BL-A-005: Chronology
BIJECTIVE LABS
BL-A-005: Chronology
Status: RELEASED
Classification: PUBLIC
Revision: 1
Date: 2026-09-07
The whole programme on one timeline, 2026-05-04 to 2026-09-07.
1. The shape of it
| Period | Days | What happened |
|---|---|---|
| 05-04 to 05-23 | 20 | Origin: retail architecture purged, migration to a microstructure system |
| 05-24 to 05-30 | 7 | The maker era. Three variants, 28,119 shadow trades, class closed |
| 05-30 to 06-12 | 14 | Taker with a wider edge. Phase gating introduced |
| 06-07 | 1 | The measurement error found |
| 06-12 to 07-02 | 21 | Five parallel machines. The apparatus built |
| 07-02 to 07-18 | 17 | The second venue. Eleven hypotheses resolved |
| 07-24 | 1 | State evaluation across both venues |
| 07-26 to 08-03 | 9 | Programme II. Six doors, eight hypotheses, one sealed gate |
| 08-03 | 1 | The anchor audit |
Roughly 90 working days from first commit to closing audit.
2. May
| Date | Event | Produced |
|---|---|---|
| 05-04 | First commit on the microstructure system | |
| 05-10 | Comprehensive system audit | Ten critical defects closed |
| 05-16 | Optimisation pass, latency and alpha quality | |
| 05-23 | External architecture critique, answered point by point | A scored self-assessment against institutional practice |
| 05-24 | Breakeven win rate measured at 122.2 percent | Pivot to passive execution. BL-IN-001 |
| 05-24 | Position tracking defect, 308.76 USD unhedged | Signed net position tracking |
| 05-26 | Exposure 84 times the declared limit after eight hours | Risk state bootstraps from the venue, not local memory |
| 05-27 | Four of eight declared instruments not streaming | Universe constitutional gate |
| 05-28 | 823 triggers, zero fills | Pivot to taker entry |
| 05-29 | Three shadow variants launched in parallel | The parallel portfolio pattern |
| 05-30 | All maker strategies halted after 28,119 trades | Phase-gated roadmap with hard criteria |
| 05-30 | Empirical gate passed on 4,800 screened cells | |
| 05-31 | 26 hours of shadow produced five trades | Rolling universe |
3. June
| Date | Event | Produced |
|---|---|---|
| 06-01 | Integrity manifest before infrastructure transfer | |
| 06-02 | 995 trades at -1.97 bps; one signal carried the whole loss | |
| 06-07 | Every verdict to date was computed gross | R-01. BL-IN-002 |
| 06-12 | Research and live container separation designed | |
| 06-14 | Three hypotheses on unused streaming data, all rejected | |
| 06-15 | Evaluation standard pre-registered | BL-D-004 |
| 06-17 | Open-interest depth assumption caught before implementation | R-22 |
| 06-22 | Rejection verdict retracted; 87 percent of sample was one instrument | R-19. BL-IN-004 |
| 06-22 | Canonical data substrate manifest | Measure, do not assume |
| 06-23 | Universe registry with denylist and liquidity floor | |
| 06-25 | Cointegration retired; walk-forward turned +180 into -55 to -68 | Walk-forward harness made mandatory. BL-IN-003 |
| 06-27 | Cash and carry retired; 13 of 14 instruments unexecutable | First venue-access verdict |
| 06-28 | Latency profile published as a sizing reference | BL-EC-002 |
4. July, first venue
| Date | Event | Produced |
|---|---|---|
| 07-01 | Unbounded event list exhausted memory; 38 hours down | Bounded sinks, class K6 |
| 07-02 | Five strategy designs and a verification runbook for the second venue | |
| 07-03 | Premium characterisation passed; cohort persistence failed | |
| 07-05 | Multi-confirmation doctrine, nine rules | BL-D-007 |
| 07-05 | Independent audit of prior venue work | Coverage below gate; an acceptance instrument found invalid |
| 07-06 | Data integrity protocol: 11 incidents, 6 classes, 8 rules | BL-D-006 |
