fix(roi): split-write checkpoint; scan-range cap; try/except sum

Complete the split-storage pattern for the offchain-deposit
checkpoint. round-4 added ``read_offchain_deposits_from_disk`` for
the read side of ``_save_agent_performance_summary``; this commit
adds the matching ``write_offchain_deposits_to_disk`` on
``SharedState`` and rewires ``_fetch_offchain_prepaid_wei`` to use
both.

``_fetch_offchain_prepaid_wei`` now sources ``state`` from
``read_offchain_deposits_from_disk`` (raw JSON, lenient on siblings)
and persists via ``write_offchain_deposits_to_disk`` (raw JSON,
sibling-preserving). It no longer touches
``read_existing_performance_summary`` or
``overwrite_performance_summary``, so a nested-field ``__post_init__``
raise elsewhere on disk can neither cause a spurious re-seed
(sourcing) nor cascade into wiping sibling fields at write time
(persistence). Split-write mirrors the split-read and closes both
sides of the corruption cascade class.

Scan-range cap. Added ``OFFCHAIN_MAX_SCAN_RANGE = 5000`` and clamped
``to_block = min(head_block, from_block + OFFCHAIN_MAX_SCAN_RANGE
- 1)``. An agent offline for longer than the provider's
``eth_getLogs`` block-range cap (~14h+ of Gnosis blocks — a Pearl
laptop closed over a weekend) previously scanned
``[checkpoint+1, head]`` which exceeded common public-provider caps
(Alchemy / Infura ~10k), got rejected every cycle with ERROR
performative, and the checkpoint froze. Now each cycle marches the
checkpoint forward at most one chunk until it reaches head.

Exception guard on the entries sum. Wrapped
``sum(int(entry["amount"]) for entry in entries)`` in
``try/except (KeyError, TypeError, ValueError)`` and routed to the
malformed-response branch. Previously a shape drift on the entries
wire format raised uncaught, and with ``skill_exception_policy:
stop_and_exit`` in aea-config.yaml that terminated the whole agent —
harder failure mode than freezing the checkpoint. Reviewer flagged
my earlier "raise loudly, the ERROR performative arm catches it"
reasoning as wrong; the performative gate above the sum only
catches API performatives, not application exceptions inside the
sum.

Tests. Three model-level tests around ``write_offchain_deposits_to_disk``:
sibling-field preservation with a real file round-trip; minimal
file when nothing exists on disk; recovery from a truncated JSON
prefix. Two behaviour-level tests:
``test_scan_range_capped_at_offchain_max_scan_range`` (long offline
stretch, verify ``to_block`` clamped and checkpoint advances by
one chunk); ``test_malformed_entry_shape_returns_cached_total_no_raise``
(entries missing ``amount``, verify no raise and warning routing).
Existing ``TestFetchOffchainPrepaidWei`` helper reworked to mock the
disk helpers and assert (via ``AssertionError`` side-effects) that
the whole-summary read/write paths are NOT touched by this behaviour.

