Changelog - singularity v1.0.0 "Cygnus"

Welcome to the first major release since v0.5.17. This release brings
experimental PDP/DDO support, a wallet/actor model split, a new database story, a trustless IPFS gateway, and a
broad dependency refresh.

Please carefully read the Migration Guide and Breaking Changes

v1.0.0 is named after Cyg X-1 and future releases will carry the names of celestial objects in this class

Highlights

PDP warm storage on FWSS registry. New --deal-type pdp schedules A new PDP tracker materializes PDPVerifier events into a local view for proofset tracking.
Experimental in v1.0 -- onboarding-to-active is verified
end-to-end on calibnet; termination tracking is pending FIP-0112.
See Infrastructure Notes.

DDO cold storage New --deal-type ddo
schedules submit allocations to a configurable on-chain contract,
paid in an ERC20 token (USDFC by default on calibnet). The legacy
f05 verified-deal flow still works and remains the default for
production mainnet workloads. Experimental in v1.0 -- onboarding
works against FIP-0109-enabled networks; Curio's mk20 onboarding
pipeline is partially in place, the full documented path is still
forming. See
Infrastructure Notes.

Trustless IPFS gateway on the content-provider: /ipfs/ endpoint backed by boxo/gateway for content CID retrieval

Preparation deletes complete in bounded time: deferred batch-based cleanup with SET NULL and reaper task instead of cascades

Devcontainer-based CI on podman: near-isomorphic (because github runners) testing between local dev and CI

Before upgrading: read the Migration Guide and
the Breaking Changes sections at the bottom of
this document. The schema migration is auto-applied by
singularity admin init (or implicitly on the next run ...
startup), but the initial run for this release can take several
minutes due to index rebuilds; halt other singularity services and
run admin init standalone the first time. Operators on MongoDB or
MySQL need an offline data migration before upgrading -- see the
migration guide.

New --no-automigrate flag to disable implicit per-process AutoMigrate if admin init is run in prestart task

Features

PDP deal type

Status: experimental in v1.0. Requires Curio v1.27.4+ (first release
where PDP is on the main release train). Termination is not yet
tracked -- deals remain state=active after the underlying proof set
is removed, until FIP-0112 activates. See
Infrastructure readiness for
the full caveats and operator gotchas.

End-to-end PDP path -- new --deal-type pdp schedule mode that
drives the FWSS pull flow via Curio's /pdp/piece/pull and
/pdp/data-sets/create-and-add endpoints. SP fetches pieces from
singularity's content provider, then submits the on-chain commit from
its own wallet with the recordKeeper (FWSS) as listener. (#678,
#628, #623, #617)
PDP shovel indexer -- materialized view of PDPVerifier events
(DataSetCreated, PiecesAdded, StorageProviderChanged,
PiecesRemoved) into pdp_proof_sets for fast tracker queries.
(#617)
PDP deal tracker -- f41 deal support, polls FWSS state, surfaces
next-challenge epoch and proof-set status per deal. (#609, #612)
PDP config knobs -- --pdp-batch-size,
--pdp-max-pieces-per-proofset, --pdp-pull-timeout,
--pdp-source-url-base, --pdp-record-keeper on deal-pusher. The
--pdp-source-url-base is required when running PDP schedules and
must point at an HTTPS host where Curio can fetch pieces by
PieceCIDv2 (<base>/piece/<v2>).
send-manual-pdp command -- single-piece push for diagnostic /
e2e use; mirrors the scheduler path through PullPiecesToFWSS.
content-provider accepts v2 CIDs -- the singularity
content-provider can now serve pieces under their PieceCIDv2 path so
Curio's pull validation succeeds. (#688)

DDO deal type

Status: experimental in v1.0. Onboarding-to-active works against
FIP-0109-enabled networks (calibnet nv27+, mainnet nv27+); SP-side
mk20 support is partially in place but the documented production
setup path is still forming. Termination is not yet tracked. Set
--duration to match the SP's sector lifetime (--duration 1555200
for Curio's default) to avoid stalled ProveCommits. See
Infrastructure readiness.

