve the existing rejection for negative limits. |
+| `AUD-610-B05` | P1 | `ask endpoints . --consumer <missing>` does not reject the missing consumer path and can produce an oversized unfiltered result. | Open. Validate consumer paths and prove the filter changes the analysed population. |
+| `AUD-610-B06` | P1 | `spring-audit` reports `spring_detected: true` for ANTLR4 after cache clear, while `migrate-check` says no Spring usage. | Open. Establish one shared stack-detection authority; do not infer Spring from unrelated XML or dependency names. |
+| `AUD-610-B07` | P2 | `pack gate --since <missing-ref>` collapses the structured base-ref error into generic `UNVERIFIED`. | Open. Preserve the cause and actionable ref error through pack aggregation. |
+| `AUD-610-B08` | P2 | Malformed CVE feeds yield `status: ok` with the parse failure buried in limitations. | Open. A broken security feed must be visibly non-successful and fail closed for automation. |
+| `AUD-610-B09` | P2 | Unknown profile names in `posture --diff` are accepted without a `PROFILE_NOT_FOUND`-style diagnostic. | Open. Validate profile identifiers against the repository's declared profile set. |
+| `AUD-610-B10` | P2 | Re-capturing the same baseline commit with another `--ref` silently overwrites the first record. | Open. Require explicit replacement or preserve both records; the series contract says history is not reconstructable. |
+
+## Contract and semantic debt
+
+These items are not all regressions, but they weaken cross-command coherence and are
+carried as follow-up acceptance criteria:
+
+- `pr-impact` root help incorrectly says `--files` consumes a unified diff from stdin;
+  it consumes changed-file paths. The help is factually wrong and must be corrected.
+- `impact-chain`, `impact` and related commands use different `not_found` risk values
+  (`0.0`, `low` and `null`). Define one shared unknown/not-found contract and its basis.
+- `endpoints --compact` publishes `endpoints: []` beside a nonzero `total`, while the
+  reason is buried in `_meta.answer_bounds`; place the bounded projection explanation
+  beside the affected field or expose a named summary field.
+- `migrate-check` can headline a Boot migration at monorepo root when only a minority
+  module is Boot-based; attribute the signal to its module and publish mixed-stack scope.
+- `custom_gate_inferred` lacks per-case provenance for inferred counts that exceed
+  direct annotation matches; document inheritance evidence or lower the claim.
+- `self-audit` and `mcp list-tools` report different populations (104 versus 110)
+  without explaining that CLI registration and MCP exposure are distinct counts.
+- `baseline diff`/`trend` fail without history while `baseline attest` returns
+  `no_history`; either document the intentional distinction or unify the family.
+- A replacement character was observed in `posture` non-coverage output. The separate
+  UTF-8 `chunk-file` check passed, so the encoding issue remains localized and needs a
+  raw reproduction before a parser change.
+- One `retrieve` invocation failed silently and then passed on immediate retry. It is
+  not promoted to a confirmed regression without a second controlled reproduction, but
+  logging/timeout evidence is warranted.
+
+## Confirmed strengths and non-regressions
+
+- `audit-report --sign-key` and `audit-verify` remain functional offline HMAC-SHA256
+  integrity checks; this is tamper-evident integrity, not public PKI or certification.
+- `baseline attest` is deterministic and explicit about non-coverage, but remains
+  unsigned; `remediation-report` describes ASK's own ledger, not the user's repository.
+- `self-audit` passed its four internal checks in the broader run; the reported count
+  discrepancy is a disclosure problem, not evidence that either registry is corrupt.
+- Spring semantic analysis, JAX-RS explanation, `verify` exit semantics, cache reuse,
+  UTF-8 chunk preservation and default dry-run/non-destructive behaviour held in the
+  tested repositories.
+- The reported cold-cache figures vary within the supplied range and are not, by
+  themselves, a performance regression.
+
+## Commercial interpretation
+
+The deterministic Spring context and agent-oriented bounding remain differentiated,
+and signed audit bundles strengthen the evidence workflow. However, the flagship
+silent failures and `selftest` nondeterminism directly undermine a reliability-led
+commercial claim. The appropriate position is **conditional pilot**, not an annual
+purchase recommendation, until `AUD-610-B01` through `B06` are fixed and independently
+retested.
+
+The external price bands ($25-40 and $35-50 per developer/month in the two reports)
+are hypotheses, not pricing evidence. No price change, compliance claim or ENS/CRA
+certification claim follows from this audit. The next commercial proof remains a named
+customer reference, signed historical evidence and validated control mapping.
+

