loki-mode 7.45.1 → 7.47.0

package/docs/P2-SPEC-ROBUSTNESS-PLAN.md

@@ -0,0 +1,192 @@
+# P2 Spec Robustness Plan (P2-1 spec interrogation gate + P2-2 assumption ledger)
+
+Status: design for implementation. No version bump, no commit in this arc.
+
+## Goal
+
+Loki must stay accurate even when the input spec is WRONG, ambiguous, or
+incomplete. Today two building blocks already detect spec defects but neither
+feeds the autonomous loop:
+
+- `autonomy/grill.sh` invokes the provider once with a Devil's-Advocate prompt
+  and writes 10-15 hardest spec questions to `.loki/grill/report.md`. It is
+  CLI-only (`grep grill autonomy/run.sh` = 0 invocations) and nothing reads its
+  output.
+- `autonomy/prd-analyzer.py` detects missing PRD dimensions and has a
+  deterministic `_make_assumption()` map, writing `.loki/prd-observations.md`,
+  which nothing reads. Its interactive Q&A is inert in non-TTY (autonomous) runs.
+
+The fix: run interrogation automatically in DISCOVERY, classify the findings,
+record every spec gap as a first-class ASSUMPTION in a tracked ledger, BLOCK
+completion while high-severity assumptions are unconfirmed-and-unacknowledged,
+and surface the ledger in the proof-of-done output. Defects are SURFACED as
+recorded assumptions, never silently autocorrected.
+
+## Core design decision: auto-acknowledgment lifecycle (prevents the trap)
+
+A naive "block completion while any high-severity assumption is unconfirmed"
+hard-blocks EVERY ambiguous run to max-iterations, because in autonomous
+(non-TTY) mode no human can ever set `confirmed=yes`. We never reach the
+"done, plus here is what I assumed" output the goal demands.
+
+Resolution: split the gate from the lifecycle.
+
+- The gate `council_assumption_ledger_gate` is a PURE function of ledger state.
+  It blocks iff an entry has `severity=high AND confirmed=false AND
+  acknowledged=false`.
+- The auto-acknowledgment lifecycle lives in run.sh (NOT in the gate). Once an
+  assumption has been written into the ledger AND injected into the build prompt
+  at least once, run.sh marks it `acknowledged=true`. Default-on; opt-out
+  `LOKI_ASSUMPTIONS_REQUIRE_CONFIRM=1` keeps a human-in-the-loop path where only
+  `confirmed=true` clears the block.
+
+This is the OPPOSITE of silent autocorrect: the assumption is recorded,
+injected into the agent's prompt, and surfaced in proof-of-done. Acknowledgment
+records "Loki has SEEN this gap and proceeded with a stated default", not "Loki
+hid it". The gate still has teeth on the first iteration (high-sev unacknowledged
+blocks) and in the require-confirm path.
+
+## Severity rule (deterministic, no LLM)
+
+grill emits no severity. Classify by section / keyword on the read side:
+
+- HIGH: security blind spots; scale/reliability blind spots; missing or
+  untestable acceptance criteria; any line containing contradiction keywords
+  (contradict, conflict, inconsistent, mutually exclusive).
+- MEDIUM: ambiguities; unstated assumptions; underspecified behavior; all
+  prd-analyzer missing-dimension assumptions.
+
+This guarantees a HIGH tier exists (so the gate has teeth) and is fully
+deterministic (so tests are reproducible).
+
+## Taxonomy mapping (classification, read-side only)
+
+grill section -> finding class:
+- "Ambiguities and missing acceptance criteria" -> ambiguous (HIGH if the line
+  references acceptance criteria / testable / measurable; else MEDIUM)
+- "Unstated assumptions" -> underspecified (MEDIUM)
+- "Security blind spots" -> missing (HIGH)
+- "Scale and reliability blind spots" -> missing (HIGH)
+- any line with a contradiction keyword (any section) -> contradictory (HIGH)
+- prd-analyzer missing dimensions -> missing (MEDIUM, deterministic default)
+
+"None identified." lines are skipped (no fabricated findings).
+
+grill output contract is NOT changed (it is parsed by the loki-grill skill).