DDO as third deal type -- --deal-type ddo schedules submit
allocations to a configurable Diamond proxy, wait for confirmation,
and create Deal rows tagged with DealTypeDDO and a payment-rail
ID. (#644)
DDO config knobs -- --ddo-contract, --ddo-payments-contract,
--ddo-payment-token, --ddo-batch-size, --ddo-confirmation-depth,
--ddo-poll-interval, --ddo-term-min, --ddo-term-max,
--ddo-expiration-offset on deal-pusher.
DDO E2E integration tests -- Anvil-based on-chain fixture that
exercises the full create-allocation -> wait -> deal-record flow.
(#655, #653)
Payment-token balance warnings -- DDO scheduling logs warnings
when the payer's USDFC (or configured token) balance is below the
one-month rate. (#656)
Term-max clamping --
reject legacy price-per-epoch flags that don't apply to DDO. (#666)
DDO docs + wallet funding prerequisite call-out in the runbook.

Trustless IPFS gateway

/ipfs/ endpoint on content-provider -- boxo-backed gateway over
the same rclone storage handler used for direct piece fetches;
content-addressed block retrieval with no spurious side-loads. (#646)
Parallel span reads for block retrieval -- the gateway blockstore
fetches each block via an offset/length span read, with bounded
parallel prefetch and an LRU block cache, rather than holding a single
reader open per storage. Scales across concurrent /ipfs requests
instead of serializing on one handle. (#683)
rclone v1.73.2 via patched fork -- per-storage handler pool, pacer
optimization, cache-invalidation and rawData fixes. (#646)

Schedules + deal-pusher

Explicit deal_type per schedule -- --deal-type market (legacy), --deal-type pdp, --deal-type ddo. (#623, #644)
create-batch simplification -- repeatable --deal-type replaces
the old --replication semantics; loop nesting fixed so policies
iterate before providers. (#652)
schedule list --group filter for grouped schedules. (#654)
Cleaner final-state handling -- stale error messages on a
schedule are cleared once it completes successfully. (#637)

Storage + dataprep

prep delete-piece -- delete specific pieces from a preparation
by CID. (#602)
Small piece padding -- pad pieces below the Curio verified deal minimum with
literal zeroes (materialized in non-inline mode and on the fly for inline) (#595,
#597)
add-piece lookup by CID -- consolidates pieces across
preparations by reusing an existing piece record when the CID matches.
(#607)
Non-deterministic inline DAG fix -- inline preparations now
produce deterministic piece CIDs across regeneration. (#657)

Wallet model

Wallet/Actor split -- a Wallet is the key-bearing entity (1:1
with a keystore entry), an Actor is the on-chain miner identity. A
preparation is wired to a single wallet (1:1, no more random
chooser). Migration runs automatically via admin init +
export-keys. (#629, #627, #622, #650)
Keystore-relative key paths -- wallets.key_path stores names
rather than absolute paths so keystore relocation works. (#661)
go-synapse signer -- replaces go-filsigner; one signer handles
both Filecoin-native signing and FEVM EIP-712 typed data via
SignDigest. (#622)
Wallet funding prerequisite docs for DDO scheduling.

Performance + correctness

Database

Nullable FKs for bulk deletes -- car_blocks.file_id,
car_blocks.car_id, etc. are nullable pointers; deletion of a parent
sets them NULL instead of cascading, and a separate reaper task
cleans up orphans in bounded batches. Makes preparation-delete and
job-delete linear in the table size you actually want to clear, not
in the orphan set. (#600)
Index on car_blocks.file_id -- fixes the FK cascade scan that
blocked completion of large preparation deletes. (#665)
Index on files.attachment_id -- same story for cleanup of
attachment trees.
SKIP LOCKED job claiming -- no more deadlocks between
concurrent workers claiming jobs; associations loaded after the claim
to keep the lock window short.
Auto sequence-reset on Postgres -- admin init detects stale
sequences (common after data imports with explicit IDs) and fixes
them. Standalone script
scripts/fix-sequences-after-import.sql for manual runs. (#583)
Piece-type auto-classification -- admin init infers
piece_type (data vs dag) for CAR files that predate the column.
Required for Curio compatibility on DAG-only pieces. (#583)
GORM v1.31 -- fixes around cascade behavior, transaction safety.
stripWalletAssignmentFKs errors checked -- previously silently
ignored. (#674)

Bug fixes

car_blocks.file_id IS NULL no longer mistaken for orphan --
null file_id is valid intermediate data and must not be deleted on
that filter alone. (#662)
Skip orphan cars during piece metadata lookup -- prevent
crashes from cars whose preparation has been deleted but whose
car_blocks remain.
Out-of-bounds in lone-error handler returns -- API path that
returned an error response in a single-element slice was indexing
before checking. (#670)
Err order + %d/string mismatch in ListPiecesHandler -- error
was constructed after the early return; format string mismatched the
value type. (#673)
dealtracker shadowed err -- inner result.Error was returned
via the wrong identifier, masking failures. (#671)
datasetworker MaxInterval default -- applied to the right
field; previously a typo silently left the interval at zero. (#672)
add-piece parser bug -- positional args were being treated as
subcommands. (#607)
urfave/cli duplicate provider flag panic -- guard before
registration.