{sourcecode-5.9.3 → sourcecode-5.9.6}/docs/DEFECT-LEDGER.md

@@ -16,6 +16,70 @@ The facts these rows are keyed to are published: `ask schema facts-v1` prints th
 
 ## Current Synchronization
 
+**Field simulation on TeaVM, 2026-08-27 — `ASK-FIELD`.** A deep, non-destructive
+run of ASK `5.9.5` against `/Users/user/Documents/workspace/testing/teavm`
+(4,024 files; 3,539 Java files) exercised the main analysis commands, packs,
+baseline lifecycle, contracts and cache lifecycle. The checkout stayed clean
+and the prior release closure set showed no regression. Three behaviours were
+reproduced and are now closed below: cache warming was not visible to subsequent
+processes, root README summary extraction emitted a Markdown-link fragment, and
+Spring presence disagreed between `spring-audit` and `migrate-check`. The run
+also confirmed that endpoint zero-result disclosure, legal input refusals,
+baseline collision protection, pack `UNVERIFIED` semantics and target
+non-mutation remain intact. This is field evidence, not an external audit; it
+must not be represented as a release certification.
+
+| ID | Severity | Status | Evidence and required closure |
+|---|---:|---|---|
+| `ASK-FIELD-01` | P1 | **closed 5.9.6 (`829f9fa`)** | Explicit warm persists synchronously in its declared namespace; the fresh-process TeaVM lifecycle battery passes. |
+| `ASK-FIELD-02` | P1 | **closed 5.9.6 (`6a659fa`)** | Member annotation imports resolve from the declaring class and confirmed Spring evidence drives presence; TeaVM and Spring control parity passes. |
+| `ASK-FIELD-03` | P2 | **closed 5.9.6 (`55b0d51`)** | Split Markdown badge continuations are discarded without changing normal inline-link unwrapping; the real TeaVM summary regression passes. |
+
+The simulation did not open the Kotlin/Java/Node classification as a defect:
+the signals were explicitly confidence-scored and may represent a legitimate
+polyglot repository. It also did not open zero endpoints, minimum token-budget
+refusal, unsupported `explain --compact`, or contractless `verify`: each matched
+the declared contract and non-coverage in the observed run.
+
+**External audit intake `AUD-610`, release 5.9.3 — 2026-08-27.** The attached
+report scored the build **7.2/10** after five parallel batches and approximately
+120 invocations. It verified a clean target repository, no Java-file changes,
+no commits and no unauthorized persistence. The report praised structured errors,
