Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 25 additions & 16 deletions docs/project_source.md
Original file line number Diff line number Diff line change
@@ -1,22 +1,22 @@
# Project Source: CompText V7

## Canonical Positioning
CompTextv7 is the deterministic replay-integrity layer for compressed operational agent traces.”
CompText V7 is a deterministic replay-validation prototype for compact operational agent/MCP traces, with a KVTC-V7 technical-log compression prototype.”

## Strategic Category
CompTextv7 is a deterministic replay-integrity validation research prototype.
CompText V7 is a deterministic replay-validation research prototype.

## Core Thesis
The core project question is:

“Can a compressed agent state still reproduce a safe operational trajectory?”
“Can compact operational trace state still reproduce a safe replay trajectory?”

The governing research direction is:

“Deterministic Replay Integrity Validation for Compressed Operational Agent Traces
“Deterministic replay validation for compact operational agent/MCP traces

## Research Alignment Principle
Every accepted change must strengthen deterministic, offline, reproducible replay-integrity validation for compressed operational traces, or be strictly maintenance-only.
Every accepted change must strengthen deterministic, offline, reproducible replay validation for compact operational agent/MCP traces, or be strictly maintenance-only.

## Current Strategic State
The project remains:
Expand All @@ -29,6 +29,15 @@ The project remains:
- audit-oriented

## Current Implementation State
Current implemented surfaces:
- curated agent trace fixtures in `tests/fixtures/agent_traces/`
- deterministic replay runner in `tests/utils/agent_trace_replay_runner.py`
- MCP replay payload layer in `src/comptext_v7/mcp/`
- evidence survival helpers in `src/validation/evidence.py`
- replay failure labels in `src/validation/replay_failure_classifier.py`
- committed artifacts such as `artifacts/agent_trace_replay_results.json`
- KVTC-V7 technical-log compression prototype in `src/core/kvtc_v7.py`

The `mcp_trace_replay` fixture family is hardened via explicit capability-boundary structures, deterministic degraded variants, and `failure_label_on_violation` handling.

A deterministic artifact evidence index maps committed evidence artifacts to their generators, evidence categories, fixture-family coverage, manifest alignment, deterministic evaluation status, LLM-free status, and external-API-free status.
Expand All @@ -37,11 +46,11 @@ A deterministic artifact evidence index maps committed evidence artifacts to the
See `NON_GOALS.md` in the repository root for the full list of prohibited expansion paths and the hard scope test.

## Hard Scope Rule
A proposal is out of scope unless it directly improves deterministic replay-integrity validation for compressed operational traces without introducing probabilistic or platform-expansion behavior.
A proposal is out of scope unless it directly improves deterministic replay validation for compact operational agent/MCP traces without introducing probabilistic or platform-expansion behavior.

## Scope Evolution & RFC Process
Scope-expanding proposals require an RFC before implementation. RFCs must:
1. define the exact replay-integrity gain,
1. define the exact replay-validation gain,
2. prove deterministic and offline reproducibility,
3. show no hidden orchestration/platform expansion,
4. describe artifact and failure-taxonomy implications,
Expand All @@ -50,21 +59,21 @@ Scope-expanding proposals require an RFC before implementation. RFCs must:
Default decision for scope-expanding proposals is rejection until the RFC is accepted.

## Correct Abstraction Level
CompTextv7 focuses on replay-integrity contracts, trace-derived artifacts, fixture families, deterministic validators, and auditable evidence outputs. It does not absorb adjacent runtime or product-platform responsibilities.
CompText V7 focuses on replay contracts, trace-derived artifacts, fixture families, deterministic validators, and auditable evidence outputs. It does not absorb adjacent runtime or product-platform responsibilities.

## Preferred Terminology
Prefer:
- deterministic replay-integrity validation
- compressed operational traces
- deterministic replay validation
- compact operational agent/MCP traces
- artifact evidence
- failure taxonomy and labels
- capability boundary
- offline reproducibility

Avoid vague or marketing abstractions that hide deterministic constraints.
Avoid vague or marketing abstractions that hide deterministic constraints. Do not frame the repository as industrial, enterprise-ready, production-ready, certification-grade, showcase-first, or a cognitive fabric.