Infrastructure

CI + dev environment

Devcontainer-based CI on podman -- replaces the older
matrix-of-images workflow with a single devcontainer that includes
PostgreSQL (and MySQL while it lasted) plus the toolchain. Same
container the maintainers use locally, so dialect-specific test
failures are reproducible. (#586, #620)
Skip codegen when inputs unchanged -- CI no longer regenerates
swagger/clients when none of the codegen input paths changed. (#619)
forge alongside anvil in the devcontainer for DDO/PDP test
fixtures. (#675)
gotestsum junit output -- test results reported back to GitHub
Actions as proper junit failures.
Cross-compiled release binaries -- goreleaser builds linux/windows
via CGO + zig cross-compilation and macOS natively. (#681)
Vendored FVM-solidity mocks -- DDO test fixtures no longer fetch
external solidity sources; mocks shipped in-repo with CBOR metadata
stripped to keep deterministic bytecode. (#669, #676, #667)
MySQL/MariaDB removed wholesale -- not just the driver but the
mysql-specific deadlock-debug tracing (database/deadlock_debug.go,
util/testutil/deadlock.go), the deadlock test fixtures in
handler/dataprep and service/healthcheck, the
.devcontainer/init-mysql.sh / start-mysql.sh startup scripts, the
CI's mysql DSN env in .github/actions/go-test-setup, the
mysql-specific FK migration branches in model/migrate.go, and the
mysql DSN branches in database/connstring*.go. Net -484 lines.
(#658)

Container image

Distroless cc-debian12 runner -- smaller image, fewer CVEs;
builder stage matched to the runner toolchain. (#592)
Container build gated on green CI or release -- avoids
publishing red images. (#611)

Dependencies

Significant bumps; full list in go.mod. Highlights:

gorm v1.31 with cascade fixes
boxo v0.35 (and lassie vendored without bitswap)
rclone v1.73.2 via parkan/rclone fork (cache + pacer fixes)
libp2p, go-swagger v0.33.1, golang.org/x/crypto/net/text
pgx v5.9.2 + a security-focused dependency refresh for the release
cut (#685)
go-synapse as the FEVM/PDP SDK
toolchain bumped to Go 1.26.5 (go.mod, release + devcontainer images)

Breaking Changes

These all require operator action; see the Migration Guide
for the exact steps.

Legacy database backends dropped

MongoDB import removed. Imports from singularity v1's MongoDB
store are no longer supported. Migrate via v0.6.0-RC2 first, then upgrade. (#588)
MySQL / MariaDB removed. Diagnosing and working around dialect quirks and deadlocks has been taking a significant developer toll with no known users. Please export your data to Postgres before upgrading.

Wallet model split

actors and wallets are now separate tables. Pre-v1.0 conflated both concepts under wallets and stored private keys in DB, blocking introduction of new deal types and creating security risks with database backups. You will be guided in exporting keys to a local keystore (#622,
#650)
No more random wallet picker. Random wallet assignment to preparations served no apparent purpose and has been removed (#629)

Schedule / deal flag changes

Schedules carry an explicit deal_type. Legacy schedules
without one are inferred as market on first read of the new
column, but new schedules must set --deal-type (one of market,
pdp, ddo).
--replication N is gone from create-batch. Use repeated
--deal-type instead (or --provider per copy).

Wire-level changes

Foreign keys flipped to SET NULL on car_blocks.car_id,
car_blocks.file_id, cars.preparation_id, cars.attachment_id,
files.attachment_id, directories.attachment_id,
jobs.attachment_id. GORM doesn't update existing constraints on
AutoMigrate, so admin init drops and re-creates them.
Deal.ClientID no longer has an FK to actors -- removed so
PDP/DDO deals can carry an EVM-derived client identifier that doesn't
correspond to an Actor row.
Deal.WalletID is the authoritative wallet pointer for new
deals.

API defaults (operator-only)

The API is meant for operator/localhost use, and its defaults now match
that:

Default bind is 127.0.0.1:9090 (was 0.0.0.0:9090). Pass
--bind 0.0.0.0:9090 to restore listening on all interfaces.
Wildcard CORS middleware removed. The dashboard and swagger UI are
served same-origin; a separate-origin console needs its own proxy.
Secret-named config values are masked in JSON API responses. Values
whose key matches IsSecretConfigName render as [redacted]; storage
persistence and backend setup are unchanged.

