Session — 2026-06-10 (daytime block 2)
What's new
The Record went public. Plus: the Phoenix Lights brief, the radar-visual field note, status history on every case, and ten icon candidates on the workshop.
Public — live on imnotlikeyouguys.com
The full Record, published
The monograph (introduction, six chapters, methodology appendix),
all 14 orphan briefs, all 3
field notes, and all four deep dives (with their image strips and
suggested watching) are live on the external site — no audio, per
direction. The static /compare tool ships with
client-side resonance scoring. Navigation is now consistent on
every page. And the chrome is nameless: the brand mark is
a bare ●, every title suffix is gone, and the only mention of the
name anywhere is the URL itself.
New writing
Two events kept apart: the 22:00 lights resolved
to the 104th FS flare drop; the 19:30 V across 300 miles open.
The Symington arc (1997 alien-costume presser → 2007 reversal)
filed as institutional behavior. With audio.
The gold-standard evidence class with its failure
modes given equal weight — Belgium as the cautionary tale of a
radar-visual case decomposing into a radar case and a visual case
that shared a night. Five of six predate digital recording;
the ore is still in the file. With audio.
New instruments + ideation
/icons — ten candidate marks
Icon ideation in the site's own glyphic language
(no letters — the site is nameless): the dot, the eclipse, the
constellation, the membrane, the open case, the token, the
redaction, the scope, the transmedium, and the editor model's
orbit — a record dot with an off-center broken ring,
three ideas in two shapes. Each at 16/32/64/128px with rationale.
Status history — on every case page
corpus_status changes are now dated, reviewable
transitions: every /g/ case page
shows its status chip and a collapsible history trail, and
tools/set_corpus_status.py is the only sanctioned way
to change a status — note required, history appended, silent
edits gone. (The /g/ pages showed no status at all before.)Engine fixes
The pipeline's polish pass now names and forbids
the fact-sheet-echo failure mode. Round-5 exposed pitch
context-blindness (a model re-pitched built work); embedding the
built-list verbatim in pitch prompts is queued.
Session — 2026-06-10 (overnight block)
Earlier — the overnight block
Seven hours with the iteration team. The record gained a radio, five new briefs, two cluster essays, a comparison engine, a continuity audit that found and fixed seventeen problems, and a feedback triage desk. Everything below is listening- or reading-ready.
6
new surfaces
7
new pieces
~5.3h
new audio
90
tests passing
Everything new, one list
- /radio — the record read aloud (124 entries, ~4.9 h)
- Field note 01 — the EM-effects cluster
- Field note 02 — the water cases
- Brief 08 — Westall, 1966
- Brief 09 — Shag Harbour, 1967
- Brief 10 — Lakenheath–Bentwaters, 1956
- Brief 11 — the Washington Flap, 1952
- Brief 12 — the Foo Fighters, 1944–45
- /team/runs — verdicts: status distribution · fifth deep dive · Levelland head-to-head · round-4 pitches
- /compare — two cases side by side
- /triage — feedback + app flags, triage desk
- /audit — continuity audit (zero highs)
- /deep-dives — now with images + suggested watching
Listen first
/radio — the record, read aloud
Every entry in the documented record now speaks: 124 Kokoro
blurbs, about 4.9 hours, in a self-advancing queue with shuffle,
corpus-status filter chips (listen to just the open cases), and
a speed control. The desk twin of the phone app's drift mode.
New writing — field notes + briefs (all with audio)
Levelland, Malmstrom, Minot, Coyne, Tehran, Cash-Landrum as one
cluster: the approach/recover envelope, what binds it, what
weakens it, and the narrow evidence that would move it from
correlation to mechanism.
Ghost rockets, Trindade, Shag Harbour, Nimitz: water as theater,
the search-finds-nothing signature, and an exclusion note worth
reading — mariana-1950 entered early drafts purely by
name-collision with the trench. A cluster built by keyword keeps
it; a cluster built by reading the entries drops it.