+We classify a COPY of its markdown; grill.sh stays byte-identical.
+
+## No-fabrication rule for ledger content
+
+A grill finding is a QUESTION, not a resolution. The ledger `assumption` field
+for a grill-derived gap is an honest "spec gives no answer; proceeding with the
+implementer default for <area>" plus `affects=<area>`. We do NOT invent a
+specific resolution the build will not actually follow. prd-analyzer assumptions
+reuse its existing deterministic `_make_assumption()` text verbatim.
+
+## Ledger schema (`.loki/assumptions/`)
+
+One JSON file per assumption: `.loki/assumptions/<id>.json`, plus a
+human-readable `.loki/assumptions/ledger.md` rollup regenerated on each write.
+Each entry:
+
+```json
+{
+  "id": "a-0001",
+  "gap": "<the spec defect / unanswered question, verbatim>",
+  "assumption": "<honest stated default Loki proceeds with>",
+  "why": "<why this assumption / where the gap came from: grill|prd-analyzer>",
+  "severity": "high|medium",
+  "class": "ambiguous|contradictory|underspecified|missing",
+  "affects": "<area, e.g. security, acceptance-criteria, data-model>",
+  "source": "grill|prd-analyzer",
+  "confirmed": false,
+  "acknowledged": false,
+  "created_at": "<iso8601>"
+}
+```
+
+Stable id = `a-` + zero-padded counter over existing files (idempotent: a second
+DISCOVERY run with the same findings does not duplicate; dedupe on the `gap`
+text hash).
+
+## Build surface (files + functions)
+
+1. NEW `autonomy/spec-interrogation.sh` (sourced by run.sh; standalone-testable):
+   - `spec_interrogation_classify_report <report.md path>`: pure classifier.
+     Reads grill markdown, emits one TSV/JSON finding per question line with
+     class + severity. Takes a file so a fixture report drives the test with no
+     `claude` call.
+   - `spec_interrogation_severity_for <section> <line>`: deterministic severity.
+   - `spec_ledger_write <gap> <assumption> <why> <severity> <class> <affects>
+     <source>`: idempotent writer (dedupe on gap hash) -> `.loki/assumptions/`.
+   - `spec_ledger_rebuild_md`: regenerate `.loki/assumptions/ledger.md`.
+   - `spec_ledger_high_unresolved_count`: count entries with
+     `severity=high AND confirmed=false AND acknowledged=false` (gate input;
+     also reused by the summary).
+   - `spec_ledger_acknowledge_all`: set `acknowledged=true` on all entries
+     (auto-ack lifecycle helper; default path).
+   - `spec_interrogation_run <spec_path>`: orchestrator. Default-on
+     (`LOKI_SPEC_GRILL=0` opts out). Provider-aware: source grill.sh, call
+     `grill_check_provider`; if provider absent, log honest message, skip the
+     grill subcall (NO fabricated questions), but STILL fold prd-analyzer
+     missing-dimension assumptions into the ledger so degrade surfaces something
+     non-blocking. On provider present: run `grill_main` (writes report.md),
+     classify it, write ledger entries. Always non-fatal to the run.
+
+2. `autonomy/run.sh` DISCOVERY (~13056, after prd-analyzer + council_init,
+   before the iteration loop): source spec-interrogation.sh and call
+   `spec_interrogation_run "$prd_path"`. This is the grep-able grill invocation
+   the task requires. Best-effort (`|| true`), never blocks startup.
+
+3. `autonomy/run.sh` auto-ack lifecycle: after the build prompt is constructed
+   each iteration (assumptions are injected into the prompt via build_prompt),
+   call `spec_ledger_acknowledge_all` UNLESS
+   `LOKI_ASSUMPTIONS_REQUIRE_CONFIRM=1`. Inject the high-severity assumption
+   list into the build prompt (so the agent sees the gaps it must respect).
+
+4. `autonomy/completion-council.sh` `council_assumption_ledger_gate` (new),
+   slotted into `council_evaluate` right after `council_evidence_gate`
+   (mirrors 2510-2513). Same defensive `COUNCIL_STATE_DIR` default, opt-out
+   `LOKI_ASSUMPTION_GATE=0`. Blocks iff `spec_ledger_high_unresolved_count > 0`.
