opencode-harness-kit 0.4.3 → 0.4.5

package/docs/KNOWN-LIMITS.md

@@ -96,8 +96,8 @@ a client cannot currently tell a short quota window from a permanent block.
 
 ## A copied plugin loads and does nothing
 
-The bundle finds `profiles/`, `prompts/`, `retention.yml` and `lanes.json` by
-walking up from its own file. A symlink resolves to the real path, so a
+The bundle finds `profiles/` and `lanes.json` by walking up from its own file,
+and `retention.yml` next to them. A symlink resolves to the real path, so a
 symlinked plugin finds the kit's data. A **copy** does not: it looks beside
 itself, finds nothing, and every profile lookup fails.
 
@@ -284,23 +284,47 @@ unhealthy and failing over at the next start is the honest behaviour.
 ## Free models get withdrawn quickly
 
 From one install's 20-day history, free model slots turned over in days:
-
-| model | observed lifetime |
-|---|---|
-| `mimo-v2.5-free` | 1 day |
-| `big-pickle` | 1 day |
-| `nemotron-3-ultra-free` | 3 days |
-| `muse-spark-1.3-contributor-free` | still running at 20 days |
-
-The tier turned over again since: four of this kit's five original profiles
-(`space-bunny-free`, `x-preview-f-free`, `deepseek-v4-flash-free`,
-`muse-spark-1.3-contributor-free`) were all withdrawn within a month of
-shipping, and the default profile pointed at a model that no longer existed.
-
-Zen states free models are available "for a limited time" with no date. Treat
-profiles as perishable: run `oc-harness doctor`, which re-checks each profile
-against live metadata and reports drift, and which also reports a lane whose
-whole chain has left the catalogue.
+`mimo-v2.5-free` lasted a day, `big-pickle` a day, `nemotron-3-ultra-free`
+three days. Zen states free models are available "for a limited time" with no
+date. Treat profiles as perishable.
+
+The README's [Free models are perishable](../README.md#free-models-are-perishable-so-check-them)
+section has the full list of withdrawn models, the four commands that check
+for drift, and what `refresh` will and will not rewrite. The short version:
+run `oc-harness doctor` before every release and whenever a lane starts
+failing with a 429 or a withdrawal.
+
+The tier turned over again since. Every one of this kit's original
+profiles has been withdrawn — `space-bunny-free`, `x-preview-f-free`,
+`deepseek-v4-flash-free`, `muse-spark-1.3-contributor-free` and
+`mimo-v2.5-free` — and the default profile pointed at a model that no longer
+existed.
+
+## The install-wide measurement is an observation, and there is a controlled one
+
+`npm run measure` compares the work before the kit landed with the work
+after. It reported the context per request going **up** by 16% on one install,
+with twice as many reads and twice as many messages in the second window.
+
+That is a number about the jobs, not about the kit. The two windows are never
+the same work, and "everything since the install" is a window that grows
+without bound, so the comparison is a different one every time the script
+runs.
+
+`npm run measure` now says so: the windows are fixed and equal around the
+split, the context drop is skipped with a note when the two sides are not
+comparable, and the gate uses the number that IS comparable.
+
+`eval/measure-prune.mjs` is that number. opencode stores the message parts
+unpruned - the pruner rewrites the outgoing request, not the history - so the
+stored parts of a real session are exactly what the pruner was given, and
+replaying the real pruner over them measures exactly what it removed. On one
+install: **80.5% of the context
+across 60 real sessions**, 6,608 parts stubbed - the number
+`npm run measure:prune` prints.
+
+The controlled comparison of the two arms is `eval/ab.mjs`, and its result is
+in `eval/RESULTS.md`.
 
 ## A model-driven compressor is not assumed to work
 
@@ -322,7 +346,7 @@ truncation directory, so re-reading is possible. But "re-run the command to see
 it again" is more expensive than "scroll up". The stub is one line precisely so
 an agent can decide whether it is worth it, and the stub says so.
 