Orphan briefs, round 2 — five open cases now covered
Westall (200+
schoolyard witnesses, the corpus's thinnest official file),
Shag Harbour
(Canada's only designated UFO crash search),
Lakenheath–Bentwaters
(the skeptics' committee produced the case's strongest sentence),
the Washington Flap
(the explanation was delivered before the analysis existed), and
the foo fighters
(adversarially corroborated — both sides filed the lights as the
other's weapon).
The team's verdicts (read at /team/runs)
The status-distribution verdict
All three roles on whether the six corpus_status buckets are
honest, which cases they'd move, and the single annotation each
would change first. First run also exposed the grounding problem
(models re-bucketed cases they misremembered) — verdicts now
argue from the embedded roster.
Which case deserves the fifth deep dive?
Three roles, three picks: Zamora/Socorro (stylist), Westall
(workhorse), the Belgian Wave (editor). The editor's argument is
structurally strongest — a mass multi-sensor multinational
military response is a vector the four existing dives don't
cover. Caveat for the read: the workhorse's "rich paper trail"
claim for Westall contradicts brief 08's documented record.
Levelland head-to-head: plasma vs ETH vs psychosocial
Forced to commit, the panel split 2–1 for plasma-atmospheric as
most parsimonious. The dissent is the better document: the editor
holds ETH strongest on the pattern while conceding the
evidence supports the behavioral description and not the
ontological conclusion. (It also misstated the case's status
despite the embedded roster — grounding still isn't solved.)
Round-4 pitches — next week's queue
Stylist: a Phoenix Lights brief. Workhorse: a radar-evidence
cross-analysis over the open cases. Editor: spectral analysis of
the Navy FLIR videos against atmospheric-scattering and
sensor-artifact models — the most ambitious and the most
falsifiable of the three.
New instruments
Two cases side by side: aligned fields with shared
values highlighted, theory anchors with shared anchors bold, and
the resonance engine's scored breakdown across the top.
The author↔corpus conversation desk. Everything
that arrives — site feedback, the phone app's ☆ flags — gets a
verdict: ack, fold (with a note saying where), or decline (with a
reason). The lifecycle is a verified statechart; nothing drops
silently.
The continuity audit: quoted numbers vs live
counts, every internal link on every page, voice anti-patterns,
unfinished threads. First run found 43 findings; all 17
high-severity were fixed in the same tick (theory-count drift in
six templates, the observatory's case links rendering to 404s,
stale about-page copy). A gate test now fails the suite if any
high-severity finding survives a run.
Team engine rebuilt — /team/runs
One engine replaces the per-round scripts: live
corpus context, draft→review→polish pipelines, and the
TeamRunMachine statechart enforced in code — empty model responses
retry and then persist as marked failures. The root cause of every
silent editor failure to date: qwen3.6 is a thinking model whose
whole token budget went to reasoning; think=false fixed it.
Verdict prompts now embed the authoritative status roster.
Deep dives gained images + suggested watching
Every dive opens with the case's image strip
(captions + attributions) and closes with the corpus's
suggested-watching grid. The phone app's detail sheet gained the
same: entry image, gold section headers, notable facts with
sources, and the watch list.
Parked — your call
Monograph theory count. The corpus has 23
theories; the monograph's introduction and chapter 2 still say 22,
and chapter 2's tier analysis was written against the 22-theory
corpus. Updating the number mechanically could desynchronize the
prose from its own argument — author decision.
App deploy pending. The detail-sheet upgrade (image, sections, watch list) is built and signed but your phone left the Mac's range before install. One command:
Audio gaps. Monograph chapters 5, 6, and the methodology appendix still have no narration (ElevenLabs quota); Kokoro can fill them on request.
Engine finding. The pipeline's polish pass sometimes regurgitates the fact sheet instead of writing prose — the brief drafts needed hand-finishing. Prompt-shape fix queued.
