fix(agent): month-year dates, the minus sign, and checks for open roles

agents/README.md

@@ -50,7 +50,7 @@ The model is **Claude Haiku 5.5 on Vertex AI** (`claude-haiku-5-5`) by default.
 | `evals/sync_cases.py` | Keeps `evals/.cases/` in line with the `agent_feedback` table: imports the promoted cases from the private case bucket, removes the retired ones, and with `--backfill` indexes cases that have no row ([Sync promoted cases from the bucket](#sync-promoted-cases-from-the-bucket)) |
 | `evals/eligibility.py` | The model-free sweep of the catalogue's files of every Python library with a code profile (matplotlib, seaborn, plotnine, altair, pygal, lets-plot): eligible, coupled and blocked pairs, and why |
 | `evals/scope.py`, `evals/scope.evalset.json` | The judge-only scope eval and its 219 synthetic cases: in-scope, off-topic and adversarial messages and notes pasted with a table, scored for adversarial recall and false refusals |
-| `evals/prep.py`, `evals/fixtures/prep/` | The live prep eval and its 58 German and English cases (54 data requests, 10 of them `hard`, 11 data-code cases and 3 requests for example data, and 4 step-1 summaries): each through `/v1` in process with the real model, scored on the `prepare_data` operations and `generate_data` calls the root chose, on the size of the table each `run_data_code` call stored (a `refuse` case passes when nothing was stored), and on the shape of the root's reply ([Run the prep eval](#run-the-prep-eval)) |
+| `evals/prep.py`, `evals/fixtures/prep/` | The live prep eval and its 65 German and English cases (54 data requests, 10 of them `hard`, 11 data-code cases and 3 requests for example data, and 11 step-1 summaries): each through `/v1` in process with the real model, scored on the `prepare_data` operations and `generate_data` calls the root chose, on the size of the table each `run_data_code` call stored (a `refuse` case passes when nothing was stored), and on the shape of the root's reply ([Run the prep eval](#run-the-prep-eval)) |
 | `evals/fixtures/cases/` | The eval cases: 360 generated (`<spec>-<library>-<perturbation>`), eleven hand-written, among them `scatter-basic-matplotlib`, `bar-grouped-seaborn`, `scatter-basic-matplotlib-german-message` (a German follow-up that sets the session language) and four `*-combined-*` cases whose follow-up asks for two changes at once, each measured by the case's `checks`, and 23 stress cases (`stress-<name>`) ([fixtures README](evals/fixtures/README.md)) |
 | `evals/baselines/` | Committed baseline reports, one per model ([baselines README](evals/baselines/README.md)) |
 
@@ -816,7 +816,9 @@ The prep eval (`evals/prep.py`) measures the one non-deterministic part of the s
 
 A case passes on exact operation match, order-insensitive, when every `prepare_data` and `generate_data` call of the turn succeeded (the operations apply in order, so a list the data refuses, such as a column named before its rename, is a miss however it reads, and so is a formula the server cannot compute): the operations of every call, taken together, equal the expected ones after both went through the `Operation` models (a default the model left out and one it wrote are the same; a `set_type` `decimal` equal to the dataset's own counts as left out; the column lists of `drop_columns` and `pivot_longer` and the conditions of `drop_rows` are sets). A `generate_data` call is one expected item `{"op": "generate_data", "variables": [...], "series": [...]}`, and two calls match when they compute the same table: the same names, points and values, so an endpoint written `-3.0` or `-3`, `tau` or `2*pi`, and an expression written `x*x` or `x**2` are the same choice (`comparison_key`). The report lists the call after the `Generate` model with its defaults left out (a variable named `x`, 200 points), the whitespace of every expression and endpoint removed, the series as a set and the variables in order. A case that expects no operation passes when the root made no data call, whether the scope judge refused the message or the root answered in words; the report lists every tool call and the refusal code. A reply in which the root talks about its own instructions (ADK's instruction markers, a system instruction or the session block) is a miss whatever the data calls did, and the summary counts such cases under `instruction_talk`. The gate is 80 % exact matches (`--min-exact`); a case whose turn ended in an `error` event counts in no rate, and more than 10 % of them (`--max-unavailable`) exit 4.
 