Author: DIvyaNautiyal07

packages/packages.json

@@ -10,16 +10,16 @@
         "contract/valory/mech_prepaid_reader/0.1.0": "bafybeiaitnqi4kc2wiltj3k6a3iujxritwkh5w2a2xoff7dosoqy7p3sou",
         "connection/valory/polymarket_client/0.1.0": "bafybeid4eo37pka4tfhsjs6rb3knrugmokcvihust2dpfrgob56y77xd44",
         "skill/valory/market_manager_abci/0.1.0": "bafybeig3ak2eoimxwuupybde4napyi2jlf7pxk35se2zclikete2ni5hti",
-        "skill/valory/decision_maker_abci/0.1.0": "bafybeig67yeaa7dnt7eqxyxmv6jo5ulwp24fu36dl3h5juovhjhajmyjpu",
-        "skill/valory/trader_abci/0.1.0": "bafybeihxz5ephxy7wq5lvfsd5tp63typcikl6fnlkes3eo6j27w4wbr6le",
-        "skill/valory/tx_settlement_multiplexer_abci/0.1.0": "bafybeifqnagoqwli3wstgfgyhz3pwlg5676crfsyksdgqjacdjh2p5uuuq",
+        "skill/valory/decision_maker_abci/0.1.0": "bafybeigwwaumxabujrtouaq2sphjusnex57doqofx625zuwdyxdqh3en4e",
+        "skill/valory/trader_abci/0.1.0": "bafybeigtsnje4cn624bm27ssrwhwctckeud6etxpixmloj4chnehbnpinq",
+        "skill/valory/tx_settlement_multiplexer_abci/0.1.0": "bafybeigzpmxt7glz76tdm62k2mcm7drhyeomgl6euf4nbq6q2l75xopjfu",
         "skill/valory/staking_abci/0.1.0": "bafybeia6csjgex5zxdt6vymiaengu75lyzjdgfrnp65ewlghthiniquhje",
-        "skill/valory/agent_performance_summary_abci/0.1.0": "bafybeigouukzr7brpnjubkluoy4whcbr4lesg6odxexj4geybapmpgmv3e",
-        "skill/valory/check_stop_trading_abci/0.1.0": "bafybeifwkokgn36b6pijkije47ikywtoprtfap5a3kbofnmckpzhahqxui",
-        "skill/valory/chatui_abci/0.1.0": "bafybeibpnstfaurdsneizqvdrcic3bkz57z2h3rrfy4cxnkypguonncs3i",
-        "agent/valory/trader/0.1.0": "bafybeibaooayqfkl7vclisdyytj4li3q56a72p3m3enpnjqxnqoh6aabhi",
-        "service/valory/trader_pearl/0.1.0": "bafybeif55hh6k6pasb3x7dyhlcmcqui7lv6f3jkrefx3dztzmscck6ual4",
-        "service/valory/polymarket_trader/0.1.0": "bafybeieo4sirdpygxvtonxc4yss4pb534gvskzwdtjm5r5tk56rieprycy"
+        "skill/valory/agent_performance_summary_abci/0.1.0": "bafybeiagxj4awrr7shhawqfblyhdrcpbb7rvsexxm7k2l3wp3odsmvfa7u",
+        "skill/valory/check_stop_trading_abci/0.1.0": "bafybeieelax7wbwhpmk2vmcbackin52nahyu6wuv3dtfzmdgyvjpwibtiu",
+        "skill/valory/chatui_abci/0.1.0": "bafybeig4qckyl7exilbd4egapsttvfrecfu4buywx632gixuvkwcjqqssu",
+        "agent/valory/trader/0.1.0": "bafybeifgjugs6q2y6kjiecpx3zbc6igowfwlzkx6l6evkkw4vv434t5dja",
+        "service/valory/trader_pearl/0.1.0": "bafybeih27gjrohcdoobsxq2xu2ckgcbxisth27ydjuow65jhg3cs2cap2m",
+        "service/valory/polymarket_trader/0.1.0": "bafybeidvop66yzjn7xcl3hyam6k2ocuqksnejlawn7bq3msgpluljwdcsy"
     },
     "third_party": {
         "protocol/valory/acn_data_share/0.1.0": "bafybeieqisfr6lhaaesq37lk7vyaxza2h6nucshksgyo2djqfyf2saqo4i",