-A `savedBytes` of 0 in `prune.jsonl` is not a second, redundant pass. It means
+A `savedChars` of 0 in `prune.jsonl` is not a second, redundant pass. It means
 `bytesAfter` did not come out below `bytesBefore` for that turn, which happens
 when a stub is no shorter than the output it replaced (the stub's own label
 costs more than a short body), or when a `reasoning` part was under the 200
@@ -346,7 +370,7 @@ compressor, or a session UI. Those exist and are good:
 |---|---|
 | model-driven compression | `@tarquinen/opencode-dcp` |
 | cross-harness context proxy | `sleev.ai` |
-| persistent memory | `opencode-supermemory` |
+| persistent memory | [`opencode-mem`](https://github.com/tickernelz/opencode-mem), which `oc-harness use --memory` turns on |
 | git worktrees | `kdcokenny/opencode-worktree` |
 | memory in git markdown | `code-yeongyu/oh-my-opencode` (now `omo`) |
 | agent DAGs | `spoons-and-mirrors/subtask2` |

package/docs/RESEARCH.md

@@ -47,13 +47,18 @@ ceiling that a normal session never reaches.
 
 5,239 historical `bash` commands were replayed against the allowlist shapes:
 
-| allowlist | prompts | share |
+| allowlist | prompts | share of 5,239 |
 |---|---|---|
 | the narrow one in use | 4,994 | 95.3% |
 | this kit's, first version | 1,204 | 23.0% |
-| this kit's, current | 1,002 | 18.5% |
+| this kit's, current | 1,002 | 19.1% |
 