| 07-09 | Gate orthogonality measured; three of four gates redundant | Specification reduced before any return computed |
| 07-10 | Cascade reversion falsified; recovery fraction negative | Continuation, not reversion |
| 07-10 | Adverse selection by identity confirmed at 0.906 bps | A fact, not a strategy |
| 07-10 | Selective quoting falsified; every arm gross negative | |
| 07-10 | Momentum follow falsified; gross negative before any fee | |
| 07-14 | Identity bursts: signal real at +3.04, net -5.96 | BL-SD-029 |
| 07-14 | Whale drift closed on cadence | |
| 07-14 | Passive post-exhaustion reversion: event-limited, not falsified | Still open |
| 07-14 | Fill matcher found undercounting 55.7 percent of adverse fills | Verdicts re-run and upheld with worse numbers |
| 07-16 | Deprecated feed found; blind for more than nine days | R-18. BL-IN-005 |
| 07-17 | Three timing levers measured and closed | |
| 07-18 | Cascade premise underpowered and coverage-limited | Recorder design named as the constraint |
| 07-24 | State evaluation: 20 of 22 families dead, 2 open |
5. Late July and August, Programme II
| Date | Event | Produced |
|---|---|---|
| 07-26 | Programme opens; Framework Alpha declared | Eight-layer chain, budget of 12, eras sealed |
| 07-27 | Substrate built: 12,970,563 rows, zero retries | |
| 07-27 | Proxy validated at five-minute resolution, capped at WEAK | |
| 07-28 | Crowding hypothesis falsified; skew sign opposite to prediction | |
| 07-28 | Four economic doors measured in one day | |
| 07-28 | Small-capital assessment issued | The three-clause test |
| 07-30 | Conditioned reversion closed on cost | |
| 07-30 | Branch closure retracted by its own authors | R-10, R-11 |
| 07-30 | Branch re-closed from four independent directions | |
| 07-30 | Stop overlay falsified; every level worse than no stop | The whipsaw anatomy |
| 07-30 | Independent audit; six corrections mandated | Volatility figure 2.6 times too small; lift 1.52 corrected to 1.49 |
| 08-01 | Positioning drift falsified on the mirror test | |
| 08-01 | Spread census corrected from 7 instruments to 676 | Median 1.31 becomes 5.26 |
| 08-01 | Market making closed on adverse selection | |
| 08-01 | Execution cost: 15.02 measured against 5.50 required | Patience is not a cost lever |
| 08-02 | Sealed era opened once | Research +8.24 against held-out -8.99 |
| 08-02 | Mechanism inversion recorded | The only such verdict in the archive |
| 08-03 | Rebalance anchor found uncontrolled across 19 pre-registrations | R-09. BL-IN-006 |
| 08-03 | Momentum headline revised from 7.3 to about 4.5 percent annualised | |
| 08-03 | Carry closed at class level | |
| 08-03 | Knowledge map published | BL-P2-003 |
6. September
| Date | Event |
|---|---|
| 09-07 | This record assembled from the archive |
7. Three things the timeline shows
The measurement error sits at the centre. It was found on day 34 of roughly 90. Everything before it produced verdicts that had to be recomputed; everything after it produced verdicts that stood. A programme that had run the cost identity on day one would have compressed the first third substantially, and that is the largest single efficiency available to anyone repeating this work.
The apparatus was built in three weeks and paid for itself in five days. Every durable process artifact was written between 06-15 and 07-06. In the stretch from 07-09 to 07-14, four hypotheses were resolved on a venue the programme had been working for under two weeks.
Retractions cluster around the strongest work. The three retracted or revised results, on 06-22, 07-30 and 08-03, all came out of periods when the programme was auditing itself hardest. A record with no retractions is not a record with no errors.
Copyright (c) 2026 Bijective Labs. All rights reserved.