+   Writes `.loki/council/assumption-block.json` on block, removes it on pass.
+   Also wired into the completion-promise route in run.sh (~14525 pattern) and
+   the code_review gate chain (~15013) so the promise path cannot bypass it.
+
+5. `autonomy/run.sh` `build_completion_summary` (~2637): emit an
+   "Assumptions recorded: N (M high-severity)" block into COMPLETION.txt and the
+   ledger list, plus the count into completion.json. So "done" means "done, plus
+   here are the N places your spec was ambiguous and what I assumed."
+
+6. NEW `tests/test-spec-interrogation.sh` (bash convention, ok/bad counters,
+   source the module, mktemp fixtures):
+   - (a) classifier on a fixture grill report writes classified findings to the
+     ledger (ambiguous/contradictory/underspecified/missing + high/medium).
+   - (b) a ledger with one high/confirmed:false/acknowledged:false entry makes
+     `council_assumption_ledger_gate` return 1 (BLOCK) and write
+     assumption-block.json.
+   - (c) clean spec (no high-sev entries, or all acknowledged) -> gate returns 0
+     (no spurious block), no block file.
+   - (d) no provider -> `spec_interrogation_run` degrades cleanly: honest
+     message, prd-analyzer assumptions still folded (medium, non-blocking), run
+     proceeds, gate passes.
+
+## Gate reachability (resolved open question)
+
+The existing gates fire from THREE sites: `council_evaluate` (~2510), the
+completion-promise route (~14525), and the code_review gate chain (~15013). The
+new gate is wired into all three so high-sev unacknowledged assumptions cannot
+slip through the promise path.
+
+## Opt-out knobs (all default-on, intelligent)
+
+- `LOKI_SPEC_GRILL=0` -> skip interrogation entirely.
+- `LOKI_ASSUMPTION_GATE=0` -> gate is pass-through.
+- `LOKI_ASSUMPTIONS_REQUIRE_CONFIRM=1` -> require human `confirmed=true`
+  (disables auto-ack); the human-in-the-loop path.
+
+No "user must decide the type" knob. Classification + severity are automatic.
+
+## Constraints
+
+No emojis, no em dashes, no version bump, no commit, no push. Provider-aware,
+degrade cleanly, no fabricated questions when provider absent.

package/docs/R9-OPEN-CORE-HOOKS-PLAN.md

@@ -3,8 +3,8 @@
 Status: SEAMS implemented (this worktree). NOT a live hosted backend.
 
 R9 in the competitive-stickiness arc is the open-core monetization layer: keep
-Loki fully open source and free, while adding the SEAMS where hosted, enterprise,
-and paid plans would attach later. R9 ships the seams only. There is no Loki
+Loki fully source-available (BUSL-1.1) and free to self-host, while adding the
+SEAMS where hosted, enterprise, and paid plans would attach later. R9 ships the seams only. There is no Loki
 hosted service, no license-verification backend, and no paid gate on any
 existing feature. Every honest stub is labeled as such.