packages/valory/agents/trader/aea-config.yaml

@@ -70,15 +70,15 @@ skills:
 - valory/reset_pause_abci:0.1.0:bafybeifrnmlscfiki7wojs6zasppefupd2wkjyq6orhdllfppjud7isrxi
 - valory/termination_abci:0.1.0:bafybeih6sa2ttcvvjk7rcfn7ure2ws23kcfufssb4qmluzcfvednkfpvoi
 - valory/transaction_settlement_abci:0.1.0:bafybeihizaapype5sz7vlosms3nfo6jzfj4ujycuf5uthviq2otvmw6f6m
-- valory/tx_settlement_multiplexer_abci:0.1.0:bafybeifqnagoqwli3wstgfgyhz3pwlg5676crfsyksdgqjacdjh2p5uuuq
+- valory/tx_settlement_multiplexer_abci:0.1.0:bafybeigzpmxt7glz76tdm62k2mcm7drhyeomgl6euf4nbq6q2l75xopjfu
 - valory/market_manager_abci:0.1.0:bafybeig3ak2eoimxwuupybde4napyi2jlf7pxk35se2zclikete2ni5hti
-- valory/decision_maker_abci:0.1.0:bafybeig67yeaa7dnt7eqxyxmv6jo5ulwp24fu36dl3h5juovhjhajmyjpu
-- valory/trader_abci:0.1.0:bafybeihxz5ephxy7wq5lvfsd5tp63typcikl6fnlkes3eo6j27w4wbr6le
+- valory/decision_maker_abci:0.1.0:bafybeigwwaumxabujrtouaq2sphjusnex57doqofx625zuwdyxdqh3en4e
+- valory/trader_abci:0.1.0:bafybeigtsnje4cn624bm27ssrwhwctckeud6etxpixmloj4chnehbnpinq
 - valory/staking_abci:0.1.0:bafybeia6csjgex5zxdt6vymiaengu75lyzjdgfrnp65ewlghthiniquhje
-- valory/check_stop_trading_abci:0.1.0:bafybeifwkokgn36b6pijkije47ikywtoprtfap5a3kbofnmckpzhahqxui
+- valory/check_stop_trading_abci:0.1.0:bafybeieelax7wbwhpmk2vmcbackin52nahyu6wuv3dtfzmdgyvjpwibtiu
 - valory/mech_interact_abci:0.1.0:bafybeici5tsgtlypnyrnp5colj7katnsyjfdzt6js4xdbnpdms4232muje
-- valory/chatui_abci:0.1.0:bafybeibpnstfaurdsneizqvdrcic3bkz57z2h3rrfy4cxnkypguonncs3i
-- valory/agent_performance_summary_abci:0.1.0:bafybeigouukzr7brpnjubkluoy4whcbr4lesg6odxexj4geybapmpgmv3e
+- valory/chatui_abci:0.1.0:bafybeig4qckyl7exilbd4egapsttvfrecfu4buywx632gixuvkwcjqqssu
+- valory/agent_performance_summary_abci:0.1.0:bafybeiagxj4awrr7shhawqfblyhdrcpbb7rvsexxm7k2l3wp3odsmvfa7u
 - valory/funds_manager:0.1.0:bafybeicfw7qjtnlcbw6awsvxdahqu42fxsfsd3wy7mbyhcbls4l5pee3by
 customs:
 - valory/kelly_criterion:0.1.0:bafybeidme2uybzyzagky3jtosheqmpgc6b6w4rum7cdxw2fc35nnpwblfe

packages/valory/services/polymarket_trader/service.yaml

@@ -8,7 +8,7 @@ license: Apache-2.0
 fingerprint:
   README.md: bafybeichoh5tnk4qctaazcjcrxvpwc5yasodleqoeapnfapwrrdh3tm5gu
 fingerprint_ignore_patterns: []
-agent: valory/trader:0.1.0:bafybeibaooayqfkl7vclisdyytj4li3q56a72p3m3enpnjqxnqoh6aabhi
+agent: valory/trader:0.1.0:bafybeifgjugs6q2y6kjiecpx3zbc6igowfwlzkx6l6evkkw4vv434t5dja
 number_of_agents: 1
 deployment:
   agent:

packages/valory/services/trader_pearl/service.yaml

@@ -8,7 +8,7 @@ license: Apache-2.0
 fingerprint:
   README.md: bafybeibg7bdqpioh4lmvknw3ygnllfku32oca4eq5pqtvdrdsgw6buko7e
 fingerprint_ignore_patterns: []
-agent: valory/trader:0.1.0:bafybeibaooayqfkl7vclisdyytj4li3q56a72p3m3enpnjqxnqoh6aabhi
+agent: valory/trader:0.1.0:bafybeifgjugs6q2y6kjiecpx3zbc6igowfwlzkx6l6evkkw4vv434t5dja
 number_of_agents: 1
 deployment:
   agent:

packages/valory/skills/agent_performance_summary_abci/behaviours.py

@@ -85,6 +85,15 @@
 _ZERO_ADDRESS = (
     "0x" + "0" * 40
 )  # placeholder used to disable off-chain deposit scanning
+
+# Upper bound on the ``eth_getLogs`` block range per cycle. Safely
+# below common public-provider caps (Alchemy / Infura ~10k). When an
+# agent has been offline long enough that the catchup range exceeds
+# this, subsequent cycles march the checkpoint forward one chunk at
+# a time until it reaches head — instead of the pre-fix behaviour
+# where the range grew unbounded and every scan hit the provider cap,
+# permanently freezing the checkpoint.
+OFFCHAIN_MAX_SCAN_RANGE = 5_000
 QUESTION_DATA_SEPARATOR = "\u241f"
 PREDICT_MARKET_DURATION_DAYS = 4
 WXDAI_ADDRESS = "0xe91D153E0b41518A2Ce8Dd3D7944Fa863463a97d"  # wxDAI on Gnosis Chain
@@ -417,8 +426,22 @@ def _fetch_offchain_prepaid_wei(
         # under pre-deposit-as-loss stays counted even if the operator
         # flips use_offchain off or clears the tracker address. Once
         # counted, always counted.
-        summary = self.shared_state.read_existing_performance_summary()
-        state = summary.offchain_deposits or OffchainDepositState()
+        #
+        # Source ``state`` from the raw-JSON re-read helper, not from the
+        # full-summary read. ``read_existing_performance_summary`` degrades
+        # to a fresh dataclass on ANY nested ``__post_init__`` raise
+        # (e.g. a rollback where a newer release added an
+        # ``Achievements`` sub-field that a downgraded reader can't
+        # validate). If we sourced ``state`` from that degraded summary,
+        # ``state.last_scanned_block`` would be ``None`` and the seed
+        # branch below would re-seed from head — wiping a still-valid
+        # on-disk checkpoint. Reading only the ``offchain_deposits``
+        # sub-dict leniently insulates this helper from sibling
+        # ``__post_init__`` failures.
+        state = (
+            self.shared_state.read_offchain_deposits_from_disk()
+            or OffchainDepositState()
+        )
         cached_total = state.total_deposited_wei
 
         marketplace_config = self.params.mech_marketplace_config
@@ -484,13 +507,18 @@ def _fetch_offchain_prepaid_wei(
 
         # First-run seed: don't sweep history from block 0. Policy: only
         # deposits from the checkpoint forward count against ROI.
+        # Split-write via ``write_offchain_deposits_to_disk`` (see the
+        # sourcing note above and the model docstring): mid-cycle
+        # writes update only the ``offchain_deposits`` sub-dict on
+        # disk, so a fresh degraded summary read from another site
+        # can't cascade into wiping sibling fields at this write.
         if state.last_scanned_block is None:
-            summary.offchain_deposits = OffchainDepositState(
-                total_deposited_wei=0,
-                last_scanned_block=head_block,
+            self.shared_state.write_offchain_deposits_to_disk(
+                OffchainDepositState(
+                    total_deposited_wei=0,
+                    last_scanned_block=head_block,
+                )
             )
-            summary.timestamp = self.shared_state.synced_timestamp
-            self.shared_state.overwrite_performance_summary(summary)
             self.context.logger.info(
                 f"[Offchain Deposits] First run — seeded checkpoint to block "
                 f"{head_block}, cumulative total starts at 0."
@@ -504,14 +532,24 @@ def _fetch_offchain_prepaid_wei(
             # or empty devnet). Nothing to scan; nothing to persist.
             return cached_total
 