+output bounding for agents and dry-run safety, but identified 17 defects or
+follow-ups. The full future closure criteria are maintained in
+[`DEVELOPMENT-ROADMAP.md`](DEVELOPMENT-ROADMAP.md); the ledger records their
+status boundary here so a later fix cannot silently close only one consumer.
+
+The report's P1 reliability cluster was `AUD-610-B01` (root YAML compact/agent
+zero-byte failure), `AUD-610-B02` (all `audit-report` formats zero-byte failure),
+`AUD-610-B03` (`risk --limit 0` legal-boundary crash) and `AUD-610-B04`
+(`timeline --since` pays for work then fails silently). None reproduced in the
+TeaVM `5.9.5` simulation; they are held as release-artifact verification items,
+not open deterministic defects.
+
+The remaining rows are: `AUD-610-B05` consumer-path admission (**closed
+5.9.4**, `86ed0a3`); `AUD-610-B06` invalid-input exit parity (**closed**,
+`ae2e8b1`, `f4b4f77`); `AUD-610-B07` command/path typo disambiguation
+(**closed**, `178aa6d`); `AUD-610-B08` pack
+gate ref admission (**closed 5.9.4**, `6690a7f`); `AUD-610-B09` inaccurate
+root help for `--files -` (**closed**, `f66cd28`, `52bcc36`); `AUD-610-B10`
+not-found envelope parity (**closed**, `c07581a`);
+`AUD-610-B11` malformed CVE feed status (**closed 5.9.4**, `abc6319`);
+`AUD-610-B12` unknown posture profiles (**closed 5.9.4**, `f7335b1`, `6e75646`);
+`AUD-610-B13` baseline overwrite (**closed 5.9.4**, `cbe4da9`);
+`AUD-610-B14` CLI/MCP population explanation (**closed**, `7c9b23e`);
+`AUD-610-B15` no-history contract parity (**closed**, `a130637`);
+`AUD-610-B16` UTF-8 replacement character (**platform verification hold**);
+and `AUD-610-B17` one-off retrieve silent failure (**investigation hold, not
+reproduced**).
+
+Commercially, the report recommends conditional pilot/adoption and a provisional
+`$35-50/developer/month` band, requiring closure of the flagship zero-byte paths
+before an annual commitment. This is recorded as external market evidence, not
+as a pricing approval or a product claim. The destructive commands excluded by
+the auditor remain outside the evidence set because only their dry-run paths were
+executed.
+
 **External `5.9.0` audit received, 2026-08-26 - `AUD-608`; corrections shipped in `5.9.1`.** The audited
 revision is the released `5.9.0` commit. The three confirmed corrections are
 closed: `AUD-608-C01` (P1, `cve` now declares its static inventory scope,