-The remaining 18.5% is the deliberately gated set, ranked by how often it
+The three rows share the denominator, and the arithmetic is theirs to be
+checked: 4,994/5,239 is 95.3%, 1,204/5,239 is 23.0%, and 1,002/5,239 is
+19.1%. The last figure was quoted as 18.5% in two places, which is a
+different number for the same count.
+
+The remaining 19.1% is the deliberately gated set, ranked by how often it
 actually came up:
 
 ```
@@ -111,7 +116,7 @@ is what actually bounds context.
 `bin/baseline.mjs` now prints all five numbers so this is re-measurable instead
 of being re-argued.
 
-**"Compaction never ran, therefore nothing is wrong."** Zero events across 156
+**"Compaction never ran, therefore nothing is wrong."** Zero events across 153
 sessions was read as "the setting is fine" rather than "nothing is being
 bounded". Both readings were half right and the compound conclusion was wrong:
 the configured `tail_turns` was inert *and* there was no bound on growth.
@@ -188,7 +193,14 @@ out.** Re-measured on 2026-10-20, lane by lane, over live calls: GLM is
 4.4-5.8 seconds and not 16-27 (the old figure came from a much larger
 prompt); MiMo reads an image in 19.2 seconds and not 8; and Nemotron Super VL
 does **not** hang on a tool read, only on a file attached to the prompt, which
-means the sweep lane is sound. The measurement also showed the safe fleet
+means the sweep lane is sound.
+
+Those were the install's numbers, measured inside a working session. A
+DIRECT measurement on 2026-10-21 - one line, the same prompt for every
+model, from the process start - puts GLM at 7.4 seconds and Beam at 6.1, so
+the two are not comparable and the routing page now says which is which.
+The image figures are still the install's: a one-line prompt is not an
+image. The measurement also showed the safe fleet
 giving up more than it needed to: two lanes were on a model the ceiling had
 forced them onto rather than the best one available for that job. The numbers,
 and what changed because of them, are in [`ROUTING.md`](ROUTING.md).
@@ -283,3 +295,6 @@ and quietly replacing it.
   session
 - The live free tier, measured 2026-10-20: one request per lane on both
   fleets, ten lanes on each, through real delegations
+- The live free tier, measured 2026-10-21, directly: `node eval/models.mjs
+  --reps 5 --effort` - the same one-line prompt to every model the profiles
+  use, from the process start, with the `low` variant as well

package/docs/ROUTING.md

@@ -4,9 +4,26 @@ The fleet is the point of this kit. One session, ten lanes, ten models, and a
 human who never switches a model by hand.
 
 This page records what the lanes are, which model serves each one, and the
-measurements behind those choices. Everything in the tables below was measured
-against the live tier on **2026-10-20** with OpenCode **1.19.0**, and is
-reproducible with `oc-harness doctor` and `oc-harness lanes`.
+measurements behind those choices.
+
+There are **two** sets of numbers here, and they measure different things:
+
+- The **install tables** are the telemetry of a real install: `oc-harness
+  lanes` over the requests a working session actually made, on
+  **2026-10-20**. They are the wait a person really had, and the `build` row
+  is not comparable with the others, because it counts the turns that did the
+  delegating.
+- The **direct table** is `node eval/models.mjs` - one line, one model, five
+  requests, whole-process wall time, on **2026-10-21**. It is the same
+  prompt for every model, so it compares models with each other, and it
+  includes the process start, which the install tables do not.
+
+Neither is the answer on its own. The install tables say what happened; the
+direct table says what a lane would wait for if the lane were all there was.
+
+Both are reproducible: `oc-harness doctor` for the catalogue, `oc-harness
+lanes` for the install, and `node eval/models.mjs --reps 5 --effort` for the
+direct table.
 
 ## The lanes
 
@@ -23,7 +40,7 @@ models. A profile says which lanes it enables and which model serves each one.
 | `vision` | read an image or a screenshot | mimo-v2.6-flash-free | minimax-m3.1-flash-free |
 | `doc` | read a long document or a spec | glm-5.3-flash-free | glm-5.3-flash-free |
 | `schema` | produce JSON or a JSON schema | deepseek-v4.1-flash-free | deepseek-v4.1-flash-free |
-| `sweep` | sweep a lot of code mechanically | nemotron-3.5-super-vl-free | longcat-2.5-preview-free |
+| `sweep` | sweep a lot of code mechanically | nemotron-3.5-super-vl-free | deepseek-v4.1-flash-free |
 | `longform` | read a megabyte, or write alternatives | longcat-2.5-preview-free | longcat-2.5-preview-free |
 
 `fleet-all` gives every lane the model built for it: ten jobs, ten models,
@@ -35,6 +52,17 @@ ceiling, and it is worth paying only when the model that takes over is the
 best one available *for that job*, which is why the two fleets order their
