"front"	"back"	"tags"
"Aider scores 32/60 — +7 over Pi. Where does the +7 come from?"	"Almost entirely from the GIT-FIRST philosophy. +2 context (repo-map), +2 memory (git commits), +3 state (git rollback), +2 verification (diff review), +1 permission (git-gated), +1 observability (git log). Git is the multiplier."	"harness-eng::dd02::analysis"
"What does 'git-as-substrate' mean for Aider?"	"One infrastructure choice (git) gives 5 rubric modules: state (git reset), memory (commits across sessions), verification (human reviews diff), permission (git-gated), observability (git log). No other harness gets this much from one decision."	"harness-eng::dd02::recall"
"State Aider's permission model and how it differs from Pi's."	"Aider = git-gated (post-hoc review + rollback via git reset). Pi = trust-the-model (no review). Aider trades real-time approval for post-hoc review with native version control. Better for coding (reviewable); doesn't catch malicious — only wrong."	"harness-eng::dd02::analysis"
"What is Aider's diff-based editing, and why is it better than Pi's write_file?"	"Search/replace blocks target specific changes, not whole-file rewrites. More reliable (less token cost, fewer hallucinated-full-file errors). 3 years of iteration have refined this to the field's best."	"harness-eng::dd02::analysis"
"What is the repo-map, and which module does it serve?"	"An AST-based file index in context. The model sees the repo STRUCTURE (symbol graph: classes, functions, signatures) without reading every file. Serves Module 3 (Context Management) — the clean middle path between Pi's nothing and a thick harness's full retrieval."	"harness-eng::dd02::application"
"Why is Aider called 'the benchmark creator'?"	"It created the Aider Polyglot Benchmark — the de facto standard for evaluating coding agents across languages. Most published coding-agent benchmark numbers trace to Aider's methodology. Aider defined how the field measures 'best.'"	"harness-eng::dd02::recall"
"Aider's git-gating catches wrong changes. Does it catch malicious ones?"	"NO — git-gating is a change-QUALITY control, not an injection DEFENSE. A prompt-injected model can commit a benign-looking 'refactor' containing an exfiltration payload. Human diff-review answers 'is this the change I asked for?' not 'is this safe to apply?' Miss → commit → exfiltration succeeds."	"harness-eng::dd02::analysis"
"Name 3 things Aider does better than any other harness in the roster."	"(1) Git-as-substrate (5 modules from 1 choice). (2) Benchmark creation (defined how the field measures). (3) Diff-based editing reliability (search/replace blocks; 3 years refined)."	"harness-eng::dd02::recall"
"Name 3 things to fix if you forked Aider."	"(1) Add Docker sandbox (blast radius = host, same as Pi — git-gating catches wrong not malicious, exfil happens at tool-exec time before diff review). (2) Add untrusted-tagging on file reads (README injection survives diff-review). (3) Add a token budget (long sessions can run away)."	"harness-eng::dd02::application"
"Aider blast radius vs Pi blast radius — same or different?"	"SAME — both have bash with full host access; no sandbox; no scoping. A compromised process can read ~/.ssh and exfiltrate in both. Aider's git-gating catches wrong edits, not malicious exfiltration (exfil happens at tool-exec, before commit review)."	"harness-eng::dd02::analysis"
"State Aider's Architect's Verdict."	"Optimizes for coding reliability via git-as-substrate — state + memory + verification + permission from one infrastructure choice. Sacrifices generality (diff-editing is coding-specific) and blast-radius safety (no sandbox, same as Pi). Build on Aider for any code-editing task where git is available; it is the field's most mature coding agent and the creator of the benchmark everyone else measures against."	"harness-eng::dd02::recall"
"State Aider's MLSecOps Relevance note."	"Aider's git-gating provides change review and rollback but NOT injection defense — a prompt-injected model can commit a benign-looking change that contains an exfiltration payload, passing human diff-review while executing maliciously."	"harness-eng::dd02::recall"
