The Corpus

The asset that makes certification meaningful — 7 real, captured-style agent trajectories across 7 domains. Every strategy must pass the gate on every one of them.

Why a corpus matters

A compression strategy is not trustworthy because it works on one example. It needs to hold across the full breadth of what agents actually do: incident response, code debugging, customer support, research synthesis, data analysis, infrastructure work, and financial reconciliation.

Domain table

Numbers are reproducible from the bundled corpus using the heuristic tokenizer. Absolute dollar figures assume claude-opus-4-8 public list pricing — verify before billing use.

Trajectory structure

The bundled trajectories are 4-turn headless agent loops; the structural minimum enforced by corpus.validate() is ≥3 turns. Three structural components:

• Cacheable stable prefix — system instructions, tool definitions, and other content that does not change between turns. This is what the cache-aware engine keeps byte-stable to maximize cache hits.
• Decision-driving volatile tool outputs — the tool results that carry the information the agent actually needs to make its next decision. These are the blocks that show up as keep in ablation.
• Causally-inert noise — speculative retrievals, stale observations, context blocks that were present but never cited by any decision. These are the blocks ablation marks as PRUNE.

Trajectories are stored as JSON files in corpus/. The manifest is at corpus/manifest.json. The loader (distil/corpus.py) resolves the corpus path from the wheel, the repo, or a $DISTIL_CORPUS environment variable.

distil bench — the CI gate

distil bench is the intended CI gate for new compression strategies. It runs the full corpus measurement loop and exits non-zero if any of the following occur:

• A trajectory fails structural validation (corpus.validate())
• The distil strategy fails the TOST non-inferiority test on any trajectory
• The gate fails to reject the aggressive lossy strategy on any trajectory (which would indicate the gate itself is broken)

$ distil bench
# Exits 0 only if every trajectory passes all three conditions above

GATE: PASS — every trajectory certified non-inferior; aggressive rejected on all.

# If a new strategy were added that degraded quality on one domain:

GATE: FAIL (1 issue(s))
  - corpus/finance-reconcile.json: distil FAILED non-inferiority

The rule in one sentence

Corpus invariants enforced by corpus.validate()

The validate() function checks structural correctness of each trajectory before it enters the gate:

• Invariant 1: ≥3 turns, and the trajectory's model must be in the pricing catalog.
• Invariant 2: Cacheable-prefix ordering — no volatile block may precede a non-volatile block within any turn.
• Invariant 3: STABLE blocks are byte-identical across every turn they appear in (breaks the cache if violated).
• Invariant 4: At least one STABLE block carries a DECISION: marker (so ablation keeps the prefix).
• Invariant 5: At least one VOLATILE, decision-relevant tool output carries a decision; non-decision-relevant volatile blocks must NOT contain a decision marker.
• Invariant 6: At least one causally-inert "noise" block — a VOLATILE, non-decision-relevant block with no decision marker — for ablation to prune.

A trajectory that fails validation is reported as a failure in bench before any cost or quality measurement runs — there is no silent skip.