docs(wiki): add topic pages to the project wiki - #27
Merged
Conversation
…ails, Experiment-Harness Results-Synthesis: one table of every headline number (pre-registration vs. observed, Newcombe CIs) synthesized across 5 vendored figure JSONs, D12–D24, PROJECT.md, and the capstone-data.json ladder. Methodology-Guardrails: the honesty machinery synthesized across DECISIONS.md (D2/D3/D7/D12/D13/D15/D16/D19/D22), CLAUDE.md, and stats.py — oracle design, CI method, seed-distinctness rule, kill-triggers, what counted as a null, what would have falsified the thesis. Experiment-Harness: how the code runs an experiment — toggle system, arm-as-config pattern, fault injectors, result paths, all reproduce commands — mined from code (agent.py, ablation.py, faults.py, chart.py) and DECISIONS.md (D11/D14–D19/D23). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What each page synthesizes and why it clears the no-restatement bar
Results-Synthesis
Synthesizes across 6 sources (5 vendored figure JSONs + D12–D24 + PROJECT.md headline table + capstone-data.json): every headline number from all five measured gaps in one table, pre-registered prediction vs. observed outcome, the injected-vs-natural distinction explained for the gap-closure figure, the full capability ladder with cross-run disclosure, and the S10 residual decomposition. No single file has all of this; the closest thing (README §12) summarises without the per-stage CIs or the residual breakdown.
Methodology-Guardrails
Synthesizes across 4 sources (DECISIONS.md D2/D3/D5/D7/D12/D13/D15/D16/D19/D22/D23, CLAUDE.md methodology section, PROJECT.md boundaries, docs/LEARNING.md CI teaching notes): the oracle design and why it differs structurally from the validator, Wilson/Newcombe/
excludes_zeroas a mechanical pre-committed gate, the "N = distinct seeds" insight (not documented outside D13), kill-triggers as pre-registration, what the S6 null actually proves about guardrail specificity boundaries, and what would have falsified the thesis. No single doc covers all of this together.Experiment-Harness
Synthesizes across code + 5 decision entries (agent.py toggle signatures, ablation.py arm configs, faults.py injectors, chart.py vendoring, DECISIONS.md D11/D14/D15/D16/D18/D19/D23): the toggle system with cost-per-arm (model turn vs. no turn), arm-as-config pattern, fault-classification boundary in
_is_retryable, where results land (gitignored runs/ vs. committed docs/figures/), all reproduce commands, and the S10 mid-experiment harness fix. The CLAUDE.md file-map has per-file descriptions but not the synthesized view of how they compose into an experiment.🤖 Generated with Claude Code