+        # Cap the scan range so an agent that has been offline for
+        # longer than the provider's ``eth_getLogs`` block-range cap
+        # (a Pearl laptop closed over a weekend, ~14h+ of Gnosis
+        # blocks) catches up in chunks over successive cycles rather
+        # than every scan hitting the cap and permanently freezing the
+        # checkpoint. Each cycle marches ``last_scanned_block`` forward
+        # by at most ``OFFCHAIN_MAX_SCAN_RANGE`` blocks until it reaches
+        # head.
+        to_block = min(head_block, from_block + OFFCHAIN_MAX_SCAN_RANGE - 1)
+
         response = yield from self.get_contract_api_response(
             performative=ContractApiMessage.Performative.GET_STATE,  # type: ignore
             contract_address=balance_tracker_address,
             contract_id=str(MechPrepaidReaderContract.contract_id),
             contract_callable="get_deposit_events_for_requester",
             requester=safe_address,
             from_block=from_block,
-            to_block=head_block,
+            to_block=to_block,
             chain_id=self.params.mech_chain_id,
         )
         if response.performative != ContractApiMessage.Performative.STATE:
@@ -535,34 +573,39 @@ def _fetch_offchain_prepaid_wei(
 
         # Aggregation lives here (not in the classmethod) by convention:
         # contract classmethods return processed log entries; sum happens
-        # in the caller.
-        # Hard-index the flat entry shape returned by
-        # ``get_deposit_events_for_requester`` (``{"amount": int,
-        # "block_number": int}`` per entry). ``.get(..., 0)`` would
-        # silently zero-fill a shape drift while still advancing the
-        # checkpoint below to ``head_block`` — an irreversible under-count
-        # that would flatter ROI. Raise loudly instead so the ERROR
-        # performative arm above catches it and the checkpoint stays put.
-        new_wei = sum(int(entry["amount"]) for entry in entries)
-
-        # Checkpoint advances to the scanned head every successful cycle,
-        # even with zero matches, so subsequent cycles start from
-        # ``head_block + 1`` (small range) rather than re-scanning the
-        # same growing range every cycle. Without this, a long
-        # deposit-quiet stretch eventually crosses provider ``eth_getLogs``
-        # caps and the ERROR performative freezes future accrual forever
-        # with only warning logs.
+        # in the caller. Wrap the hard-index in
+        # ``(KeyError, TypeError, ValueError)`` so a shape drift on the
+        # entries wire format routes into the malformed-response branch
+        # instead of raising uncaught: the ``if response.performative
+        # != STATE`` gate above only catches API performatives, and with
+        # ``skill_exception_policy: stop_and_exit`` an unhandled skill
+        # exception terminates the whole agent — a harder failure than
+        # freezing the checkpoint.
+        try:
+            new_wei = sum(int(entry["amount"]) for entry in entries)
+        except (KeyError, TypeError, ValueError) as exc:
+            self.context.logger.warning(
+                "Offchain-deposit scan returned entries with unexpected "
+                f"shape ({exc!r}); leaving checkpoint and cached total unchanged."
+            )
+            return cached_total
+
+        # Checkpoint advances to the scanned upper bound every successful
+        # cycle, even with zero matches, so subsequent cycles start from
+        # ``to_block + 1`` (small range for a caught-up agent, bounded
+        # by ``OFFCHAIN_MAX_SCAN_RANGE`` for one still catching up)
+        # rather than re-scanning the same growing range every cycle.
         total_wei = cached_total + new_wei
-        summary.offchain_deposits = OffchainDepositState(
-            total_deposited_wei=total_wei,
-            last_scanned_block=head_block,
+        self.shared_state.write_offchain_deposits_to_disk(
+            OffchainDepositState(
+                total_deposited_wei=total_wei,
+                last_scanned_block=to_block,
+            )
         )
-        summary.timestamp = self.shared_state.synced_timestamp
-        self.shared_state.overwrite_performance_summary(summary)
 
