Skip to content

Git designed for AI agents.

graft: the recommended auto-landing gate (two vendors agree, clean reports, verify passes) in RESOLVE.md and PLAN10 15.13

Federico Eliab5b3a2fd056a
diff --git a/doc/graft/PLAN10.md b/doc/graft/PLAN10.md
index 3450b9f..0f24aea 100644
--- a/doc/graft/PLAN10.md
+++ b/doc/graft/PLAN10.md
@@ -1809,6 +1809,14 @@ verification: `verify.enabled` false, `verify.command`, `verify.timeout` 5 min,
`verify.max_tree_bytes` 2 GiB, `verify.max_tree_depth` 64. Nothing lands from a model unless `verify.enabled` is set
and the command passes on the exact tree.
+**The recommended gate for an operator who wants auto-landing.** G1 of 15.12, as a recipe and not as server code:
+two models from different vendors give the byte-identical resolution, both composition reports are clean, and
+`verify.command` passes on the exact tree (`verify.enabled` is mandatory with it, and the wrapper is a `command`
+strategy with `origin = "model"`; RESOLVE.md has the recipe). Measured on the OSS held-out files it lands 59.8% of the
+files with 15.5% of those wrong under two independent judges (upper bound 20.1%), before the operator's tests; on conflicts
+of agent PRs 45.8% of the files with 43.1% wrong. The 20%-wrong bar it is compared with was set after the numbers were
+known (15.12).
+
**The default model.** The shipped example is `openai/gpt-6-luna` at effort low. On the stratified 100 the one other
model that passed the smoke check, `z-ai/glm-5.3-flash` pinned to Z.AI, is as good (15.12) and costs the same per
file; the models that reason without a bound on their effort ended their calls past `call_timeout` or `max_tokens`
diff --git a/doc/graft/RESOLVE.md b/doc/graft/RESOLVE.md
index 45de4c7..561d443 100644
--- a/doc/graft/RESOLVE.md
+++ b/doc/graft/RESOLVE.md
@@ -691,6 +691,27 @@ origin = "model" # its files are proposals, and land only if verify.command pa
`origin` is `"rule"` (the default) or `"model"`. A model command's answers are
not recorded by `rerere`.
+**The recommended gate for auto-landing a model's resolution.** The `llm` strategy alone produces a proposal that lands
+only if `verify.command` passes on the exact tree. An operator who wants model resolutions to land without an agent in
+the loop gets the best measured result (PLAN10 15.12) from this gate, which the server does not implement and an
+operator builds with the `command` strategy above:
+
+1. **Two models from different vendors resolve the file** (two `graft-resolve` configurations, each with one `llm`
+ strategy and its own provider), and the resolution is used only if their results are byte-identical after
+ normalizing line endings and trailing whitespace;
+2. **the composition report of both is clean** (no line saying that the answer drops or duplicates a change of a side);
+3. **`verify.command` passes** on the exact tree: `verify.enabled = true` is mandatory with this gate, and the strategy
+ must be `origin = "model"` as above, so that the server itself refuses to land it otherwise.
+
+The wrapper (the program of the `command` strategy) writes `{out}` only when 1 and 2 hold and fails otherwise; the chain
+then goes on, and the agent gets the conflict package with the proposals as usual. The product ships no such wrapper.
+What it was worth, measured on 487 held-out files of 13 open-source repositories (Luna and glm-5.3-flash, with the
+verify simulation of the bench in place of the operator's tests): 59.8% of the files landed, and 15.5% of those differed
+from what two independent reviewing models accept (11.3% for one of them; Wilson upper bound 20.1% for both), against 43%
+on conflicts of merged agent PRs. These figures are before the operator's own tests, which are the one signal that
+caught more errors where we could run them; the gate does not replace `verify.command`, it only chooses which proposals
+are worth verifying. Cost: two model calls per region.
+
**The `rerere` store and the upgrade.** Entries are versioned and record their
origin. The stores of v0.1.0 and v0.2.0 recorded the `llm` answers as well, with
no origin: after an upgrade `graft-resolve` never replays an entry without