"What is OpenHarness's (HKUDS) defining objective function?"	Inspectability under reproducibility. The diagnostic test: can a researcher point at the exact mechanism that produced this decision, vary it, and measure the effect? NOT capability under constraints (the production objective).	harness-engineering::dd13::recall
"What is OpenHarness's origin and category?"	HKUDS lab (academic), April 2026. The academic harness — built for research reproducibility and inspectability, NOT for production shipping. The clean-room comparison baseline.	harness-engineering::dd13::recall
"What three properties define OpenHarness's 'academic' posture?"	(1) Inspectability — every component modular and individually examinable. (2) Auto-compaction visibility — first-class named parameterized component. (3) Transparent decisions — logged at source with inputs, not reconstructed after the fact.	harness-engineering::dd13::recall
"What is ohmo?"	OpenHarness's built-in personal agent. The reference demonstration that exercises every component so a researcher can vary a parameter and measure the effect. ohmo is to OpenHarness what Pi (DD-01) is to the minimal-harness category.	harness-engineering::dd13::recall
"Inspectability-as-product vs observability-as-feature — what is the distinction?"	Production harnesses emit rich telemetry about a black box (observability-as-feature). OpenHarness makes the box itself open — every component examinable in isolation (inspectability-as-product). Related but distinct properties.	harness-engineering::dd13::analysis
"Name the four parts of OpenHarness's auto-compaction mechanism."	(1) Trigger predicate — declared condition deciding WHEN. (2) Selection policy — what summarized, preserved, dropped. (3) Summarization step — model-based, inputs and outputs logged. (4) Non-destructive record — original retained for analysis.	harness-engineering::dd13::recall
"Why is the non-destructive record (Part 4) the load-bearing detail?"	The original context survives compaction, so a researcher can ask 'did compaction change the outcome?' — a question a production harness cannot answer without instrumentation. Mirrors DD-21's replaces_entry_ids and Module 8's event-sourcing.	harness-engineering::dd13::analysis
"Why does Module 3 cite OpenHarness as the compaction reference?"	Because the four-part mechanism is STUDIABLE rather than merely present. In production harnesses compaction is entangled with session store, token accounting, retry logic. In OpenHarness each part is individually examinable.	harness-engineering::dd13::analysis
"What are OpenHarness's three explicit memory tiers?"	(1) Working memory — current context window post-compaction. (2) Episodic memory — prior turns/sessions, retrievable not in-context. (3) Semantic memory — distilled facts keyed for retrieval.	harness-engineering::dd13::recall
"What architectural choice distinguishes OpenHarness's memory from production harnesses?"	Separate stores with declared interfaces, NOT layers of the same buffer. This buys individually-auditable read and write paths (Module 4.3 write-gating). The cost is integration friction — production conflates the tiers for speed.	harness-engineering::dd13::analysis
"Why is OpenHarness's tool count research-minimal?"	A large tool set is noise — it confounds 'what did the model do?' with 'what did the tools enable?' OpenHarness keeps the surface small so the model's decisions are the signal. Opposite of a production harness's incentive.	harness-engineering::dd13::analysis
"OpenHarness scores 30/60. Why is it still valuable?"	The score reflects what OpenHarness is NOT (production-ready), not a failure of what it IS (the cleanest mechanism reference in the roster). Same pattern as DD-01 Pi — low absolute score reflecting 'not trying to be production.' Value is pedagogical and methodological.	harness-engineering::dd13::analysis
"On which modules does OpenHarness score highest?"	Module 3 Context (4/5) — auto-compaction reference. Module 4 Memory (4/5) — explicit tiered stores. Module 10 Observability (4/5) — inspectability is the product (below 5 only because human-readable over machine-readable).	harness-engineering::dd13::recall
"On which modules does OpenHarness score lowest, and why?"	Module 5 Sandbox (1/5), Module 6 Permission (2/5), Module 9 Verification (1/5), Module 11 Security (1/5). These are deliberately minimal or absent because they are NOT the research question.	harness-engineering::dd13::recall
"What is the OpenHarness-vs-Tau (DD-21) inspectability axis?"	OpenHarness = legibility-of-BEHAVIOR (what the harness did). Tau = legibility-of-CODE (what the harness is). Both inspectability-first, optimizing for complementary properties. Shared design: the non-destructive record.	harness-engineering::dd13::analysis
"What does 'we can defend why the run went the way it did' mean, and why is it stronger than 'we can reproduce the run'?"	OpenHarness logs every decision at source with inputs (first-class record). 'Reproduce the run' is after-the-fact trace reconstruction — weaker. 'Defend the run' means you can point at the decision and the inputs that produced it.	harness-engineering::dd13::analysis
"What are the three things OpenHarness does better than other harnesses?"	(1) Compaction as a designed four-part mechanism (Module 3 reference). (2) Inspectability-as-product — every decision logged at source with inputs. (3) Non-destructive context record — original survives compaction (mirrors DD-21, Module 8).	harness-engineering::dd13::recall
"What are three things to fix if you misuse OpenHarness as production?"	(1) Add a sandbox — bash/exec run without containment. (2) Add a permission model — below floor. (3) Emit machine-readable decision logs — currently human-readable; a structured schema would meet the DD-14 observability standard.	harness-engineering::dd13::recall
"For MLSecOps, what two questions can only OpenHarness answer?"	(1) Compaction-induced drift — does the summarizer drop a security-critical instruction? (2) Memory-tier poisoning — can a write to semantic memory survive compaction? Both require visible mechanisms that production harnesses do not expose.	harness-engineering::dd13::analysis
"Why is inspectability NOT a security property?"	A visible compaction mechanism is not a safe compaction mechanism. Inspectability means 'I can see what happens,' not 'what happens is safe.' OpenHarness has no sandbox, no permission model, no security model. Read it as a research property.	harness-engineering::dd13::analysis
"What gap would OpenHarness need to close to hit 5/5 on Module 10 Observability?"	Emit machine-readable decision logs. Currently the inspectability is human-readable. A structured schema would make it analyzable at scale and bring it to the DD-21/DD-14 event standard.	harness-engineering::dd13::analysis
"What is the shared design between OpenHarness and DD-21 (Tau)?"	The non-destructive record. Both refuse to overwrite the on-disk truth. OpenHarness retains the original context; Tau retains the CompactionEntry replacement chain. The event-sourcing principle (Module 8) as a convergent design.	harness-engineering::dd13::recall