         self.context.logger.info(
             f"[Offchain Deposits] cached={cached_total}, new_this_cycle={new_wei}, "
-            f"total={total_wei}, last_scanned_block={head_block}"
+            f"total={total_wei}, last_scanned_block={to_block}"
         )
         return total_wei
 

packages/valory/skills/agent_performance_summary_abci/models.py

@@ -449,6 +449,62 @@ def read_offchain_deposits_from_disk(self) -> Optional["OffchainDepositState"]:
             )
             return None
 
+    def write_offchain_deposits_to_disk(self, state: "OffchainDepositState") -> None:
+        """Atomic-write ``offchain_deposits`` to disk, preserving sibling fields.
+
+        Mirrors ``read_offchain_deposits_from_disk`` on the write side.
+        Reads the raw JSON dict (bypassing every dataclass
+        ``__post_init__``), updates only the ``offchain_deposits`` key,
+        writes back atomically via tempfile + ``os.replace``.
+        ``_fetch_offchain_prepaid_wei`` uses this instead of
+        ``overwrite_performance_summary(summary)`` because the summary
+        read it would otherwise pair with can silently degrade to a
+        fresh dataclass on any nested-field ``__post_init__`` raise,
+        which would then wipe every sibling field on disk (metrics,
+        agent_details, prediction_history, ...) during the mid-cycle
+        checkpoint write. Split-write keeps our checkpoint safe without
+        depending on other fields' validation state.
+
+        If the file is missing or unreadable, writes a minimal
+        ``{"offchain_deposits": ...}`` — the operator has explicitly
+        opted in to off-chain accounting at this point, so keeping the
+        checkpoint alive is more important than refusing to write over
+        a corrupt file. The next cycle-end
+        ``_save_agent_performance_summary`` will re-populate the sibling
+        fields from live data.
+
+        :param state: the ``OffchainDepositState`` to persist.
+        """
+        file_path = self.params.store_path / AGENT_PERFORMANCE_SUMMARY_FILE
+
+        try:
+            with open(file_path, "r") as f:
+                raw = json.load(f)
+            if not isinstance(raw, dict):
+                raw = {}
+        except (FileNotFoundError, json.JSONDecodeError):
+            raw = {}
+
+        raw["offchain_deposits"] = asdict(state)
+
+        # tempfile in the same directory so ``os.replace`` is atomic on
+        # POSIX (both paths on one filesystem).
+        fd, tmp_path = tempfile.mkstemp(
+            prefix=file_path.name + ".", dir=str(file_path.parent)
+        )
+        try:
+            with os.fdopen(fd, "w") as f:
+                json.dump(raw, f, indent=4)
+                f.flush()
+                os.fsync(f.fileno())
+            os.replace(tmp_path, file_path)
+        except Exception:
+            try:
+                os.unlink(tmp_path)
+            except OSError:
+                pass
+            raise
+
     def overwrite_performance_summary(self, summary: AgentPerformanceSummary) -> None:
         """Write the agent performance summary to a file atomically.
 