 

package/docs/architecture/STATE-MACHINES.md

@@ -972,7 +972,7 @@ Source: `run.sh:7880-7881` (checklist_should_verify, checklist_verify)
 
 ## 7. Quality Gates
 
-### 7.1 Nine-Gate Pipeline
+### 7.1 Eight-Gate Pipeline
 
 Source: `skills/quality-gates.md`
 
@@ -980,40 +980,39 @@ Source: `skills/quality-gates.md`
   Code Change
       |
       v
-  Gate 1: Static Analysis (CodeQL, ESLint)
-      |──BLOCK (critical findings)──> [REJECTED]
+  Gate 1: Static Analysis (CodeQL, ESLint/Pylint, type-checker)
+      |──BLOCK (severity ladder)──> [REJECTED]
       v
-  Gate 2: Type Check (tsc --noEmit)
+  Gate 2: Test Suite (pass/fail; red blocks; coverage % not measured this release)
       |──BLOCK──> [REJECTED]
       v
-  Gate 3: Unit Tests (>80% coverage, 100% pass)
-      |──BLOCK──> [REJECTED]
-      v
-  Gate 4: Integration Tests
-      |──BLOCK──> [REJECTED]
-      v
-  Gate 5: 3-Reviewer Blind Review (see 7.3)
+  Gate 3: Blind 3-Reviewer Review with severity blocking (see 7.3)
       |──BLOCK (Critical/High severity)──> [REJECTED]
       v
-  Gate 6: Anti-Sycophancy Check
-      |──BLOCK (devil's advocate finds issues)──> [REJECTED]
+  Gate 4: Anti-Sycophancy Devil's Advocate (on unanimous PASS)
+      |──BLOCK (devil's advocate Crit/High findings)──> [REJECTED]
       v
-  Gate 7: Security Scan
-      |──BLOCK──> [REJECTED]
+  Gate 5: Mock Integrity Detector
+      |──BLOCK (HIGH findings)──> [REJECTED]
       v
-  Gate 8: Performance Check
-      |──BLOCK──> [REJECTED]
+  Gate 6: Test Mutation Detector
+      |──BLOCK (HIGH findings)──> [REJECTED]
       v
-  Gate 9: E2E / Playwright
+  Gate 7: Documentation Coverage
       |──BLOCK──> [REJECTED]
       v
+  Gate 8: Magic Modules Debate
+      |──BLOCK (BLOCK-severity findings)──> [REJECTED]
+      v
   [APPROVED]
 ```
 
+Backward-compatibility is a conditional healing-mode auditor, not a numbered gate.
+
 Gate status values: `passed`, `failed`, `skipped`
 Persistence: `.loki/dashboard-state.json` field `qualityGates`
 Severity levels: `critical`, `high`, `medium`, `low`
-Blocking threshold: Critical and High always block; Medium blocks by default.
+Blocking threshold: Critical and High block; Medium and Low are advisory.
 
 ### 7.2 Model Escalation
 

package/docs/architecture/bmad-loki-voice-agent-council-analysis.md

@@ -57,5 +57,5 @@ architecture, and adversarial review -- complementing Loki Mode's autonomous exe
 1. P0 must ship independently and prove value before P1/P2 begin
 2. No runtime dependency on BMAD repo -- adapter reads BMAD output artifacts only
 3. Zero regression on existing non-BMAD workflows
-4. All code must pass existing 9-gate quality system
+4. All code must pass existing 8-gate quality system
 5. Context budget: BMAD additions must stay under 15K tokens per iteration

package/docs/auto-claude-comparison.md

@@ -120,21 +120,24 @@ Loki Mode implements CONSENSAGENT (ACL 2025):
 **Verdict: Loki Mode wins** - Research-backed quality assurance.
 
 ### 5. Quality Gates
-Loki Mode has 14 quality gates:
+Loki Mode runs 8 deterministic quality gates plus full SDLC phase coverage.
+
+The 8 deterministic quality gates: static analysis (CodeQL, ESLint), test suite (pass/fail), blind 3-reviewer review with severity blocking, anti-sycophancy Devil's Advocate, mock-integrity, test-mutation, documentation coverage, and Magic Modules debate. (Backward-compatibility is a conditional healing-mode auditor, not a numbered gate.)
+
+Beyond the gates, the SDLC pipeline covers these phases:
 1. Static analysis (CodeQL, ESLint)
-2. Unit tests (>80% coverage)
+2. Unit tests (test suite passes; coverage % not measured this release)
 3. API/Integration tests
 4. E2E tests (Playwright)
 5. Security scanning (OWASP)
-6. SAML/OIDC/SSO integration
-7. Parallel code review (3 reviewers)
-8. Performance/load testing
-9. Accessibility (WCAG)
-10. Regression testing
-11. UAT simulation
-12. Anti-sycophancy check
-13. Scale-aware review intensity
-14. Continuous monitoring
+6. Parallel code review (3 reviewers)
+7. Performance/load testing
+8. Accessibility (WCAG)
+9. Regression testing
+10. UAT simulation
+11. Anti-sycophancy check
+12. Scale-aware review intensity
+13. Continuous monitoring
 