{sourcecode-5.9.3 → sourcecode-5.9.6}/docs/DEVELOPMENT-ROADMAP.md

@@ -3,12 +3,37 @@
 *Chief-engineer development record. This is the internal, dev-facing counterpart to the
 customer-facing `docs/EXECUTIVE-ROADMAP.md`.*
 
-> **CURRENT QUEUE — release `5.9.3`; external reconciliation `AUD-608` is implemented.**
+> **CURRENT QUEUE — release `5.9.6`; external reconciliation `AUD-610` is triaged and the reproducible fixes are implemented.**
+> The 2026-08-27 external review adds `AUD-610-B01` through `AUD-610-B10`, led by
+> silent `selftest` nondeterminism, silent root/audit-report failures, a legal-zero
+> `risk --limit 0` crash, an unvalidated consumer path and contradictory Spring
+> detection. The reproducible boundary defects are closed in `5.9.4`: consumer
+> admission (`86ed0a3`), CVE feed integrity (`abc6319`), gate ref admission
+> (`6690a7f`), posture profile admission (`f7335b1`, `6e75646`), baseline
+> collision protection (`cbe4da9`), compact endpoint disclosure (`8a1936f`)
+> and unresolved impact exits (`c07581a`). The remaining external-only reliability
+> observations are held for release-artifact verification and must not be presented
+> as product defects. Full triage is in
+> [`AUDIT-2026-08-27-EXTERNAL-5.9.3.md`](AUDIT-2026-08-27-EXTERNAL-5.9.3.md).
 > The 2026-08-27 external review adds `AUD-609-B01` (overlapping workspace paths
 > double-count Java files) and `AUD-609-B02` (missing workspace paths accepted as
 > empty repositories); both are fixed in `5.9.3`. It reports no regression in the
 > prior closure set. Evidence and the commercial interpretation are in
 > [`AUDIT-2026-08-27-EXTERNAL-5.9.2.md`](AUDIT-2026-08-27-EXTERNAL-5.9.2.md).
+
+> **FIELD SIMULATION — TeaVM, 2026-08-27.** A deep real-use simulation of the
+> released `5.9.6` build was run against
+> `/Users/user/Documents/workspace/testing/teavm` (4,024 files, 3,539 Java
+> files, clean checkout before and after, 96 persisted artefacts). It exercised
+> root JSON/YAML, `spring-audit`, `migrate-check`, `endpoints`, `risk`, `impact`,
+> `impact-chain`, `posture --diff`, audit reports, self-audit, packs, contracts,
+> baselines, attestation and cache lifecycle. No regression was found in the
+> previously closed contract fixes, and the target repository was not modified.
+> It did, however, confirm three follow-up defects: cache persistence across
+> processes, a split Markdown badge fragment in the README summary, and divergent
+> Spring-presence signals. All three are now closed below as `ASK-FIELD-01` through
+> `ASK-FIELD-03`; the fix evidence does not imply that TeaVM is a supported Spring
+> target.
 > `AUD-608-C01` is closed in `fe1f624`: `cve` declares static
 > dependency-inventory coverage and non-coverage. `AUD-608-P01` is closed in
 > `b479465`: `pr-risk --since` constructs its base context and surfaces failures
@@ -119,6 +144,77 @@ triage is in [`AUDIT-2026-08-27-EXTERNAL-5.9.2.md`](AUDIT-2026-08-27-EXTERNAL-5.
 | `AUD-609-C01` | P2 | Signed evidence and historical attestation use different integrity contracts. | Decide whether `baseline attest` gains the existing signing mechanism or clearly publishes that it is unsigned; document the choice in help, schema and buyer-facing material. Do not call unsigned output a certificate or compliance proof. | **Queued, commercial validation.** No implementation claim until the signing boundary and key model are explicit. |
 | `AUD-609-C02` | P3 | Translate technical evidence into concrete ENS/CRA controls. | Define a named control mapping with source evidence, missing evidence and `legal_claim: false`; validate it with a named compliance buyer or auditor before presenting it as a market capability. | **Queued, buyer-gated.** `audit-report --pack ens` remains evidence mapping only. |
 