Migration Guide

From v0.5.x

Stop all singularity processes and double-check with ps/pgrep. We also recommend pausing all schedules before this (and resuming them after migration completion).
HIGHLY RECOMMENDED: Back up your database. The upgrade process has been battle tested in large-scale production and best efforts have been made to ensure atomicity and idempotency, however major upgrades always carry inherent risks. If you unable to snapshot your DB and are concerned about critical data loss please describe the situation in Slack. An automated rollback path is not provided.
If you are on MySQL/MariaDB: dump your data and restore into
Postgres or YugabyteDB before upgrading. pgloader has been tested successfully for this.
Run singularity admin init standalone to:
- rename the legacy wallets table to actors
- create wallets table and migrate wallet_assignments
- fix Postgres sequence positions if they drifted (typical after
  data imports)
- infer piece_type for cars that predate the column
- drop and re-create FK constraints as SET NULL
- add new indexes on car_blocks.file_id, files.attachment_id,
  and the jobs/preps FK columns
  This will best-effort resume if interrupted, but try to avoid terminating the process. May take a while depending on data volume and DB backing, be patient.
Perform the private key export: assuming you had existing wallets in database, you should see the following message

  Error: <N> actor(s) have private keys in the database that are not usable by
  current code.
  Run 'singularity wallet export-keys' to migrate them to the filesystem keystore

Please pay attention during this process, it is straightforward but destructively redacts private keys from the database and may require manual key injection into your secret store in advanced setups (managed cluster deployments) and has thus been broken out into an interactive flow.

$ singularity wallet export-keys
  exported: <N>
  skipped:  <M> (wallet already exists)         # on repeated runs (action is idempotent until drop confirmation) 
  errors:   <K>                                 
    - actor f1xxx: invalid key format: ...      # only on malformed keys; stop and seek support if encountered
    - actor f1yyy: keystore write failed: ...
    ...
  Drop private_key column? [y/N]

Keys will be written to ~/.singularity/keystore in lotus hex format (overridable with $SINGULARITY_KEYSTORE or falls back to ./.singularity/keystore if default/supplied path isn't writeable) on the machine running the singularity client. In a single-node cluster, simply verify that expected keys have been written and answer "y[ENTER]".

ADVANCED:
In a cluster or remote management setup, you will want to pass through the keystore to your daemons, by copying them to the worker or injecting via secrets manager. Only deal-pusher loads the keys for security reasons. For example, in a gitlab + nomad setup, you might write something like this (with a corresponding NOMAD_VAR_WALLET_KEY_f1xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx):

  template {
    data        = "${var.WALLET_KEY_f1xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx}"
    destination = "secrets/f1xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
    perms       = "0600"
    change_mode = "restart"
  }

Please consult the documentation for your orchestration system for more details. Further advanced options are available in the help text for wallet export-keys.

At this time you may also adjust your orchestration to run admin init as a prestart task for the pod/group and pass --no-automigrate to the individual daemons. This is a minor optimization as the no-op automigrate is very lightweight.

OPTIONAL: enable pdp-tracker: if you intend to make PDP deals, enable the new pdp-tracker service, which uses shovel to track onchain PDP lifecycle events.
Resume services and continue onboarding!
Lastly, once normal operation is confirmed (deals being successfully made, etc) you may want to review your backup strategy. A secure, offline backup of the keystore is highly recommended. The retention policy of the previous DB snapshots, which contain private keys, should be carefully evaluated as per your risk profile.

Optional: run the standalone scripts manually

Some helper scripts are available, though you should not need to use them directly if the migration process was carried out correctly:

scripts/fix-sequences-after-import.sql -- Postgres sequence reset
scripts/infer_piece_type.sql -- classify untyped pieces as
data / dag
scripts/cid_decode.sql -- PL/pgSQL helpers to decode bytea CID
columns to canonical CIDv0/v1 strings inside a SELECT (#687)

Enabling PDP

Before turning this on, read
Infrastructure readiness.
PDP in v1.0 is experimental; termination tracking is pending
FIP-0112. Suitable for pilots, not mainnet revenue workloads.

Required: a Curio v1.27.4+ instance running default NullAuth (the
shipping default), registered in ServiceProviderRegistry with a PDP
product whose serviceURL is reachable over HTTPS, and a payee that
matches the SP's wallet. The recordKeeper allowlist on Curio must
include the calibnet/mainnet FWSS contract; both are on the allowlist
by default.