 **Auto-Claude:** Single QA validation loop (up to 50 iterations).
 
@@ -236,7 +239,7 @@ Loki Mode now incorporates proven patterns from Cursor's large-scale agent deplo
 3. **Anti-Sycophancy** - Blind review prevents false positives
 4. **Full SDLC** - Business, marketing, growth automation
 5. **Published Benchmarks** - Verify claims with reproducible tests
-6. **MIT License** - More adoption-friendly
+6. **Source-available (BUSL-1.1)** - Inspect and self-host the full code
 
 ---
 
@@ -253,7 +256,7 @@ Loki Mode now incorporates proven patterns from Cursor's large-scale agent deplo
 - Full spec-to-product lifecycle (not just coding)
 - 41 specialized agent roles
 - Anti-sycophancy measures
-- MIT license
+- Source-available (BUSL-1.1) license
 - No subscription requirement
 - Verified benchmarks
 

package/docs/certification/01-core-concepts/lesson.md

@@ -105,19 +105,20 @@ Full agent type definitions are in `references/agent-types.md`.
 
 ## Quality Gates
 
-Loki Mode enforces a 9-gate quality system. Code must pass all applicable gates before moving forward:
+Loki Mode enforces an 8-gate quality system. Code must pass all applicable gates before moving forward:
 
 | Gate | Name | Purpose |
 |------|------|---------|
-| 1 | Input Guardrails | Validate scope, detect injection, check constraints |
-| 2 | Static Analysis | CodeQL, ESLint/Pylint, type checking |
-| 3 | Blind Review System | 3 specialist reviewers in parallel, blind to each other |
-| 4 | Anti-Sycophancy Check | If reviewers unanimously approve, run a Devil's Advocate reviewer |
-| 5 | Output Guardrails | Validate code quality, spec compliance, no secrets |
-| 6 | Severity-Based Blocking | Critical/High/Medium = BLOCK; Low/Cosmetic = TODO |
-| 7 | Test Coverage Gates | Unit: 100% pass, >80% coverage; Integration: 100% pass |
-| 8 | Mock Detector | Flags tests that mock internal modules instead of real code |
-| 9 | Test Mutation Detector | Detects assertion value changes alongside implementation changes |
+| 1 | Static Analysis | CodeQL, ESLint/Pylint, type checking |
+| 2 | Test Suite (pass/fail) | Red blocks; coverage % not measured this release |
+| 3 | Blind Code Review (3-reviewer council + severity blocking) | 3 specialist reviewers in parallel, blind to each other; Critical/High = BLOCK; Medium/Low advisory |
+| 4 | Anti-Sycophancy / Devil's Advocate | If reviewers unanimously approve, run a Devil's Advocate reviewer |
+| 5 | Mock Integrity Detector | Flags tests that mock internal modules instead of real code |
+| 6 | Test Mutation Detector | Detects assertion value changes alongside implementation changes |
+| 7 | Documentation Coverage | README exists, docs freshness, API docs for packages |
+| 8 | Magic Modules Debate | Spec-vs-implementation debate on generated Magic Modules |
+
+A conditional backward-compatibility / legacy-healing auditor also runs in healing mode (not one of the 8 numbered gates).
 
 The blind review system (Gate 3) selects 3 reviewers from a pool of 5 named specialists:
 
@@ -179,4 +180,4 @@ Every Loki Mode project uses these files in the `.loki/` directory:
 
 ## Summary
 
-Loki Mode is an autonomous multi-agent system that follows the RARV cycle to build software from PRDs. It uses 41 agent types organized into 8 domains, enforces quality through 9 gates with blind peer review, and maintains episodic/semantic/procedural memory for continuous learning. Projects are classified into simple, standard, or complex tiers that determine the number of phases executed.
+Loki Mode is an autonomous multi-agent system that follows the RARV cycle to build software from PRDs. It uses 41 agent types organized into 8 domains, enforces quality through 8 gates with blind peer review, and maintains episodic/semantic/procedural memory for continuous learning. Projects are classified into simple, standard, or complex tiers that determine the number of phases executed.

package/docs/certification/01-core-concepts/quiz.md

@@ -45,7 +45,7 @@ D) test-coverage-auditor
 A) 3
 B) 5
 C) 7
-D) 9
+D) 8
 
 ---
 
@@ -67,12 +67,12 @@ D) complex
 
 ---
 
-**Question 8:** What is the minimum test coverage required by Gate 7 (Test Coverage Gates)?