App deploy pending. The detail-sheet upgrade (image, sections, watch list) is built and signed but your phone left the Mac's range before install. One command:
bash tools/deploy_ios.sh with the phone unlocked.
Audio gaps. Monograph chapters 5, 6, and the methodology appendix still have no narration (ElevenLabs quota); Kokoro can fill them on request.
Engine finding. The pipeline's polish pass sometimes regurgitates the fact sheet instead of writing prose — the brief drafts needed hand-finishing. Prompt-shape fix queued.
Session — 2026-06-09 (evening)
The register membrane, a front door, a third deep dive, a fully annotated case catalogue, and the Operator's Token made printable.
The register membrane + front door
Everything on the site now lives in one of three declared
registers — ● THE RECORD (the non-fiction
corpus), ◇ THE WINDOW (the fiction), and
▣ THE WORKSHOP (the production layer). Every
page carries its register badge; links that cross registers get
the target's glyph; fiction nodes in the constellation render
dashed. A new front door at
/ opens the three doors;
the concept wiki moved to /wiki. Formal model
and invariants in docs/REGISTER-MEMBRANE.md; 14
verification gates enforce the membrane against drift.
Research depth
Third deep-dive case study. Ten timeline entries across the two
December 1980 nights and the documentary aftermath; the Halt memo
and tape treated as the contemporaneous spine; radiation readings
reported honestly as near background; the lighthouse account
given its full strength; the post-2010 binary-code material
flagged as a later, contested layer.
/observatory/gaps — annotation sweep + status lens
Every historical case now carries a
corpus_status
and a one-to-two-sentence defensible note — the analyzer's
biggest gap (48 unannotated, 72.7%) closed in one sweep. A new
status-distribution lens reads the result: 22 cases
phenomenologically open, 17 apparatus-events, 13
under-corroborated, 8 disputed optical artifacts, 3 explained,
3 documented hoaxes. Procedure + controlled vocabulary in
docs/backlog/corpus-status-procedure.md.
Off the screen
/ledger — the Operator's Token
Backlog E.F1.T1, done. A printable 30 mm disk per ledger
batch: batch code raised on the front, a seed-derived procedural
mark engraved on the back — no two batches carry the same
symbol — and a paper-insert groove in the rim. STL downloads on
every ledger entry page, parametric OpenSCAD source alongside.
Session — 2026-06-09 (overnight block)
Wake-up surface from the autonomous block of the night. Eight new analytical surfaces, an audiobook, seven monograph briefs, a global map, a gap analyzer, and a connection map between Skinwalker and the rest of the corpus. Tests: 26 passing at the start of the block, 49 passing at this writing.
12
new routes
~57m
of new audio
~18k
new prose words
58
tests passing
Codex v2 monograph
A fresh-slate non-fiction monograph in a welcoming Carl Sagan
register, written after the corpus aggregation work surfaced its
actual shape. Six chapters, an introduction, and a methodology
appendix. Reports what the corpus is, not what it would like to
be.
Seven short analytical briefs on the orphan cases — the cases
the corpus tagged
demands_new_theory during the
42-case sweep. One brief each on Robertson Panel, AOIMSG,
Mantell, Gimbal, GoFast, the Mariana Film, and Operation
Mainbrace. Filed in the v2 voice.
Observatory
The Corpus Gap Analyzer. Turns the observatory's own counts back
on themselves to surface what the corpus does NOT have:
five weakest theories (all tier 4, all with zero anchored cases),
eight orphan-case classifications (all homeless — no theory
shares their tags), eight thin-anchored theories, and the
single biggest gap it surfaced — 48 of the 66 cases then catalogued (72.7%) unannotated
by corpus_status — was closed by the annotation sweep of 2026-06-09;
the lens now reads zero and stands guard against regression.