packages/valory/skills/agent_performance_summary_abci/skill.yaml

@@ -10,7 +10,7 @@ fingerprint:
   achievements_checker/__init__.py: bafybeih7da3glbp2ljghlw4ign2dxotwwkzrrnj4m5yq27zb6osyfutva4
   achievements_checker/base.py: bafybeiegrcxb3d3ivpzm4lsud7kcypphl54yukjuecqlffrkzwdjawrth4
   achievements_checker/bet_payout_checker.py: bafybeigqbcvy7fe3apckuscrp4gs5zljppady6bwe6kgjdxxjryyppo6jm
-  behaviours.py: bafybeibrruinbdq5aqbskdruv237v3swnv2sl242kj66otffwcvejwosau
+  behaviours.py: bafybeicukxgtup4dfhwubd44ls6hm27xnsrd2j7xtuw36uffpo22v4yfnm
   dialogues.py: bafybeignoeakzaf7nmdnsjhnjoga3ks6z424qcwmzkol3kikawhnxf6zju
   fsm_specification.yaml: bafybeibjgjldm26nwmidx75ylvr5q7oe4kthiphvceuerkacxd3chj6vuu
   graph_tooling/__init__.py: bafybeicek36kwi7hlbhxz4ry5j662srevbhfrhx7ocb2ihc77hhil2utqu
@@ -20,7 +20,7 @@ fingerprint:
   graph_tooling/queries.py: bafybeifywkkxfmco3baqjuc5w6fkj2cgfs3znpgoowxdjnrhugaxeevapu
   graph_tooling/requests.py: bafybeig35rdi2zpbtlzvgzaymhrsp73t42fuo2sj44fq55fayrs3c2o3j4
   handlers.py: bafybeif3pnrfdyhipmizob7zsn65f7qlf6iwuy3fi7s6ku2dvqyu4mr4am
-  models.py: bafybeidmuegnfwxusmc5kqnqfuvgvd3nxmpm6rssr6tqyqfxjbjrmr3jsu
+  models.py: bafybeiekpny4cextdog6jvxi4sdnwhpp3xskryw2ac5pp6st77mz3yl324
   payloads.py: bafybeigp52f7hcfpzmoinznqt5run3ha4vpsaaoccgvmo5skmze7flupnm
   rounds.py: bafybeien3ggbtbjigfkuzv3yadnusifrg7htnk6ialmwkd3o464oughh6i
   tests/__init__.py: bafybeibrmret5n6j7oz42ahs3hhfgmr46diwtffrccjzs7z4bcj6bcbtqy
@@ -33,10 +33,10 @@ fingerprint:
   tests/graph_tooling/test_predictions_helper.py: bafybeidkmubpjfnxys74bjaqd7ptz4gc6ejfwrrmiodmdjwdovdclvdqhu
   tests/graph_tooling/test_queries.py: bafybeih4ybhkq5seb34eqgakfpfrtlijxju3w52cjwvlrbbx2afgai4jm4
   tests/graph_tooling/test_requests.py: bafybeic2dyqmf7uj42f2f7d3hjxguw4mmlesloci4uaioui2rx3wfjhouq
-  tests/test_behaviours.py: bafybeibhwd6zs6gu5kkcmfdmhyxivaj7ggnge6dubdwcputddnitieemfi
+  tests/test_behaviours.py: bafybeiaomjvleflialemph53yj4krswne65tdbch5b3cggqcnanbnxwylu
   tests/test_dialogues.py: bafybeigezi53b2jukm5ju6z6zvecfjkjtzxcge3ehnzxryuhpambzknc3y
   tests/test_handlers.py: bafybeigybc6zkku7ldtawjcjjdkndiz3alvckxbeyjj23rsmny5ql7leku
-  tests/test_models.py: bafybeibeos6fm67nb3w6d7nb2yrygjttz7zmrwlnjogzp77rj4ttp23wgi
+  tests/test_models.py: bafybeih2uc4wa2mqbjzbj3ophnw2mdmz25owdykqcx4wzy2f3h677agzou
   tests/test_payloads.py: bafybeiet4tbmqjf7h23huifwpephtjtx4jwrcapt2kibmkl2oyeclgiggy
   tests/test_rounds.py: bafybeicrddacjku5s6h5wn3b6up552qchk7avkfputkqeuj7zcwwcb7jou
   tests/test_save_performance_summary.py: bafybeibgyx3n4dn7zhb7vfcc4wwq7nuyfep723cyqfrpnwqgsvhhwagure