Module DD-13 — OpenHarness: Academic Baseline (Inspectability as Product)

OpenHarness: Academic Baseline (Inspectability as Product)

HKUDS lab, April 2026. Research reproducibility, not production. Auto-compaction as a designed four-part mechanism. The harness Module 3 points at when it teaches compaction. 30/60 — the score reflects what it is not (production), not a failure of what it is (the cleanest mechanism reference in the roster).

minutes

artifacts

sub-sections

OpenHarness optimizes for inspectability under reproducibility — can a researcher point at the exact mechanism that produced this decision, vary it, and measure the effect? A different objective function than production harnesses, not a weaker version of the same one. Every component is modular and individually examinable, every decision is logged at source with the inputs that produced it, and the auto-compaction mechanism is a designed four-part mechanism (trigger predicate, selection policy, summarization step, non-destructive record) rather than an emergent recovery behavior. The closest peer on inspectability is DD-21 (Tau): OpenHarness optimizes legibility-of-behavior, Tau optimizes legibility-of-code. The shared design is the non-destructive record.

Key Claims

Load-Bearing Claims

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): a low absolute score that reflects 'not trying to be production.' The rubric scores production-readiness across all eleven modules; OpenHarness optimizes for inspectability, not capability. Its value is pedagogical and methodological — the harness that makes every other harness's claims falsifiable.

Auto-compaction is a designed four-part mechanism — the Module 3 reference. Trigger predicate (declared condition for when), selection policy (what summarized/preserved/dropped), summarization step (model-based, inputs and outputs logged), non-destructive record (original retained for analysis). The mechanism is studiable rather than merely present. Production harnesses fold all four parts into an emergent recovery behavior entangled with session store, token accounting, and retry logic.

The non-destructive record is the load-bearing detail — it makes falsifiable research claims possible. 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 CompactionEntry with replaces_entry_ids and Module 8's event-sourcing principle: the on-disk truth is never rewritten, only the in-context view is swapped.

Explicit tiered memory (working, episodic, semantic) as separate stores — not layers of the same buffer. Buys individually-auditable read path and write path (Module 4.3 write-gating). The cost is integration friction; production harnesses conflate the tiers because doing so is faster and cheaper. OpenHarness pays the friction to buy the inspectability.

After This Module

State OpenHarness's defining objective function — inspectability under reproducibility — and explain why it optimizes for a different goal than production harnesses (capability under constraints).

Describe the auto-compaction implementation as a designed four-part mechanism (trigger predicate, selection policy, summarization step, non-destructive record) and explain why Module 3 cites it as the reference.

Distinguish inspectability-as-product (OpenHarness — the box itself is open) from observability-as-feature (production harnesses — rich telemetry about a black box).

Score OpenHarness 30/60 and defend the interpretation: the score reflects what it is not (production-ready), not a failure of what it is (the cleanest mechanism reference in the roster).

Articulate the OpenHarness-vs-Tau (DD-21) inspectability axis — legibility-of-behavior vs legibility-of-code — and the shared non-destructive-record design.

Artifacts

Teaching Document

Teaching document — the academic thesis (inspectability-as-product, not production), the three academic properties, the architecture (swappable modular components), the auto-compaction four-part mechanism (Module 3 reference), explicit tiered memory, ohmo the reference agent, the 30/60 score profile with interpretation, the OpenHarness-vs-Tau axis, anti-patterns, key terms, references

READ

Diagrams

7 Mermaid/n8n diagrams — inspectability as design goal, swappable modular architecture, the four-part compaction mechanism, explicit tiered memory, the OpenHarness-vs-Tau inspectability axis, the production gap, the n8n compaction workflow

READ

Slide Deck

11 slides — reveal.js, dark theme, design-system teal; covers the inspectability-vs-capability thesis, three academic properties, swappable architecture, the four-part mechanism, tiered memory, the 30/60 score interpretation, the OpenHarness-vs-Tau axis, MLSecOps relevance, three better/three to fix

READ

Teaching Script

Verbatim teaching transcript with [SLIDE N] cues, ~2,000 words spoken across 11 slide cues