## Roadmap From Here
1. Expand deterministic replay-integrity fixture coverage.
1. Expand deterministic replay-validation fixture coverage.
2. Strengthen artifact-evidence traceability and reproducibility checks.
3. Improve contract-level validation clarity and failure classification precision.
4. Add narrowly scoped capabilities only via approved RFC.
Expand All @@ -73,11 +82,11 @@ Avoid vague or marketing abstractions that hide deterministic constraints.
- Documentation-only governance changes should stay small and focused.
- Core logic changes must include targeted tests.
- No merge-ready state until CI is green and review threads are resolved/outdated.
- Keep docs, showcase, and core refactors separated.
- Keep unrelated docs, demos, and core refactors separated.

## PR Template Additions
PR descriptions should explicitly confirm:
- deterministic replay-integrity alignment,
- deterministic replay-validation alignment,
- non-goals preserved,
- artifact regeneration expectations,
- evidence-index impact (updated or unchanged),
Expand All @@ -93,10 +102,10 @@ PR descriptions should explicitly confirm:
Governance is conservative, boundary-enforcing, and evidence-led. CompTextv7 accepts incremental improvements over broad platform ambitions.

## Current Priority Stack
1. Deterministic replay-integrity correctness.
1. Deterministic replay-validation correctness.
2. Artifact evidence traceability.
3. Failure-label clarity and consistency.
4. Scope-bound documentation and review discipline.

## One-Sentence Rule
If a change does not directly improve deterministic replay-integrity validation for compressed operational traces, it should not merge into core.
If a change does not directly improve deterministic replay validation for compact operational agent/MCP traces, it should not merge into core.
15 changes: 10 additions & 5 deletions docs/validation.md
Original file line number Diff line number Diff line change
Expand Up @@ -21,10 +21,6 @@ convenience. It does not define workspaces, dependencies, or a root Node app.
The dashboard remains the only Node application in this repository, with its
dependency management in `dashboard/app`.

The embedded showcase was removed from Comptextv7 and is maintained separately
in `ProfRandom92/comptext-v7-monaco-showcase`. Historical Comptextv7 showcase
files are preserved on the `archive/showcase-before-extraction` branch.

Root npm scripts use `npm --prefix` to delegate to the dashboard directory and
use `pytest` for Python validation. No root `node_modules` directory or root npm
dependencies are required for the wrapper itself.
Expand Down Expand Up @@ -67,7 +63,7 @@ npm run smoke:release-health
Use these commands for dashboard TypeScript changes, release-health UI changes,
and `dashboard/app/src/core/foundation/` modules.

## Python validation from the repository root
## Python replay validation from the repository root

Run Python tests from the repository root:

Expand All @@ -80,6 +76,12 @@ pytest tests/test_paper_replay_bench.py tests/test_agent_trace_replay.py tests/t
The focused replay command validates the deterministic paper replay, agent trace
replay, and replay continuity surfaces without changing benchmark logic.

Agent trace replay is fixture-bound: curated traces live in
`tests/fixtures/agent_traces/`, the deterministic runner is
`tests/utils/agent_trace_replay_runner.py`, and committed replay output is stored
in `artifacts/agent_trace_replay_results.json`. Validation is local and does not
use embeddings, vector databases, LLM judges, or external APIs.

Install the Python test dependency set:

```bash
Expand All @@ -93,3 +95,6 @@ python tests/utils/paper_replay_runner.py
python tests/utils/agent_trace_replay_runner.py
python benchmarks/run_replay_continuity.py --iterations 250 --output-dir reports/replay_continuity
```

The KVTC-V7 technical-log compressor in `src/core/kvtc_v7.py` is a deterministic
prototype, not a production-readiness or certification claim.
2 changes: 1 addition & 1 deletion pyproject.toml
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
[project]
name = "comptextv7"
version = "0.1.0"
description = "CompText V7 Cognitive Fabric prototype"
description = "Deterministic replay-validation prototype for compact operational agent/MCP traces"
requires-python = ">=3.11"
license = "Apache-2.0"
license-files = ["LICENSE"]
Expand Down
Loading