+### Field Follow-up — TeaVM simulation, 2026-08-27
+
+These rows are engineering follow-ups from a controlled real-repository
+simulation, not findings from an external auditor. They are intentionally
+separate from `AUD-610`: the earlier release audit's non-reproduced observations
+remain non-reproduced, while these three behaviours were reproduced directly on
+the TeaVM checkout. The simulation also confirms that the normal structured-error
+paths, pack verdicts, baseline collision protection, output bounds and target
+repository cleanliness continue to behave as documented.
+
+| ID | Priority | Finding | Acceptance criteria | Status/order |
+|---|---:|---|---|---|
+| `ASK-FIELD-01` | P1 | `cache warm` failed to make the next process observe persisted cache state. | Use one canonical cache namespace and target identity; a fresh process must observe the declared warm state and persisted entries. | **Closed in `5.9.6` (`829f9fa`).** Explicit warm persistence is synchronous and the TeaVM lifecycle battery passes. |
+| `ASK-FIELD-02` | P1 | TeaVM exposed divergent Spring-presence signals between `spring-audit` and `migrate-check`. | Resolve member annotation imports and require confirmed Spring evidence for cross-command presence. | **Closed in `5.9.6` (`6a659fa`).** TeaVM and a Spring control repository now agree; no heuristic broadening was added. |
+| `ASK-FIELD-03` | P2 | Root compact summary leaked a split Markdown Discord badge continuation. | README summaries must never emit URL delimiters or badge fragments while preserving normal inline links. | **Closed in `5.9.6` (`55b0d51`).** The TeaVM README regression and summarizer battery pass. |
+
+**Observed but not opened as defects:** TeaVM was classified as Kotlin
+full-stack with Java as a low-confidence secondary signal and Node.js from a
+manifest; this may be expected multi-language evidence but needs corpus review
+before changing classification. `endpoints` returning zero was accompanied by
+the declared zero-result/non-coverage explanation. `repo-ir --budget-tokens`
+refusing a budget below its minimum estimate, `explain --compact` rejecting an
+unsupported flag, and `verify` returning `UNVERIFIED` without contracts all
+matched their documented contracts. These behaviours must remain in regression
+coverage, but are not roadmap bugs from this run.
+
+### External Audit Intake — `AUD-610` detail, release 5.9.3
+
+The attached external report audited ASK `5.9.3` in five parallel batches
+(approximately 120 invocations), scored it **7.2/10**, and found the audited
+repository clean: no Java files, commits or unauthorized persisted files were
+changed. Its strongest positive evidence was structured error quality (9.0),
+agent-oriented output bounding (9.5) and default non-destructiveness (9.0).
+Its commercial conclusion was **conditional adoption/pilot**, with a suggested
+price band of `$35-50/developer/month`, contingent on closing the flagship
+silent-failure paths before an annual commitment. That price is an external
+hypothesis, not an ASK pricing decision.
+
+The detailed intake below preserves every reported error and its future closure
+boundary. Status is evaluated against the current `5.9.6` documentation and the
+TeaVM simulation above, not blindly copied from the audited `5.9.3` build.
+
+| ID | Severity | Reported issue | Roadmap disposition and future closure |
+|---|---:|---|---|
+| `AUD-610-B01` | P1 | Root YAML compact/agent reportedly returned zero bytes and progress reportedly exited 127. | **Not reproduced in `5.9.5` TeaVM; held as release-artifact verification, not an open product defect.** Current usage/error batteries require structured stderr on failure. |
+| `AUD-610-B02` | P1 | `audit-report` reportedly returned zero bytes across formats and output modes. | **Not reproduced in `5.9.5` TeaVM; held as release-artifact verification.** JSON, YAML, Markdown and file-output paths completed with evidence. |
+| `AUD-610-B03` | P1 | `risk --limit 0` reportedly crashed silently. | **Not reproduced in `5.9.5` TeaVM; held as release-artifact verification.** TeaVM accepted zero as a legal empty-list bound and the boundary battery passes. |
+| `AUD-610-B04` | P1 | `timeline --since HEAD~5` reportedly paid for work and then returned empty output. | **Not reproduced in `5.9.5` TeaVM; held as release-artifact verification.** Timeline failures publish structured diagnostics and sampled failures retain per-sample reasons. |