"Aider's tool count (~8) vs Pi's (4). Does Aider violate the 4-tool philosophy?"	"NO — the 4-tool philosophy is a noise-reduction strategy for GENERAL tool-use. Aider's ~8 tools are coding/git-specific and purpose-built (read, write, edit, bash, search, git ops) — they add capability WITHOUT dispatch noise. Evaluate tool count against use case, not against a universal limit."	"harness-eng::dd02::analysis"
"What is the single biggest module gain Aider achieves over Pi? By how many points?"	"Module 8 (State): Pi scores 1, Aider scores 4. +3 — the largest single-module delta in the deep-dive roster. Source: git commits as native rollback. Every change is a recoverable checkpoint."	"harness-eng::dd02::recall"
"Aider's stop conditions vs Pi's stop conditions — compare."	"Both: end_turn + max-iterations + human interrupt (Ctrl+C). NEITHER has a token budget or error threshold. Aider's maturity (3 years) makes the absence of a token budget surprising — long sessions can still run away."	"harness-eng::dd02::analysis"
"Why does Aider score -1 vs Pi on Module 12 (Prompt Assembly)?"	"Aider's prompt is ~2,000 tokens (editing format + git conventions) vs Pi's <1,000. Denser, more capable, but less 'thin.' The cost of specialization — the editing format must be specified. Still thin by industry standards."	"harness-eng::dd02::analysis"
"What is the injection case that defeats Aider's git-gating? Walk through it."	"Model reads a README containing ASI01 payload → prompt-injected → produces a benign-looking 'refactor' that also reads ~/.ssh/cred and curls to attacker → human reviews diff, sees a legitimate refactor, PASSES it (reviewing for 'is this what I asked for,' not 'is this safe') → commit executes → exfiltration succeeds. Git-gating is downstream of the damage."	"harness-eng::dd02::analysis"
"Which 5 rubric modules does git-as-substrate yield? Name them."	"M4 Memory (commits = checkpoints), M8 State (git reset = rollback), M9 Verification (diff IS the verification), M6 Permission (git-gated review), M11 Observability (git log = trace). Plus M3 Context via the repo-map (AST index, not from git directly)."	"harness-eng::dd02::recall"
"Aider's LOC vs Pi's LOC. What does the delta tell you?"	"Aider ~25,000 LOC vs Pi ~1,200 LOC — ~20x. The delta is the cost of git-integration, diff-editing machinery, repo-map AST parsing, and 3 years of accumulated features. Still readable; no longer 'read in an afternoon.'"	"harness-eng::dd02::recall"
"Construct the 'malicious-but-benign-looking' diff that defeats git-gating. What does it contain?"	"A legitimate refactor (rename function, move import, add helper) + a buried exfil payload (read ~/.ssh/cred, curl to attacker). The refactor is real and correct; the payload hides inside it. Human diff-review sees a valid refactor and approves — the exfil executes at commit time."	"harness-eng::dd02::application"
"Why is Aider the field's reference for coding-agent performance, even though thicker harnesses outscore it?"	"Because Aider created the benchmark. The Aider Polyglot Benchmark is the de facto standard — every coding-agent number traces to it. Aider defined how the field measures 'best.' Being the reference is about measurement authority, not raw rubric score."	"harness-eng::dd02::analysis"
"What does 'the diff IS the verification' mean for Module 9?"	"Aider commits after each change. The human reviews `git diff` — the change itself becomes the quality gate. Implicit verification: if the diff looks right, the change is accepted. No separate test-pass requirement (that would be a thicker harness's approach). Clever coding-specific pattern; not a general verification framework."	"harness-eng::dd02::application"
"Aider has no subagents (M10). Is this a defect or a decision?"	"A DECISION consistent with the git-first philosophy. Aider is a single-agent pair-programmer — one model, one git history, one diff-reviewer. Subagents would multiply commit noise and complicate the diff-review workflow. The absence is coherent with the design."	"harness-eng::dd02::analysis"
"Compare Aider's context management (M3) to Pi's."	"Pi: none — relies on minimal tool output, no compaction. Score 2. Aider: repo-map (AST-based symbol graph) gives structure without reading every file. Score 4 (+2). The repo-map is the clean middle path between nothing and full vector retrieval."	"harness-eng::dd02::analysis"