Five improvements that eliminate most of the agent-in-the-loop friction
observed while closing out the 988-entity WoN evaluation (C.1):
1. Gemini adapter now retries on 429 + 5xx with exponential backoff
(same pattern already used by OpenRouter/OpenAI). Removes the need
for shell-level retry wrappers when hitting free-tier rate limits.
2. evaluate CLI prints the underlying error ("ERROR — HTTP 503 …")
instead of a bare "ERROR", so agents don't have to drop into Python
to diagnose transient failures.
3. --entity/--chapter now respect existing evaluation files by default
(previously only the full-collection pass did). New --force flag
opts into re-evaluation. Stops silently burning free-tier quota on
re-runs of the same slug.
4. --entity accepts hyphenated slugs (matching entity filenames) and
normalizes them to the underscore form used on disk. On a miss the
CLI suggests near matches instead of a bare "not found".
5. eval-summary --update-metrics is no longer destructive:
read_metrics_file/write_metrics_file preserve structured values
(type_distribution) and don't flatten ints to floats. Fixes a
silent data loss observed on every run.
Bonus: the evaluator field in written evaluation frontmatter now
falls back from run_config.model_name to the adapter's resolved model
(or the model echoed back in the API response), so rows no longer
show `evaluator: null` when --model is omitted.
Tests: new tests/unit/llm/test_gemini.py covers retry behavior;
tests/unit/infospace/test_history.py gains a round-trip test that
pins the type_distribution / int-preservation invariants.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Stage 1 — Decouple:
- Move RunConfig + LLMResponse to markitect/llm/models.py (canonical)
- Move LLMAdapter + Mock/ErrorLLMAdapter to markitect/llm/adapter.py
- markitect/prompts/execution/models.py and llm_adapter.py become re-export shims
- All 4 adapters + factory.py updated to import from markitect.llm.*
- Parameterize app_name in toml_config.py (resolve_llm, get_default_layers,
get_preference_layers): paths and env var now derived from app_name arg
- Add tests/test_llm_isolation.py: 7 isolation + backward-compat tests
Stage 2 — Extract:
- Standalone llm-connect package created at ~/llm-connect/
- All 18 llm files copied; markitect.* imports replaced with llm_connect.*
- LLMError base inlined in llm_connect/exceptions.py (no markitect dep)
- llm-connect installed into markitect-venv; declared in pyproject.toml
Smoke test: markitect llm-check succeeds (live Gemini API call).
Backward compat: markitect.prompts.execution.{models,llm_adapter} still work.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- SourcePipeline: retry split_entities stage once when 0 entity delimiters
are found (free-tier models intermittently return short non-formatted
responses); save raw LLM response to <stage>-raw.md alongside prompts
- Return None (pause pipeline) rather than writing empty view file when
no entities found after max retries
- _http.py: wrap json.JSONDecodeError in LLMAPIError with body preview
- extract-entities.md: add explicit H2-heading format example to Output
Format section to prevent models from using inline "Section:" format
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add OpenAI-compatible embedding support (works with both OpenAI and
OpenRouter), file-based embedding cache with content-digest invalidation,
and pure-Python cosine similarity utilities for downstream redundancy
detection.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add TOML-based config resolution with 7-level priority chain:
CLI flags > env var > user preference > directory preference >
directory default > user default > hardcoded fallback.
New commands: llm-default (view/set/clear defaults), llm-preference
(view/set/clear preferences). Each shows only its own scope. llm-check
now displays source attribution for resolved provider/model.
Existing commands (llm-helper, llm-check) refactored to use
resolve_llm() instead of manual resolution. Hardcoded fallback
changed from openrouter/aurora-alpha to gemini/gemini-2.5-flash
due to persistent OpenRouter 502 errors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add OpenAIAdapter for the OpenAI chat completions API (apikey-chatgpt.txt
or OPENAI_API_KEY). Set default model to arcee-ai/trinity-large-preview:free
for the infospace pipeline and increase max_tokens from 4096 to 8192.
Reprocess chapter 05 with Trinity Large (was Gemini: 1 truncated entity,
now 19 complete entities). Process chapters 06 (Aurora Alpha, 10 entities)
and 07 (Trinity Large, 15 entities including regenerated violent-policy.md).
Canonical set now at 85 unique entities.
Add entity archive policy: entities are never silently deleted. Retired
entities move to output/entities/archive/ with a dated reason header.
New CLI option: --archive-entity <slug> --reason "...". The --list
output shows the archive count alongside the canonical set.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add GeminiAdapter calling Google's Generative Language REST API
(default model: gemini-2.5-flash). Register "gemini" as third
provider in the factory and CLI. Add rate-limit retry with
exponential backoff to the pipeline's _call_llm helper. Increase
default max_tokens from 2000 to 4096.
Process book-1-chapter-05 via Gemini free tier — 1 new entity
extracted (necessaries-conveniencies-and-amusements-of-life),
41 existing entities correctly skipped by dedup. Canonical set
now at 42 unique entities.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements markitect/llm/ package with concrete LLMAdapter implementations:
- OpenRouterAdapter: HTTP via urllib with retry/backoff on 429/5xx
- ClaudeCodeAdapter: subprocess-based Claude CLI with stdin piping
- Factory pattern: create_adapter("openrouter") or create_adapter("claude-code")
- API key resolution chain: constructor > env var > project-root key file
- 42 unit tests, 2 integration tests (gated on API key / CLI availability)
Also adds the infospace-with-history example with Wealth of Nations VSM
analysis pipeline, templates, schemas, source chapters, and processed
output for chapters 1-2. process_chapters.py now supports --provider
and --model flags for automatic LLM-driven processing.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