+| `AUD-610-B05` | P1 | `endpoints --consumer <missing-path>` reportedly behaved like a valid consumer path and produced the same oversized result. | **Closed in `5.9.4` (`86ed0a3`).** Consumer paths now have explicit admission and fail closed; retain the missing-path and valid-path differential battery. |
+| `AUD-610-B06` | P2 | Semantic input errors reportedly had inconsistent process exits. | **Closed by the shared error contract (`ae2e8b1`, `f4b4f77`).** `INVALID_INPUT` is exit 1 and parser/grammar failures are exit 2; usage and subprocess parity batteries pass. |
+| `AUD-610-B07` | P2 | Unknown subcommands were reportedly swallowed as repository paths. | **Closed (`178aa6d`).** Registry-derived dispatch rejects the typo before path fallback and preserves valid lone path arguments. |
+| `AUD-610-B08` | P2 | `pack gate --since <missing-ref>` reportedly collapsed the actionable Git ref error into generic `UNVERIFIED`, unlike direct `pr-impact`. | **Closed in `5.9.4` (`6690a7f`).** Unresolved refs are rejected at admission with structured input evidence; preserve direct-vs-pack parity coverage. |
+| `AUD-610-B09` | P2 | Root help reportedly described `--files -` ambiguously. | **Closed (`f66cd28`, `52bcc36`).** Root and subcommand help state path-list stdin, file and comma-separated shapes; real unified diffs are rejected as invalid input. |
+| `AUD-610-B10` | P2 | Missing symbols reportedly produced divergent impact risk/score semantics. | **Closed (`c07581a` plus `tests/test_direct_callers_unit.py`).** Unresolved paths use the shared `not_found` vocabulary, unknown risk, null score where arithmetic is not applicable and consistent non-zero exit. |
+| `AUD-610-B11` | P2 | Malformed CVE feed reportedly yielded `status: ok` with the parse problem buried in `limitations`. | **Closed in `5.9.4` (`abc6319`).** Feed integrity is fail-closed and statically scoped; preserve malformed, empty and mixed-validity feed tests so a security feed cannot silently degrade to “clean”. |
+| `AUD-610-B12` | P2 | Invented posture profile names were accepted and returned a normal-looking result. | **Closed in `5.9.4` (`f7335b1`, `6e75646`).** Unknown declared populations are rejected; retain profile-admission and cross-command tests. |
+| `AUD-610-B13` | P2 | Repeated baseline capture at the same commit reportedly overwrote the prior evidence without warning or `--force`. | **Closed in `5.9.4` (`cbe4da9`).** Same-commit collisions are rejected; retain duplicate, explicit overwrite and cross-kind tests. |
+| `AUD-610-B14` | P2 | `self-audit` reported a different CLI/MCP count without explaining the populations. | **Closed in `5.9.6` (`7c9b23e`).** `self-audit` publishes both populations, native MCP additions and the relationship formula; TeaVM process regression passes. |
+| `AUD-610-B15` | P3 | `baseline diff`/`trend` failed on no history while `baseline attest` returned `no_history`. | **Closed in `5.9.6` (`a130637`).** Diff, trend and attest now return `status: no_history`, exit 0 and explicit `non_coverage` for empty history. |
+| `AUD-610-B16` | P3 | A replacement character (`�`) reportedly appeared in production JSON around a UTF-8 boundary in posture non-coverage text. | **Open reproduction follow-up.** Run byte-level UTF-8 tests on the exact payload and supported platforms, including em dash/arrow boundaries. If confirmed, fix serialization/truncation at code-point boundaries; if not, record the platform cause and keep the claim withdrawn. |
+| `AUD-610-B17` | P3 | One `retrieve endpoint-security` invocation reportedly failed silently, then succeeded immediately on retry; no second reproduction was obtained. | **Investigation only.** Add correlation IDs, phase-level failure capture and a repeated/concurrent cache-control test. Do not classify a transient single sample as a deterministic defect until reproduced. |
+
+**Audit exclusions and interpretation:** destructive `rename-class` without
+`--dry-run` and `migrate-apply --write` were not executed for safety; the latter
+was additionally blocked by the auditor's sandbox heuristic, not by an ASK
+failure. The report found no regression in the previously closed surfaces and
+confirmed dry-run/non-destructive behaviour. Until `AUD-610-B01` through
+`AUD-610-B04` and the remaining consistency rows are independently closed, ASK
+must retain conditional-pilot positioning and must not promise silent-failure
+freedom or annual-grade reliability.
+
 ### Explicitly Rejected
 
