MAIN-PROC-002: HOW THIS LIBRARY HANDLES UNCERTAINTY

Read This Once. Everything Else Will Make More Sense.

Regenerative Toronto | July 2026

Document ID: MAIN-PROC-002 Who this is for: anyone encountering this library for the first time — a reviewer, a journalist, a candidate's staffer, a future version of the process that built this. Five minutes here saves confusion later. What this is not: a financial-figure-specific guide (see REF_005_ROI_Lock_Memo for that) or a list of open issues (see TRACK_001_Master_Issues). This is the underlying method both of those apply.

THE THREE MARKS YOU'LL SEE EVERYWHERE

Every factual claim in this library carries one of three marks:

• ✅ — independently verified against a primary source (a government document, a peer-reviewed study, a named institution's own published figures). The source is cited next to the claim.
• ⚠️ — a real figure this library could not independently verify to the same standard: a single-model estimate, a derived calculation, a figure attributed to a source that couldn't be directly confirmed. Not "probably wrong" — "not yet confirmed to the same standard as a ✅ claim." Treat these as directionally useful, not citation-ready without further work.
• REMOVED or ~~struck through~~ — a claim that appeared in an earlier version of this library and was later found unverifiable, wrong, or attributed to a source that doesn't say what it was claimed to say. These are kept visible, not deleted, specifically so anyone checking this library's history can see what was tried and rejected, rather than wondering whether the same bad figure might reappear elsewhere.

If you're building something from this library — a speech, an ad, a deputation — treat ✅ claims as safe to cite directly and ⚠️ claims as needing your own follow-up before they go anywhere public.

WHY THIS LIBRARY SOMETIMES REPORTS ITS OWN EVIDENCE AS WEAKER THAN A READER MIGHT WANT

Three real examples, because the pattern matters more explained through cases than through a general statement:

The At Home/Chez Soi employment finding. This library's own core evidence base for Housing First (the At Home/Chez Soi randomized controlled trial) found that housing people without also providing supported employment did not improve employment or income — and in fact showed lower employment odds than the comparison group. A less disciplined document would have quietly left this out, since it complicates a "housing fixes everything" narrative. This library states it directly (see HMK-029 Part 10, and REC S-12, which exists specifically because of this finding) and built a distinct, better-grounded recommendation around it instead of working around the inconvenient result.

The STAR Centre's null result. Toronto's own Recovery Education Centre — the first of its kind in Canada, and the closest real-world precedent to this library's own daytime-engagement recommendation (REC DT-01) — was rigorously evaluated, and that evaluation found no statistically significant improvement in its primary outcome. This library cites that result directly in HMK-036 and in the recommendation text itself, rather than only citing the parts of STAR's story that support the recommendation. A recommendation built on a precedent whose own evidence is honestly presented is more defensible than one built on a precedent whose weaknesses were quietly omitted.

The "tentative" physical activity evidence. A systematic review this library cites for physical-activity interventions among people experiencing homelessness describes its own findings as supporting only "a tentative recommendation." This library uses that exact word rather than upgrading it to something stronger. Where the evidence is qualitatively promising but not yet quantitatively proven, this library says both things, not just the more convenient one.

The pattern across all three: when this library's own evidence complicates its own argument, that finding is kept in, named clearly, and used to build a better or more honestly-scoped recommendation — not smoothed over. This is a deliberate choice, not an accident of incomplete editing, and it is part of why this library's claims are more defensible than a version that only reported favourable findings would be.

THE CLAIM LEDGER, IN ONE PARAGRAPH

Every factual claim in this library that carries meaningful weight has a corresponding row in MAIN_DATA_001_Claim_Ledger.csv — the claim's text, its source, a status (VERIFIED / VERIFY / REMOVED), and a note on how it was confirmed. This exists so that any claim's provenance can be traced without re-doing the original research, and so that an automated test suite (see MAIN_PROC_008_Test_Suite_Documentation.md) can check that claims marked REMOVED don't silently reappear as if they were still live, that recommendation counts stay consistent across every document that states them, and that specific figures this library has previously gotten wrong don't quietly resurface. The test suite (check test_suite.py for the current exact count, since this changes too often to track here) has caught real errors in this library's own history — not a theoretical safeguard, a demonstrated one.

WHAT "GAPS" MEAN IN THIS LIBRARY

TRACK_001_Master_Issues.md tracks every outstanding item in the library honestly, organized into tiers by why something is still open:

• Tier 0 (Blocking): cannot be closed by more research — needs a lawyer, a filer, or an actual relationship with an affected organization.
• Tier 1 (Research Gaps): could in principle be closed with more research, but hasn't been yet, or is blocked on data access this process's tooling can't reach.
• Tier 2/3: bounded, smaller items — specific recommendations not yet costed, specific facts not yet reconciled.
• Tier 4 (Held): deliberately not actioned pending explicit confirmation — not an oversight.
• Tier 5: standing discipline, not a checklist — things to keep doing, not things to finish.

A library that claims to have no open gaps is either much smaller than this one or not being honest about its own limits. This library's gap list is long because the library itself is large and because gaps are logged rather than hidden — that's a feature of the discipline, not evidence the work is incomplete in a concerning way.

IF YOU FIND SOMETHING THAT LOOKS WRONG

It might be. This library has found and corrected real errors in its own prior work repeatedly — a wrong DOI, a stale figure that crept back into a document after being fixed elsewhere, an internal contradiction between two sections of the same document. Check MAIN_TRACK_003_Corrections_Log.md for the history of what's been caught and fixed; if what you've found isn't there, it's a genuinely new finding and should be treated as such, not assumed to already be known.

MAIN-PROC-002 | Version 1.0 | July 2026 | Regenerative Toronto Cross-reference: REF_005_ROI_Lock_Memo (financial-figure-specific tiering), TRACK_001_Master_Issues (the current gap list), MAIN_PROC_008_Test_Suite_Documentation.md (the automated discipline), MAIN_TRACK_003_Corrections_Log.md (the full correction history).