 chains differently (see below).
 
+The `sweep` lane is the one that was worth changing rather than accepting.
+A bulk mechanical search is the one job where reasoning is pure waste, and
+the model it ran on reasoned 420 tokens a message at 16.6 seconds - the
+most expensive way to ask a model to grep. It runs on DeepSeek now, which is
+zero-retention, publishes a `low` level, and measures 19 reasoning tokens at
+3.4 seconds.
+
+Under that ceiling every lane that asks for a tier GETS one. The safe fleet
+has no lane with a dropped tier request; the unrestricted fleet has three, and
+the measurement says they are the three cheapest reasoners on the tier.
+
 ## Measured, per lane
 
 Every number here is a live call, not a catalog reading. A **request** is one
@@ -77,11 +105,88 @@ that did the delegating, so they count the work of the other nine lanes. A
 one-line `build` reply on its own measured **3130 ms** and **21137 ms** on a
 cold start.
 
+## Measured directly, one line, one model
+
+    node eval/models.mjs --reps 5 --effort
+
+Every model gets the same prompt - one line, one word back - and the same
+five requests. `wall` is the whole `opencode run`, so it includes the
+process start and the plugin load. `reasoning` is the tokens the model
+spent thinking, which on a prompt this small is close to nothing for most of
+them: the difference between a tier and no tier shows up on real work, not
+on a one-line reply.
+
+| model | wall avg | median | max | reasoning | wall @low | reasoning @low |
+|---|---|---|---|---|---|---|
+| `beam-free` | 7,800 | **6,118** | 13,782 | 24 | 6,308 | 28 |
+| `minimax-m3.1-flash-free` | 7,114 | 6,903 | 8,448 | 0 | **6,292** | 0 |
+| `nemotron-3.5-lightning-free` | 8,140 | 6,882 | 13,554 | 86 | — | — |
+| `glm-5.3-flash-free` | 10,604 | 7,408 | 21,002 | 0 | 10,718 | 12 |
+| `nemotron-3-ultra-free` | 9,282 | 7,847 | 15,402 | 36 | — | — |
+| `nemotron-3.5-super-vl-free` | 10,404 | 8,353 | 19,763 | 15 | 10,988 | 17 |
+| `mimo-v2.6-flash-free` | 8,875 | 9,345 | 9,635 | 8 | — | — |
+| `big-pickle` | 9,983 | 10,438 | 11,372 | 8 | — | — |
+| `deepseek-v4.1-flash-free` | 15,840 | 10,659 | 25,805 | 0 | 14,193 | 46 |
+| `longcat-2.5-preview-free` | 13,099 | 11,996 | 16,991 | 16 | — | — |
+
+**What this says that the install tables do not.** The margin between the
+fastest model and the fourth is under a second - 6.1 s for Beam against 7.4 s
+for GLM - so a plan that has to be quick does not need the fastest model, it
+needs any of the first four. The spread at the other end is real: LongCat
+costs twice what Beam costs, on the same one-line question.
+
+**And the `low` column is not a reason to ask for `low` on a small job.**
+On a prompt this small the difference is inside the noise, and on one of them
+it is the wrong way round. The tier earns its keep on real work, and that
+is the measurement below, on the install's own messages.
+
+A dash in the `@low` columns means the model publishes **no `low` level**,
+so there is nothing to ask for: it runs at its natural rate. The script
+reads the catalogue to know which is which, rather than asking every model
+and reporting an answer it did not use.
+
+The model ids are shown without their provider; the script prints
+`opencode/beam-free`, and either spelling works with `--model`.
+
+**The free tier is noisy, and the table says so.** Five requests is five
+requests: `deepseek` had a 25.8 s outlier in a 10.7 s median, and Beam had a
+13.8 s one in a 6.1 s median. The median is the number to read, and a
+single request is not a measurement of anything.
+
 **What the safe ceiling costs, measured rather than assumed.** Planning is the
-lane that pays: 2624 ms on Beam against 5106 ms on DeepSeek, about twice the
-wait. Reviewing barely pays at all: 4716 ms on Nemotron 3 Ultra against 5757
-ms on GLM. Locating and verifying get *faster*, because DeepSeek publishes a
-`low` effort level and Big Pickle publishes none at all.
+lane that pays: the install measured 2624 ms on Beam against 5106 ms on
+DeepSeek, and the direct measurement says the same thing a different way -
+6.1 s against 10.7 s, about three quarters more.
+
+Reviewing does not pay at all. The install measured 4716 ms on Nemotron 3
+Ultra against 5757 ms on GLM, and the direct measurement has them level, with
+GLM marginally ahead: 7.4 s against 7.8 s. Either way, the model built for
+reviewing is not faster than the one the ceiling puts in its place, and
+neither is slow.
+
+Locating and verifying get *faster*, because DeepSeek publishes a `low` effort
+level and Big Pickle publishes none at all.
+
+**What a tier is worth, measured on this install rather than assumed.** The
+effort level is the cheapest lever there is: DeepSeek measures 320 reasoning
+tokens a message at `max` and 19 at `low`, a factor of 17, and a lane that