Singularity-side:

HTTPS content-provider reachable as <base>/piece/<pieceCidV2>.
singularity run content-provider accepts both v1 commP and v2 piece
CIDs (#688); put it behind a TLS-terminating reverse proxy (caddy,
nginx, etc.)
singularity run deal-pusher --eth-rpc ... --pdp-source-url-base https://<base> ...
A schedule with --deal-type pdp and --provider <SP's f4 address>

Enabling DDO

Before turning this on, read
Infrastructure readiness --
DDO in v1.0 is experimental and the Curio mk20 onboarding path is
still partially documented. Set --duration to match your SP's
sector lifetime (--duration 1555200 for Curio defaults) to avoid
stalled ProveCommits with locked precommit deposits.

A wallet funded with the configured payment token (USDFC on calibnet
by default) and a Diamond proxy contract address from your DDO
deployment
singularity run deal-pusher --eth-rpc ... --ddo-contract 0x... ...
A schedule with --deal-type ddo and --duration 1555200 (or
whatever matches your SP's sector lifetime)

See DDO Contract Deal Guide for the full setup.

Infrastructure readiness (PDP and DDO)

PDP and DDO ship in v1.0 as experimental. The onboarding-to-active
path works end-to-end and is verified against calibnet; lifecycle
observability (termination, faults) is not yet provided by the
underlying network primitives, so it's not surfaced in singularity
either. Production mainnet workloads sized for revenue should continue
on legacy f05 until upstream FIPs activate.

What works

PDP: --deal-type pdp schedules drive Curio's /pdp/piece/pull
and trigger the SP-side on-chain commit; pdp-tracker materializes
PDPVerifier events locally and flips proposed -> active once the
SP ProveCommit's the first piece. Verified against Curio v1.27.4.
DDO: --deal-type ddo schedules allocate against the configured
Diamond contract and create payment rails. With FIP-0109 (activated
nv27, Sept 2025), the contract receives a dataSetCreated callback
on sector commit; deal-tracker reads it to flip proposed -> active. Verified against a local lotus-miner SP.

What does not work yet

Termination tracking is missing for both PDP and DDO. The
miner actor does not push a termination notification to the DDO
contract or PDPVerifier; pull-style ValidateSectorStatus
(FIP-0112) is accepted but not activated as of the v1.0 release. Deals
stay state=active after the underlying sector is terminated, and
FilecoinPay rails keep accruing. To stop the accrual, call
FilecoinPay.terminateRail(railId) manually. Design parked at
PLAN_DDO_TERMINATION_TRACKING.md for when FIP-0112 lands.
SP-side software is in transition. PDP is on Curio's main
release train as of v1.27.4 (April 2026); earlier Curio versions
do not have it. Curio mk20 DDO onboarding is partially in place
but the documented production setup path is still forming -- local
testing against the lotus-miner DDO plugin works today.
Public PDP subgraphs have known indexing gaps on the FWSS listener
field. Tracking issue: FilOzone/pdp-explorer#103. Use
singularity's Shovel-indexed pdp-tracker as the source of truth
for deal status rather than the public dashboards.

Operator gotchas

issue	what an operator must do
DDO `--duration` must match the SP's sector lifetime. Mismatch leaves the SP's precommit deposit locked until the precommit expires.	Use `--duration 1555200` (~540 days) to match Curio's default sector lifetime, or coordinate a shorter `CommittedCapacitySectorLifetime` with your SP.
SP worker wallet funding is not preflighted by singularity.	Monitor your SP's worker balance independently. `lotus-miner sectors list` showing `PreCommitFailed` is the symptom.
Curio PreCommit batching defaults to a 24-hour wait -- a single-sector DDO schedule won't ProveCommit until that window closes.	On the SP, set `[Sealing] BatchPreCommits = false` and `AggregateAboveBaseFee = "0"`, or run `lotus-miner sectors batching precommit --publish-now` manually.
Sector termination is not observable via `StateSectorGetInfo` because compaction is voluntary and most SPs never run it.	For termination/liveness checks, use partition bitfields (`AllSectors AND NOT LiveSectors`) rather than the `SectorOnChainInfo` lookup.

Recommendation by use case

use case	recommendation
Mainnet production deals, sized for revenue	Legacy f05 market deals. Mature observability, well-understood failure modes, reliable termination tracking.
Pilot / staging workloads on PDP or DDO	Supported in v1.0 with manual operational discipline. Good for ecosystem validation and building operational familiarity.
Mainnet production PDP/DDO at scale	Wait. Track FIP-0112 activation and the FilOzone PoRepService PR (#459). When both are in production, singularity will ship the termination tracker per `PLAN_DDO_TERMINATION_TRACKING.md`.