 These are not roadmap work: `--format` and `--compact` remain separate contracts;

{sourcecode-5.9.3 → sourcecode-5.9.6}/docs/EXECUTIVE-ROADMAP.md

@@ -5,7 +5,7 @@
 > `pack assessment` and `pack gate`, and make evidence and safe change the
 > product surface. This roadmap must not introduce a competing product thesis.
 
-> **CURRENT UPDATE — release `5.9.3`, following the external `5.9.0` reconciliation (`AUD-608`).**
+> **CURRENT UPDATE — release `5.9.6`, following the external `5.9.3` reconciliation.**
 > The commercial increment adds stateless context budgets, a reproducible context-cost harness,
 > source-level migration SARIF and an opt-in workspace entitlement pilot. Advisory behaviour remains
 > the default; named-buyer validation is still required before broad enforcement.
@@ -34,6 +34,26 @@
 > Workspace entitlement is therefore safer to measure, but remains advisory by default;
 > no pricing or enforcement decision is authorized by this fix alone.
 
+> **EXTERNAL REVIEW — 5.9.3, received 2026-08-27.** Two independent lenses score the
+> build **7.2/10** and **6.5/10**. The reports confirm clean, non-destructive execution
+> across five batches and strong structured errors on normal validation paths, but open
+> a critical reliability issue: `selftest` intermittently hangs or exits silently with
+> no output. They also reproduce silent failures in the root YAML path and all
+> `audit-report` formats, plus a legal-zero `risk --limit 0` crash, an unvalidated
+> `endpoints --consumer` path and contradictory Spring detection on ANTLR4. These are
+> release-blocking reliability concerns for an annual buyer commitment. The commercial
+> posture is conditional pilot until the flagship failures are fixed and re-tested.
+> Full triage: [`AUDIT-2026-08-27-EXTERNAL-5.9.3.md`](AUDIT-2026-08-27-EXTERNAL-5.9.3.md).
+
+> **5.9.6 reliability correction.** The reproducible contract defects from the
+> latest review are closed: missing consumer paths and malformed CVE feeds now
+> fail closed; `pack gate` rejects unresolved refs; posture profile diffs reject
+> unknown declared-population names; baselines refuse same-commit overwrite;
+> compact endpoint output explains omitted rows; and unresolved impact queries
+> return non-zero even without Java sources. The commercial position remains a
+> conditional pilot because the independent `selftest`, root, `audit-report`,
+> `risk --limit 0` and ANTLR4 observations still require reproduction or closure.
+
 > **HISTORICAL STATUS — release `5.9.0`, following two external audits of `5.8.32`, 2026-08-26.** Two
 > independent rounds scored the release **9.7/10** (reverification) and **8/10** (comparison against
 > `5.8.31`). **Neither round reported a regression**, and both maintained the same commercial

{sourcecode-5.9.3 → sourcecode-5.9.6}/docs/MANUAL-USUARIO.md

@@ -1,6 +1,6 @@
 # Manual de usuario — ASK Engine
 
-> Versión de la herramienta: **5.8.21**  ·  El comando que se ejecuta es `ask`.
+> Versión de la herramienta: **5.9.6**  ·  El comando que se ejecuta es `ask`.
 > (`sourcecode` sigue funcionando como alias en desactivación y reenvía a `ask`.)
 > Público: personas que van a usar `ask` por primera vez, sin conocimiento previo.
 
@@ -8,6 +8,21 @@
 
 ## 1. ¿Qué es ASK Engine?
 
+### Contratos de fiabilidad de la versión 5.9.6
+
+Las entradas inválidas se rechazan antes del análisis costoso mediante el
+envelope estructurado `INVALID_INPUT`: un `--consumer` inexistente, un feed CVE
+mal formado o una referencia `pack gate --since` que no existe nunca se tratan
+como un repositorio vacío ni como `UNVERIFIED`. `posture --diff` valida los
+perfiles desconocidos cuando el repositorio declara su población de perfiles.
+
+`baseline capture` no sobrescribe la captura de un mismo commit; en pipelines
+puede usarse `--auto` para omitirla. En `endpoints --compact`, una lista
+`endpoints: []` significa que las filas se omitieron deliberadamente: consulta
+`endpoints_note`, `endpoints_omitted` y `total`, porque no significa que no haya
+rutas. `impact` e `impact-chain` devuelven código de salida 1 si no resuelven
+el objetivo, también cuando el repositorio no contiene fuentes Java.
+
 **ASK** significa **Actionable Software Knowledge** (conocimiento accionable sobre el
 software); el producto es el *Actionable Software Knowledge Engine*, abreviado
 **ASK Engine**, y el comando que se ejecuta es `ask`.

{sourcecode-5.9.3 → sourcecode-5.9.6}/docs/PRODUCT-ROUTE.md

@@ -49,6 +49,15 @@ customer reference, concrete control translation and buyer evidence are still
 required before changing the price, activating workspace enforcement or claiming
 compliance-market fit. External price opinions are hypotheses, not pricing data.
 