+asks for a tier and lands on a model that publishes none runs at that model's
+NATURAL rate instead. So the tier request is worth protecting.
+
+The three lanes in `fleet-all` whose model cannot serve the tier - `reviewer`
+on Nemotron 3 Ultra, `locate` on Nemotron 3.5 Lightning, `verify` on Big
+Pickle - are the fleet's CHEAPEST reasoners: 116, 59 and 18 reasoning tokens a
+message, against 289 for `build` and 198 for `plan`, which keep their tier on
+purpose. The tier is the lever, and it is not where the tokens are. The
+config header says so where it happens, in the form `locate=unset(asked low)`.
+
+One lane was not cheap, and it was fixed. `sweep` is a bulk mechanical
+search, so reasoning on it is pure waste, and it ran on LongCat: 420 reasoning
+tokens a message at 16.6 s, the most expensive way to ask a model to grep. No
+zero-retention model publishes a `none` level, so the lane could not even turn
+reasoning off when it asked to. It runs on DeepSeek now, which is
+zero-retention, publishes a `low` level, and measures 19 reasoning tokens at
+3.4 s.
 
 And one lane gets faster rather than slower. MiMo reads an image in 19.2 s;
 MiniMax reads the same image in 7.7 s. The unrestricted fleet put `vision` on
@@ -93,16 +198,48 @@ choice anyway.
 
 | model | measured | why it owns that lane |
 |---|---|---|
-| minimax-m3.1-flash-free | 3.1 s for a one-line reply, 512k output cap, reads an image in 7.7 s | The widest output cap on the tier and a stable tool call, so the lane that writes code fits, and it is also the fastest measured image reader. |
-| beam-free | **2.6 s average over 7 requests, the fastest on the tier**, 1M context, structured output | Planning is a reading-and-reasoning job with a long context and a structured answer. Nothing else on the tier does it faster. |
-| nemotron-3-ultra-free | 4.7 s average, but **34 s on a cold first request** | The largest open MoE available, which is what a two-pass review needs. The cold start is worth knowing about before blaming the route. |
-| nemotron-3.5-lightning-free | 3.7 s average over 4 requests, 5.5 s at worst, no selectable effort | No effort level, so it spends nothing on reasoning, and it can emit a file larger than 128k tokens. |
-| big-pickle | 6.5 s average over 2 requests, structured output, no effort levels, no images | Stable, and it answers in a shape a program can read. |
-| mimo-v2.6-flash-free | **19.2 s average with an image**, 23.8 s at worst - the slowest lane on the tier | It reads images correctly, which is the job. It publishes audio input too, and that path is unreachable (see below). |
-| glm-5.3-flash-free | 4.4-5.8 s, structured output, 3 effort levels | Careful and quick enough. A previous measurement put it at 16-27 s on a larger prompt; on the work a lane actually does it is five seconds. |
-| deepseek-v4.1-flash-free | 3.4-5.2 s, structured output, reasoning + agentic coding | The natural lane for JSON a program has to parse, and the best all-rounder inside a zero-retention ceiling. |
-| nemotron-3.5-super-vl-free | 4.2 s average over 3 requests, **the only model publishing a `none` effort level** | A lane that turns reasoning off entirely is a job of its own, and no other model can do it. |
-| longcat-2.5-preview-free | 9.4-13.2 s, 1M context, reads images, temperature on | The long-context lane. A preview endpoint: it can be withdrawn without a rename. |
+| minimax-m3.1-flash-free | 6.9 s median for a one-line reply, 512k output cap - the widest on the tier, reads an image in 7.7 s | The widest output cap on the tier and a stable tool call, so the lane that writes code fits, and it is also the fastest measured image reader. |
+| beam-free | **6.1 s median, the fastest of the ten**, 1M context, structured output | Planning is a reading-and-reasoning job with a long context and a structured answer. It is the fastest measured, but only by under a second over the next three - so the reason to pick it is the 1M context and the structured output as much as the speed. |
+| nemotron-3-ultra-free | 7.8 s median, but **34 s on a cold first request** | The largest open MoE available, which is what a two-pass review needs. The cold start is worth knowing about before blaming the route. |
+| nemotron-3.5-lightning-free | 6.9 s median, 13.6 s at worst, no selectable effort | No effort level, so it spends nothing on reasoning, and it can emit a file larger than 128k tokens. |
+| big-pickle | 10.4 s median, structured output, no effort levels, no images | Stable, and it answers in a shape a program can read. |
+| mimo-v2.6-flash-free | 9.3 s median for a line of text, **19.2 s average with an image** | It reads images correctly, which is the job. It publishes audio input too, and that path is unreachable (see below). |
+| glm-5.3-flash-free | 7.4 s median, structured output, 3 effort levels | Careful and quick enough. A previous measurement put it at 16-27 s on a larger prompt; on the work a lane actually does it is about seven seconds. |