Case Similarity Engine. For each historical case,
surfaces its five most structurally similar cases scored along
four interpretable dimensions: shared anchoring theories
(weight ×3), classification overlap (×2), category match (×1),
decade proximity (×1). The Nimitz Tic-Tac top match — the
USS Theodore Roosevelt encounters of 2014-15 — is anchored in
three identical theories. Pick a case from the dropdown.
Triple-convergent build from round-2 brainstorm (Magnum, Llama,
Qwen all pitched a comparison/similarity tool independently).
Cytoscape view of the constellation filtered to the gap analyzer's
flagged nodes — tier-4 orphan theories, thin-anchored theories,
demands_new_theory cases, unannotated cases. 66 ghost nodes
(25% of the full constellation) with 28 interconnecting edges.
The cluster shape reveals where the corpus's structural support
thins out. Sister surface to /observatory/gaps.
Built from Qwen's round-2 #2 pitch.
Horizontal-bar chronology of all corpus theories on a 1940→2026
axis, color-coded by support tier. Reveals which mid-20th-century
theories survived to the present (ETH, demonic, black-projects)
and which contemporary frameworks are still consolidating
(apparatus-response, AI / non-biological-intelligence).
Maps + deep dives
Global Anomaly Atlas. Every coord-anchored place in the corpus
(55 entries from
places.yaml) plus ten curated
high-profile case sites (Rendlesham, Nimitz, Phoenix Lights,
Trinity, Roswell, Kecksburg, Ariel School, Tehran, Varginha,
Skinwalker). Click a category pill to toggle that layer.
A generic YAML-driven case-study framework, opened with the
second corpus deep dive after Skinwalker. The Nimitz Tic-Tac
entry includes 14-day timeline, named personnel (Fravor,
Dietrich, Underwood, Day, Mellon, Elizondo), the full sensor
record (Princeton SPY-1, ATFLIR, F-18 radar, visual), theories
invoked vs not invoked with rationale, and four conventional
explanations with corpus stance on each.
Dual-axis chronology: the 22 corpus theories on the 1940→2026
timeline (top half), with the Skinwalker pre-show institutional
phases (Sherman 1994-96, NIDS 1996-2004, AAWSAP 2007-10,
Fugal 2016-20) and the 7 History Channel seasons (2020-2026)
overlaid as separate rows. The bottom section answers "what
theories were current during which Skinwalker phase" by
computing range intersection. Convergent on Qwen r2 #5.
Bipartite connection map between Skinwalker episodes and the
broader corpus. Three lenses: the 21-entry anomaly catalog
with episode coverage per pattern (uap leads at 13 episodes);
13 episodes with explicit outward cross-references; 6 corpus
items pointed at by those references, with the inverse view.
URL resolution validated — every link goes to a routable target
or renders as bare text.
Public-facing
Public-facing landing page introducing the project to a first-
time visitor. Explains the corpus, the methodology, and the
reader's possible entry points.
Test coverage
tests/test_api_smoke.py
Added 23 new pytest cases covering /about, /codex-v2,
/codex-v2/listen, /codex-v2/orphan-briefs/*,
/observatory/theories-timeline, /observatory/gaps, /deep-dives,
/deep-dives/nimitz, /api/deep-dives/nimitz, /api/audio/*,
/atlas, /api/atlas, /skinwalker/crosswalk, /api/skinwalker/crosswalk.
Each new API endpoint has a shape-sanity test (not just a
200-OK check) — the gap analyzer test asserts tier-4 overlap
with weakest theories, the atlas test asserts plausible
lat/lon on every pin, the crosswalk test asserts the Nimitz
xref resolves to the deep-dive page.
This page is curated, not auto-generated. The autonomous block
scheduled itself across multiple ticks via
ScheduleWakeup,
doing one substantive build per tick. A round-2 team brainstorm
is queued — when the user wakes, the next pitch set will be sitting
in docs/backlog/team-pitch-ideas-round2.json ready to
inform what gets built next.