+The 5.9.3 external review changes the commercial gate: ASK may be evaluated in a
+short, bounded pilot, but the product must not make a reliability-led annual
+commitment while `selftest` can fail silently or hang and flagship evidence paths
+can return zero bytes. The differentiator remains credible; the purchase proof is
+not. The route therefore requires the confirmed silent-failure and stack-detection
+items in [`AUDIT-2026-08-27-EXTERNAL-5.9.3.md`](AUDIT-2026-08-27-EXTERNAL-5.9.3.md)
+to close and survive independent retesting before price or compliance positioning
+is strengthened.
+
 ## Individual commands and packs
 
 Every existing command remains supported for users who want a precise answer:

{sourcecode-5.9.3 → sourcecode-5.9.6}/docs/USER_GUIDE.md

@@ -46,15 +46,30 @@ CLI commands        — impact, endpoints, spring-audit, explain, … each a pro
 The key idea: the extraction is **content-addressed**. Commands reuse the parse cache and,
 where their analysed scope matches, the shared Canonical IR; `ask cache model` names what a
 warm buys for each command rather than implying that every projection costs the same. In
-5.9.3, `validation` enters through that shared CIR and `data-exposure` reuses one semantic
+5.9.6, `validation` enters through that shared CIR and `data-exposure` reuses one semantic
 model across all declared label seeds. (The extraction and consumption contract is fixed in
 the architecture ADRs 0001–0004 under `docs/architecture/`.)
 
 The current external pack audit intake is tracked in
 [`AUDIT-2026-08-26-EXTERNAL-5.8.28-PACK.md`](AUDIT-2026-08-26-EXTERNAL-5.8.28-PACK.md);
-the two pack fixes are included in release 5.9.3, with release-artifact verification
+the two pack fixes are included in release 5.9.6, with release-artifact verification
 remaining for detached execution and pack evidence.
 
+### Reliability contracts in 5.9.6
+
+Invalid analysis inputs fail before expensive work and use the structured
+`INVALID_INPUT` envelope: a missing `--consumer` repository, a malformed CVE
+feed, or an unresolved `pack gate --since` ref is never treated as an empty or
+`UNVERIFIED` result. `posture --diff` reports unknown profiles when the project
+declares profiles. `baseline capture` does not replace a snapshot already stored
+for the same commit; use `--auto` when a pipeline should skip it.
+
+With `endpoints --compact`, `endpoints: []` means the rows were deliberately
+omitted. Read `endpoints_note`, `endpoints_omitted` and `total`; an empty list is
+not evidence that the API has no routes. `impact` and `impact-chain` return exit
+code 1 when the target is unresolved, including when the selected repository
+contains no Java sources.
+
 ---
 
 ## The semantic engine
@@ -221,7 +236,7 @@ pipx install sourcecode   # isolated install, no venv needed
 
 # Verify
 ask version
-# ask 5.9.3
+# ask 5.9.6
 ```
 
 Requires Python 3.9+.

{sourcecode-5.9.3 → sourcecode-5.9.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sourcecode"
-version = "5.9.3"
+version = "5.9.6"
 description = "Persistent structural context and ultra-fast repeated analysis for AI coding agents"
 readme = "README.md"
 requires-python = ">=3.9"

{sourcecode-5.9.3 → sourcecode-5.9.6}/src/sourcecode/__init__.py

@@ -4,4 +4,4 @@ ASK Engine is the product. ``ask`` is the canonical CLI command; ``sourcecode``
 the legacy compatibility alias and the Python/PyPI package name. See
 docs/PRODUCT_IDENTITY.md (normative)."""
 
-__version__ = "5.9.3"
+__version__ = "5.9.6"

{sourcecode-5.9.3 → sourcecode-5.9.6}/src/sourcecode/architectural_baseline.py

@@ -445,11 +445,15 @@ def _announce_ask_dir(ask_dir: Path) -> None:
 
 
 def write_baseline(baseline: dict, out_dir: Path) -> Path:
-    """Write `baseline` as deterministic JSON under `out_dir`; return the file path."""
+    """Write a new baseline without replacing an existing commit snapshot."""
     _readonly.guard(out_dir, what="baseline history directory")
     out_dir.mkdir(parents=True, exist_ok=True)
     protect_ask_dir(out_dir)
     path = out_dir / _baseline_filename(baseline)
+    if path.exists():
+        raise FileExistsError(
+            f"Baseline already exists for commit {baseline.get('commit') or '-'}: {path}"
+        )
     path.write_text(
         json.dumps(baseline, sort_keys=True, indent=2) + "\n", encoding="utf-8"
     )