+**Question 8:** What does Gate 7 (Documentation Coverage) check?
 
-A) 50%
-B) 60%
-C) 80%
-D) 100%
+A) That unit test coverage is at least 80%
+B) That every function has an inline comment
+C) That a README exists, docs are fresh within 10 commits, and packages have API docs
+D) That cyclomatic complexity stays under 10
 
 ---
 

package/docs/certification/05-troubleshooting/lesson.md

@@ -17,30 +17,40 @@ This module covers diagnosing and resolving common issues in Loki Mode: gate fai
 
 ## Quality Gate Failures
 
-When a quality gate fails, identify which gate triggered the failure:
+When a quality gate fails, identify which gate triggered the failure (the 8-gate
+system is detailed in `skills/quality-gates.md`):
 
-**Gates 1-6 (Review gates):**
+**Gates 1-2 (Static analysis and test suite):**
+- Gate 1 (Static Analysis): fix CodeQL/ESLint/Pylint/type-checker findings
+- Gate 2 (Test Suite): the test runner must pass; red blocks. Coverage % is not
+  measured this release. Fix failing tests before proceeding (never delete or
+  skip tests)
+
+**Gates 3-4 (Review gates):**
 - Check the review output for severity levels
-- Critical/High/Medium = BLOCK (must fix)
+- Critical/High = BLOCK; Medium/Low advisory (recommended to fix)
 - Low/Cosmetic = TODO (informational)
 - If all 3 reviewers pass unanimously, Gate 4 runs Devil's Advocate
 
-**Gate 7 (Test coverage):**
-- Unit tests must have 100% pass rate and >80% coverage
-- Integration tests must have 100% pass rate
-- Fix failing tests before proceeding (never delete or skip tests)
-
-**Gate 8 (Mock detector):**
+**Gate 5 (Mock integrity detector):**
 - Runs `tests/detect-mock-problems.sh`
 - Flags tests that mock internal modules instead of using real code
 - Flags tautological assertions and high internal mock ratios
-- Disable with `LOKI_GATE_MOCK_DETECTOR=false` (not recommended)
+- Disable with `LOKI_GATE_MOCK=false` (not recommended)
 
-**Gate 9 (Test mutation detector):**
+**Gate 6 (Test mutation detector):**
 - Runs `tests/detect-test-mutations.sh`
 - Detects assertion values changed alongside implementation (test fitting)
-- Detects low assertion density and missing pass/fail tracking
-- Disable with `LOKI_GATE_MUTATION_DETECTOR=false` (not recommended)
+- Detects low assertion density
+- Disable with `LOKI_GATE_MUTATION=false` (not recommended)
+
+**Gate 7 (Documentation coverage):**
+- Checks README presence, docs freshness within 10 commits, and API docs for packages
+- Disable with `LOKI_GATE_DOC_COVERAGE=false` (not recommended for packages)
+
+**Gate 8 (Magic Modules debate):**
+- Runs the spec-vs-implementation debate on generated Magic Modules
+- BLOCK-severity findings block; disable with `LOKI_GATE_MAGIC_DEBATE=false`
 
 ## Circuit Breaker System
 

package/docs/certification/05-troubleshooting/quiz.md

@@ -67,11 +67,11 @@ D) Removes the entire `.loki/` directory
 
 ---
 
-**Question 8:** Which environment variable disables Gate 8 (Mock Detector)?
+**Question 8:** Which environment variable disables Gate 5 (Mock Integrity Detector)?
 
-A) `LOKI_SKIP_MOCK_CHECK=true`
+A) `LOKI_GATE_MOCK=false`
 B) `LOKI_GATE_MOCK_DETECTOR=false`
-C) `LOKI_DISABLE_GATE_8=true`
+C) `LOKI_DISABLE_GATE_5=true`
 D) `LOKI_NO_MOCK_DETECTION=true`
 
 ---

package/docs/certification/README.md

@@ -73,7 +73,7 @@ docs/certification/
 
 ## Cost and Licensing
 
-This certification program is **free and open source**, released under the same license as Loki Mode. No registration or payment is required.
