BL-A-003: Data Dictionary

BIJECTIVE LABS

BIJECTIVE LABS . SEP 7, 2026

BL-A-003: Data Dictionary

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

What each stream in the substrate carries, and the conventions a reader needs to interpret the published artifacts.

1. Historical streams, first venue

Stream	Granularity	Instruments	Size
`kline_1h`	Hourly bars	154	215.0 MB
`kline_1d`	Daily bars	154	11.0 MB
`funding`	Per settlement	154	6.9 MB
`oi_1h`	Hourly open interest	154	81.3 MB
`oi_5m`	Five-minute open interest	25	0.5 MB

Bars carry open, high, low, close, volume and turnover against a bar-open timestamp. The convention throughout this record is that a bar timestamped t covers the interval ending at t, so a signal computed on that bar is measurable at t and not before.

That convention is not cosmetic. Reading a bar as covering the interval starting at t shifts every signal one period into the future and produces a look-ahead result that passes every other check.

2. Live capture streams, first venue

Stream	Trigger	Files	Size
`liq`	Liquidation event	15,224	43.1 MB
`ob_snap`	Liquidation event	15,219	321.1 MB
`trades`	Liquidation event	15,223	207.2 MB

All three are event-triggered rather than scheduled. That is the design decision documented in BL-TR-002 section 4, and it is why post-event reversion coverage came out at 29 percent.

2.1 The liquidation side field

The field requires care and it caused an incident.

The venue's earlier topic reported the side of the order performing the liquidation. Its replacement reports the side of the position being liquidated. The two conventions invert:

old topic:   is_long = ( side == "Sell" )     order-side convention
new topic:   is_long = ( side == "Buy" )      position-side convention

Any reader working with liquidation data across that boundary must know which convention their source uses. Getting it wrong flips the direction of every cascade event. BL-IN-005 is the case study.

3. Second venue streams

Stream	Content	Size
`positions`	Wallet-level open positions	2,921.4 MB
`wallets`	Wallet enumeration and metadata	331.5 MB
`trades`	Public trade tape
`book`	Order book snapshots
`assetctx`	Per-instrument context, including funding
`candle`	Bars
`funding_xvenue`	Cross-venue funding comparison
`userfills_raw`	Per-participant fill records

The position stream is what makes this venue distinct. Each record carries the holder, the instrument, the signed position size, the entry price, the liquidation price, margin used, and maximum leverage.

Two of those fields carry information no anonymous venue can provide. The liquidation price makes forced-flow pressure directly observable rather than inferred, and the entry price makes a participant's unrealised position knowable.

Position records are sampled at a 93-second median latency. That is a declared cadence, not a data property, and BL-FM-002 section 5.1 explains the trade it represents.

4. Derived capture

Stream	Content	Size
`leadlag events`	Paired large-cap and small-cap flow windows	817.1 MB

Captured at one-minute resolution. The hypothesis it served found price adjustment contemporaneous at that resolution, which closes the question at one minute and leaves it open below one minute.

5. Redaction in the published artifacts

Every artifact under _evidence/ is published as the instrument emitted it, with two changes.

Formatting. JSON re-serialised with indentation. No key, value or ordering altered.

Identifier redaction. Third-party wallet identifiers replaced by a stable pseudonym.

original    0x...
published   wallet_<8 hex characters of a hash of the original>

The pseudonym is stable, so two records referring to the same participant remain identifiable as such, which is what the identity research depended on. The identity itself is not recoverable.

Artifact	Fields redacted	Keys affected
`E-0055`	305	`addr`, `user_a`
`E-0057`	1	`hlp_addr`

No other artifact contained a wallet identifier, a host address, an email address or a credential-shaped string. The published set was re-scanned after redaction and returns zero residual identifiers across all 59 files.

Redaction never touched a value carrying a result.

6. Units and sign conventions

Quantity	Unit	Sign convention
Returns, costs, spreads, markout	Basis points	Negative means a loss to the position holder
Funding	Basis points per settlement	Positive means longs pay
Funding accrual in the cost identity	Basis points	Signed by position direction; a short receives when funding is positive
Position size	Signed units of the instrument	Negative is short
Markout	Basis points	Negative means the price moved against the passive side
Open interest	Contract units, per instrument
Timestamps	UTC, always

The funding sign is the one most often reversed by readers coming from a different source. In this record a positive funding rate means the long side pays, so a short position accruing through a positive settlement receives.

7. Era boundaries

Era	Range
Held-out A	Substrate floor to 2021-12-31
Research	2022-01-01 to 2025-06-30
Held-out B	2025-07-01 onward

Boundaries are inclusive of the stated dates and are enforced in code rather than by convention. BL-IS-007 is the guard.

The Programme II research panel drawn from the research era is 12,970,563 rows across 154 instruments and four streams, occupying 314 MB, built with zero retries.

8. What a gap means

Lake partitions are written atomically with explicit gap marking. A partition that failed to write is recorded as a gap rather than being silently absent.

This distinction matters more than it appears. Without it, a downstream reader cannot tell no data from no events, and on a sparse event stream those look identical. The programme lost more than nine days to exactly that confusion on a different feed, and the case is BL-IN-005.

9. Cross-references

BL-TR-002 the substrate. BL-A-001 the evidence register. BL-FM-002 sections 5 and 7 for cadence and substrate as declared parameters. BL-IS-007 for era enforcement.