The matrix above is the operator-facing summary; the full version with
FIP citations and deeper background lives in the maintainers' internal
infrastructure-readiness notes.

PRs Included (v0.5.17..v1.0.0)

PR	Title
#689	PDP pull fixups: addPieces nonce/payload/metadata + shovel null blocks
#688	content-provider: accept v2 piece CID in /piece/
#687	psql helpers for CID inspection
#686	pre-release follow-up fixes
#685	chore: dep bump for final release cut
#684	cleanup: pre-release sweep
#683	blockstore: parallel span reads for content retrieval
#682	use self-hosted forest RPC instead of GLIF
#681	goreleaser: CGO + zig cross builds
#678	refactor PDP to use SP-side pull
#676	sol: strip CBOR metadata from mock bytecode
#675	devcontainer: add forge alongside anvil
#674	migrate: check errors in stripWalletAssignmentFKs
#673	dataprep: fix err order and %d/string in ListPiecesHandler
#672	datasetworker: apply MaxInterval default to the right field
#671	dealtracker: return result.Error, not shadowed outer err
#670	api: fix out-of-bounds on lone-error handler returns
#669	vendor fvm-solidity mocks; convert DDO test off MockAllocationFacet
#667	Consolidate FVM precompile test helpers
#666	ddo: clamp term-max via schedule.Duration; reject legacy price flags
#665	index car_blocks.file_id to fix FK cascade scans
#662	skip orphan cars in piece metadata lookup
#661	store keystore-relative names in wallets.key_path
#660	log wallet_assignments migration progress in export-keys
#658	remove mysql support
#657	fix non-deterministic piece CID for inline DAG regeneration
#656	Add DDO payment token low-balance warnings
#655	Add DDO E2E integration tests with Anvil fork helpers
#654	Add --group filter to schedule list
#653	Add DDO integration test skeleton and EVM wallet funding helper
#652	Simplify create-batch: replace --replication with repeatable --deal-type
#650	sane migration path for wallet/actor mapping
#647	complete key migration code
#646	add trustless IPFS gateway, use patched rclone, remove bitswap
#645	fix flaky TestIntegration_FullResync
#644	add DDO as third deal type
#642	remove --pdp-contract-address flag
#641	remove stale workflows
#640	pdp: propose proof set transfer to sp after filling
#639	fix PDP quirks
#637	fix: clear stale schedule error message after successful completion
#629	simplify wallet-preparation to 1:1, remove random chooser
#628	Wire deal-pusher to on-chain PDP adapter
#627	add Deal.WalletID, decouple from Actor FK
#624	Use writable MariaDB tmpdir in devcontainer scripts
#623	Add explicit deal type support for schedules
#622	wire go-synapse signer into keystore (vs go-filsigner)
#620	upgrade github-actions to v3, enable podman storage cache
#619	skip go generate when codegen inputs unchanged
#617	feat/pdp shovel indexer
#612	chore/pdptracker-rpc-wiring
#611	gate container build on green CI or release
#609	feat/f41 deal tracker
#608	bump go-swagger to v0.33.1
#607	fix add-piece to lookup existing pieces by CID
#606	dedupe fields in storage types codegen
#605	remove hardcoded @host in swagger
#602	feat: add prep delete-piece command
#601	hotfix: avoid re-scanning known valid FKs
#600	give in and use nullable FKs/pointers for massive deletes
#599	harmonize download flags, closes #460
#598	fix: correctly rename cars on S3 type remotes
#597	hotfix: use local tmpfile for padding small pieces
#596	fix remaining deadlocks + prep deletion
#595	Pad very small pieces with literal zeroes
#592	update runner image
#591	improve lotus API test survivability
#589	chore/update
#588	remove legacy mongodb
#586	Devcontainer-based CI flow
#526	fix docker-compose.yml singularity_init container env var
#519	chore: bump version to v0.6.0-RC3
#516	Fix: restore version.json to v0.6.0-RC2
#479	Feat/allow small pieces
#467	chore/linter
#460	harmonize download flags (closed via #599)

For posterity: an SP Pool replication policy and an f05-paid scaffolding
were prototyped on this branch in March 2026 and then reverted; the v1.0
deal-type model in #644/#652 is the path that stuck.

v1.0.0