+This certification program is **free and source-available**, released under the same license as Loki Mode (BUSL-1.1). No registration or payment is required.
 
 ## Version
 

package/docs/certification/answer-key.md

@@ -12,10 +12,10 @@ This file contains answers for all module quizzes and the final certification ex
 | 2 | C | 41 agent types: 37 domain + 4 orchestration |
 | 3 | B | After 5 failures, the task moves to `.loki/queue/dead-letter.json` |
 | 4 | C | architecture-strategist is always one of the 3 selected reviewers |
-| 5 | D | 9 quality gates (Input Guardrails through Test Mutation Detector) |
+| 5 | D | 8 quality gates (Static Analysis through Magic Modules Debate); backward-compatibility is a conditional healing-mode auditor, not one of the 8 |
 | 6 | B | Episodic, semantic, and procedural memory |
 | 7 | B | Simple tier uses 3 phases |
-| 8 | C | Gate 7 requires >80% unit test coverage |
+| 8 | C | Gate 7 (Documentation Coverage) checks README presence, docs freshness within 10 commits, and API docs for packages; coverage % is not measured this release |
 | 9 | C | Claude Code supports full features; Codex and Gemini run in degraded mode |
 | 10 | B | If all 3 reviewers unanimously approve, a Devil's Advocate reviewer runs |
 

package/docs/certification/certification-exam.md

@@ -49,7 +49,7 @@ D) test-coverage-auditor
 A) 3
 B) 5
 C) 7
-D) 9
+D) 8
 
 ---
 
@@ -71,12 +71,12 @@ D) complex
 
 ---
 
-**Question 8:** What is the minimum test coverage required by Gate 7?
+**Question 8:** What does Gate 7 (Documentation Coverage) check?
 
-A) 50%
-B) 60%
-C) 80%
-D) 100%
+A) That unit test coverage is at least 80%
+B) That every function has an inline comment
+C) That a README exists, docs are fresh within 10 commits, and packages have API docs
+D) That cyclomatic complexity stays under 10
 
 ---
 
@@ -439,11 +439,11 @@ D) Removes the entire `.loki/` directory
 
 ---
 
-**Question 48:** Which environment variable disables Gate 8 (Mock Detector)?
+**Question 48:** Which environment variable disables Gate 5 (Mock Integrity Detector)?
 
-A) `LOKI_GATE_MOCK_DETECTOR=false`
+A) `LOKI_GATE_MOCK=false`
 B) `LOKI_SKIP_MOCK_CHECK=true`
-C) `LOKI_DISABLE_GATE_8=true`
+C) `LOKI_DISABLE_GATE_5=true`
 D) `LOKI_NO_MOCK_DETECTION=true`
 
 ---

package/docs/competitive/bolt-new-analysis.md

@@ -409,7 +409,7 @@ These are bolt.new weaknesses that Loki Mode already solves or can emphasize:
 
 #### R5: Advertise Production Readiness as Key Differentiator
 - **bolt.new's gap**: 70% done code, no tests, no review, $5-20K remediation
-- **Loki Mode's advantage**: RARV cycle, 10 quality gates, 3-reviewer system, automated testing
+- **Loki Mode's advantage**: RARV cycle, 8 quality gates, 3-reviewer system, automated testing
 - **Action**: Create comparison content showing: "bolt.new gives you a prototype. Loki Mode gives you a product."
 - **Messaging**: "From PRD to production, not PRD to prototype"
 
@@ -529,7 +529,7 @@ The "vibe coding" market -- AI tools that generate code from natural language --
 | Lovable | Lovable | $100M ARR in 8 months | Design-quality prototypes |
 | Vercel | v0 | Part of Vercel ($3.5B+) | UI component generator |
 | Replit | Replit | $1.16B valuation | Browser IDE + AI |
-| Autonomi | Loki Mode | Open source, early stage | PRD-to-production system |
+| Autonomi | Loki Mode | Source-available (BUSL-1.1), early stage | PRD-to-production system |
 
 ### 11.4 Strategic Implication for Loki Mode