+| deepseek-v4.1-flash-free | 10.7 s median, structured output, reasoning + agentic coding | The natural lane for JSON a program has to parse, and the best all-rounder inside a zero-retention ceiling. It was the slowest of the ten on the one-line measurement - the tier is free, and free is not fast. |
+| nemotron-3.5-super-vl-free | 8.4 s median, **the only model publishing a `none` effort level** | A lane that turns reasoning off entirely is a job of its own, and no other model can do it. |
+| longcat-2.5-preview-free | 12.0 s median - the slowest of the ten, 1M context, reads images, temperature on | The long-context lane. A preview endpoint: it can be withdrawn without a rename. |
+
+## What each model publishes
+
+From the live catalogue, 2026-10-21. `oc-harness doctor` re-checks all of
+it, and a model that is withdrawn says so rather than being measured.
+
+| model | effort levels | context | output cap | reads |
+|---|---|---|---|---|
+| `beam-free` | low, medium, high, xhigh, max | 1M | 131k | text |
+| `big-pickle` | **none** | 200k | 32k | text |
+| `deepseek-v4.1-flash-free` | low, high, max | 200k | 32k | text, image |
+| `minimax-m3.1-flash-free` | low, medium, high, xhigh, max | 1M | **512k** | text, image, video |
+| `nemotron-3-ultra-free` | **none** | 1M | 128k | text |
+| `nemotron-3.5-lightning-free` | **none** | 262k | **262k** | text |
+| `mimo-v2.6-flash-free` | **none** | 200k | 32k | text, image, audio, video |
+| `glm-5.3-flash-free` | low, high, max | 200k | 32k | text, image, video, pdf |
+| `nemotron-3.5-super-vl-free` | **none**, minimal, low, medium, high, xhigh | 1M | 131k | text, image |
+| `longcat-2.5-preview-free` | **none** | 1M | 131k | text, image |
+
+Two things fall out of this table that the prose above rests on.
+
+**Only one model can turn reasoning off entirely**, and it is
+`nemotron-3.5-super-vl-free`. The `sweep` lane is a bulk mechanical search,
+so reasoning on it is pure waste - and the model that could switch it off
+publishes `none` and the four zero-retention models do not, so under the
+ceiling that lane cannot ask for it. It runs on DeepSeek, which publishes
+`low`, and that is the closest thing available.
+
+**A model that publishes no effort level still spends what it spends.** It
+runs at its natural rate. That is not a defect: it is the reason Big Pickle
+is cheap to use, and the reason the kit asks for a tier only where the tier
+is worth having.
 
 ## What the tier cannot do
 
@@ -147,6 +284,11 @@ mis-served by that:
 - `locate` asked for `low` effort and landed on a model that publishes no
   effort levels at all, so the request was dropped on the floor. It now lands
   on DeepSeek, which publishes `low`, and measures 3612 ms.
+- `sweep` is a bulk mechanical search, so reasoning on it is pure waste. It
+  ran on LongCat at 420 reasoning tokens a message and 16.6 s - and no
+  zero-retention model publishes a `none` level, so the lane could not turn
+  reasoning off even when it asked to. It runs on DeepSeek now: 19 reasoning
+  tokens at 3.4 s.
 - `vision` landed on the same model, at 9-13 s, when MiniMax reads the same
   image in 7.7 s.
 
@@ -170,14 +312,22 @@ base profile would reveal.
 
 ### What the safe fleet gives up
 
-`sweep` is the one lane that loses its reason to exist under the ceiling.
+`sweep` was the one lane that lost its reason to exist under the ceiling.
 Super VL is the only model on the tier that publishes a `none` effort level,
-so "reasoning off" is a job of its own; and Super VL is a may-train provider,
-so the ceiling excludes it. Under `fleet-safe` the lane falls back to
-LongCat, which publishes no effort levels at all, and it is a second `locate`
-with a 1M context window. The lane is kept, because `oc-harness lanes` should
-decide that from measurements rather than from a guess - but the docs say out
-loud that the capability is gone.
+so "reasoning off" is a job of its own; and Super VL is a may-train
+provider, so the ceiling excludes it. Under `fleet-safe` the lane ran on
+LongCat, which publishes no effort levels at all, and it was a second
+`locate` with a 1M context window.
+
+It runs on DeepSeek now. Not LongCat: LongCat measured 420 reasoning tokens
+a message at 16.6 seconds, which is the most expensive way to ask a model to
+grep, and DeepSeek publishes a `low` level and measures 19 reasoning tokens
+at 3.4. The lane is still not the unrestricted fleet's lane, and the
+reasoning-off capability is still out of reach under a zero-retention
+ceiling - but the cost of the substitute is no longer the whole story.
+
+The lane is kept, because `oc-harness lanes` should decide that from
+measurements rather than from a guess.
 
 ## Measuring, and cutting
 
@@ -187,19 +337,27 @@ is to count the requests:
 ```
 $ oc-harness lanes
 