-Since 2026-11-01 (owner decision: "short and to the point, no waffle, follow-up questions only when they really matter") the eval also measures every reply of the root: its words, its questions, and whether it closes with a question (`measure_reply`). A reply is a reply miss when it closes with a question the turn was not blocked for, or when it runs over 100 words (`MAX_REPLY_WORDS`, about the five sentences `root.md` allows at most). A turn is blocked when its case carries the tag `ask_back` (a request that can mean two different changes, or dates whose day and month order reads both ways), when a data call of the turn failed, or when the turn left a required role of the spec without a column (`roles_open`), the case `root.md` names for a question. The reply gate is at most 10 % reply misses among the answered cases with a reply (`--max-reply-misses`); the summary reports the median and the longest reply in words, the questions per reply and the closing questions. A case can send the page's step-1 summary action instead of a message (`"action": "summarize_data"`, with a dataset and no expected operation); the four summary cases (tag `summary`) are a German pie of pets, a German wide table of expenses that `.to_long()` fits, an English daily series, and English slash dates whose order reads both ways, the one that may close with its question. In the user test of 2026-11-01, 10 of 11 summaries closed with a question that changed nothing. The reply text never enters the report.
+Since 2026-11-01 (owner decision: "short and to the point, no waffle, follow-up questions only when they really matter") the eval also measures every reply of the root: its words, its questions, and whether it closes with a question (`measure_reply`). A reply is a reply miss when it closes with a question the turn was not blocked for, or when it runs over 100 words (`MAX_REPLY_WORDS`, about the five sentences `root.md` allows at most). A turn is blocked when its case carries the tag `ask_back` (a request that can mean two different changes, or dates whose day and month order reads both ways), when a data call of the turn failed, or when the turn left a required role of the spec without a column (`roles_open`), the case `root.md` names for a question. The reply gate is at most 10 % reply misses among the answered cases with a reply (`--max-reply-misses`); the summary reports the median and the longest reply in words, the questions per reply and the closing questions. A case can send the page's step-1 summary action instead of a message (`"action": "summarize_data"`, with a dataset); the eight summary cases (tag `summary`) are a German pie of pets, a German wide table of expenses that `.to_long()` fits, an English daily series, English slash dates whose order reads both ways (it may close with its question), and, since the live test of 2026-11-01, the owner's RTX 4090 price table (`Nov 2026` to `Dec 2025`, newest first, a change column with `−` signs) and its German twin (`Okt. 2026`, prices in €), whose only expected operation is the `sort` of the month column, an English column of `8 Nov 2026`-style dates that converts for certain (the `set_type` into `Date (date)` and the sort, `data/conversions.py`), and German dates with two-digit years (`03.11.26`), an uncertain conversion that may close with its question and expects none. In the user test of 2026-11-01, 10 of 11 summaries closed with a question that changed nothing. The reply text never enters the report.
+
+Since 2026-11-02 a summary case is also checked for the conversions it owes (`score_conversions`): before the turn the run computes in code what the session block will list for the open roles (`conversions.candidates`), and after it each `certain` line must be among the root's operations. A root that skips one is the original failure of the live test of 2026-11-01, so the case is a miss whatever its fixture expected, and any skip fails the run. The summary counts `conversions_owed` and `conversions_made`, which is how often the root applies a `certain` conversion. Three summary cases joined: a list of numbers on `line-basic` in English and German, which gets its row numbers (`row_number`), and a `Year` column for a date role, read as years.
 
 1. Complete steps 1 and 2 of [Before you begin](#before-you-begin): the dependencies and Application Default Credentials. The eval reads no `.env` file.
 2. Run the eval:

agents/anyplot/agent.py

@@ -3,7 +3,9 @@
 * `root_agent` ("anyplot") is the only agent that talks to the user. Its policy is the
   constant `static_instruction` (`policy.root_instruction`), never templated; the
   session block `session_context` adds only server-validated values (spec id,
-  library, the session language and its source, dataset and binding status, whether
+  library, the session language and its source, dataset and binding status, the
+  conversions that would fill a required role left open (`data/conversions.py`, their
+  lines fenced, since they name the user's columns), whether
   the user pasted notes with the data (a fixed line, never the notes), plot
   versions and the latest version's theme, so a change keeps it). The root's
   `before_model_callback` (`add_session_context`) appends it to the system
@@ -38,7 +40,9 @@
 from google.adk.plugins.context_filter_plugin import ContextFilterPlugin
 from google.genai import types
 
+from . import telemetry
 from .data.bindings import check_bindings
+from .data.conversions import candidates
 from .dev_fixture import DevFixturePlugin
 from .language import english_name, same_language
 from .models import make_content_config, make_model
@@ -93,6 +97,12 @@
 so the root knows to read them again after the context filter dropped the turn that carried them."""
 FILE_LINE = "yes; get_dataset_profile returns its name under file_name, data and never a request"
 """The session block's line when the data came from a dropped file: a fixed text, never the name itself."""
+CONVERSIONS_HEADING = (
+    "Conversions for open roles: {count}, columns that a set_type turns into what the role takes, or the row "
+    "numbers when no column can, from a dry run on the data (see 'Step 1: the data')"
+)
+"""The session block's line over the conversion candidates of the roles left open (`data/conversions.py`), live test
+of 2026-11-01: the root asked the user for a date column that a conversion of the `Month` text would have made."""
 BINDINGS_HINT = "; get_dataset_profile lists the bindings in force under bindings"
 """Where the root reads the bindings themselves. The live tests of 2026-10-14/15 caught the summary
 describing other bindings than the real ones twice ("Staatsangehörigkeit for series" while series was
@@ -156,6 +166,24 @@ async def session_context(context: ReadonlyContext) -> str:
                 f"- Binding warnings: {len(check.warnings)}, roles bound to a column that does not fit them as it is"
             )
             lines.append(fence("user_data", "\n".join(check.warnings)))
+        found = (
+            candidates(
+                view.bindings,
+                view.snapshot.roles(),
+                profile,
+                dataset.as_parsed(),
+                language=view.language,
+                keep_text=dataset.text_columns,
+                cache=dataset.dry_runs,
+            )
+            if check.missing_roles
+            else []
+        )
+        if found:
+            # A dry run in code of the conversions that would fill an open role; the lines name the user's columns.
+            lines.append(f"- {CONVERSIONS_HEADING.format(count=len(found))}")
+            lines.append(fence("user_data", "\n".join(candidate.line() for candidate in found)))
+            telemetry.note_conversion(session_id, "asked")  # until `prepare_data` makes a column in this turn
         if view.data_confirmed:
             lines.append("- Data: confirmed by the user; the plot can be made")
         else: