Golden Chain Pipeline

The 17-link Golden Chain development cycle — spec-driven, benchmark-gated, self-assessing.

pipeline

Execute the full Golden Chain development cycle.

Aliases: chain

tri pipeline run <task>        # Execute full 17-link cycle
tri pipeline status            # Show current pipeline state
tri pipeline resume            # Resume from checkpoint

Subcommands

Subcommand	Description
run [task]	Execute full 17-link Golden Chain cycle
status	Show current pipeline state
resume	Resume from last checkpoint

The 17 Links

Links are numbered 0-16 (17 total). The CLI banner displays "16 Links" counting from Link 0.

#	Link	Description	Critical
0	TVC_GATE	Search corpus, return cached or continue	Yes
1	BASELINE	Analyze previous version v(n-1)
2	METRICS	Collect v(n-1) performance metrics
3	PAS_ANALYZE	Research patterns and science
4	TECH_TREE	Build technology dependency graph
5	SPEC_CREATE	Create .vibee specifications
6	CODE_GENERATE	Generate code from specs (vibee gen)
7	TEST_RUN	Run test suite (zig build test)	Yes
8	BENCHMARK_PREV	Compare to v(n-1) — regression gate	Yes
9	BENCHMARK_EXTERNAL	Compare to llama.cpp / vLLM
10	BENCHMARK_THEORETICAL	Gap to theoretical maximum
11	DELTA_REPORT	Generate improvement report
12	OPTIMIZE	Fix if needed (optional)
13	DOCS	Generate documentation + proofs
14	TOXIC_VERDICT	Self-assessment (Russian)
15	GIT	Commit + push changes
16	LOOP_DECISION	Decide next iteration	Yes

Example output:

Golden Chain Pipeline - 16 Links
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Link  0: TVC_GATE [CRITICAL]
  [OK] TVC_GATE (50ms)

Link  1: BASELINE
  [OK] BASELINE (100ms)

...

Link 14: TOXIC_VERDICT [CRITICAL]
  [OK] TOXIC_VERDICT (25ms)

================================================================
              GOLDEN CHAIN CLOSED
================================================================

Completed: 15/17 links
Improvement: 1.23%
Threshold: 61.80% (phi^-1)

KOSCHEI IMMORTAL! Needle is sharp.

Needle Status (Immortality Check)

The pipeline uses the Koschei Needle metaphor. Improvement rate is measured against the golden threshold (phi^-1 = 0.618):

Status	Condition	Message
Immortal	rate > 61.8%	KOSCHEI BESSMERTEN! Igla ostra.
Mortal	0 < rate < 61.8%	Uluchshenie est', no Igla tupitsya.
Regression	rate <= 0	REGRESSIYA! Igla slomana.

Improvement Rate Formula

Weighted across four dimensions:

Dimension	Weight	Metric
Performance	40%	current_tps / prev_tps - 1
Memory	30%	prev_mem / curr_mem - 1
Test Coverage	20%	curr_tests / prev_tests - 1
Accuracy	10%	curr_accuracy - prev_accuracy

decompose

Break a task into sub-tasks (Links 3-4 of the pipeline).

tri decompose <task description>
tri decompose "add user authentication"

Example output:

Task Decomposition (Links 3-4)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Task: add user authentication

Sub-tasks identified:
  1. Analyze existing codebase
  2. Create .vibee specification
  3. Generate code from spec
  4. Write tests
  5. Run benchmarks
  6. Document changes

Use 'tri pipeline run' to execute full cycle

plan

Generate .vibee specifications from sub-tasks (Link 5).

tri plan
tri plan --file tasks.json

spec-create

Create a .vibee specification template (Link 6).

Aliases: spec_create

tri spec-create <name>
tri spec-create my_module

Generates a template file at specs/tri/<name>.vibee with the standard VIBEE structure (name, version, language, module, types, behaviors).

loop-decide

Loop decision gate (Link 17) — determines whether to continue iterating or exit.

Aliases: loop_decide

tri loop-decide              # Auto mode (evaluate EXIT_SIGNAL)
tri loop-decide auto         # Same as above
tri loop-decide continue     # Force CONTINUE
tri loop-decide exit         # Force EXIT

The loop decision evaluates the EXIT_SIGNAL criteria:

EXIT_SIGNAL = (
    tests_pass AND
    spec_complete AND
    critical_assessment_written AND
    tech_tree_options_proposed AND
    achievement_documented AND
    dashboard_widget_updated AND
    committed
)

Condition	Check
tests_pass	zig build test succeeds
spec_complete	.vibee spec exists for all generated code
critical_assessment_written	Toxic verdict (Link 14) completed
tech_tree_options_proposed	3 options for next iteration listed
achievement_documented	Report written in docsite/docs/research/
dashboard_widget_updated	Canvas mirror widget reflects new state
committed	Changes committed to git

verify

Run tests + benchmarks (Links 7-11).

tri verify

Example output:

Verification (Links 7-11)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Link 7: Running Tests...
  [OK] Tests passed

Link 8: Running Benchmarks...
  [OK] Benchmark: 125us (1000 iterations)

Verification complete
  Tests: PASS
  Benchmarks: No regression detected

verdict

Generate a toxic self-assessment verdict (Link 14).

tri verdict

Produces a structured Russian-style self-critique:

Example output:

TOXIC VERDICT (Link 14)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  WHAT WAS DONE:
    - Implemented VSA bind/unbind optimization
    - Added 12 new test cases
    - Benchmark: 1.4 Gops/s (↑23%)

  WHAT FAILED:
    - Memory usage still 2x theoretical minimum
    - No WASM target coverage

  TECH TREE OPTIONS:
    A) Optimize memory layout (packed trits)
    B) Add WASM compilation target
    C) Implement GPU offload via Metal

  NEEDLE STATUS:
    Improvement: 23.4%
    Threshold:   61.8% (phi^-1)
    Status:      Igla tupitsya (needle dulling)

Error Recovery

The pipeline handles errors with different strategies:

Error Type	Strategy	Example
Critical link failure	Abort	Tests fail, benchmark regression
Benchmark timeout	Retry	External benchmark slow
Non-critical failure	Skip	External benchmark unavailable
Git conflict	Manual	Wait for user intervention

Constants

PHI           = 1.618033988749895
PHI_INVERSE   = 0.618033988749895   (Needle threshold)
TRINITY       = 3.0                 (phi^2 + 1/phi^2)