-  lane            reqs     avg ms  model
-  ----------------------------------------------------------------
-  plan                7       2624  beam-free
-  build              21      10104  minimax-m3.1-flash-free
-  reviewer            2       4716  nemotron-3-ultra-free
+  lane             reqs     avg ms     avg ctx  model
+  ----------------------------------------------------------------------
+  plan                0          -           -  deepseek-v4.1-flash-free        never used
+  build              87       9935           -  minimax-m3.1-flash-free, deepseek-v4.1-flash-free
+  reviewer            0          -           -  glm-5.3-flash-free              never used
   ...
-  8 lane(s) were never used    cut candidates, not defects
+  9 lane(s) were never used cut candidates, not defects
+  one request is one assistant message; avg ms covers the requests that finished
 ```
 
 A lane that is never used is a description the model reads on every turn and a
 free model doing nothing. Cut it from the profile's `roles`, and the lane
 leaves the config with it.
 
+That is also the counter that says a lane is NOT earning its place, and it
+is the honest one to act on: it counts the requests a working session
+actually made. `node eval/models.mjs` answers the other question - how long
+a lane would wait - and the two together are what a routing decision needs.
+The counter alone would cut a lane that is slow but used, and the
+measurement alone would keep a lane nobody calls.
+
 The counter reads `.space/telemetry.jsonl` in the project, written by the
 plugin: one record per assistant message, with the lane, the model and the
 duration. Two things about it are worth knowing, because both were wrong at
@@ -250,5 +408,7 @@ directory is exactly as dangerous as one in the plural one.
   config, so it is sized by the **narrowest** lane: 24,000 tokens for
   `fleet-safe`, whose narrowest lane has a 200k window, and 19,200 for
   `fleet-all`, whose narrowest lane is Big Pickle at a 160k input cap.
+  (Big Pickle's CATALOGUE context is 200k; the profile pins its input cap to
+  160k, and the cap is the number that governs.)
   Sizing it by the widest model instead would reserve 125,829 tokens and
   leave every 200k lane with 74k usable.